Welcome to our new website!

Physical transfer of membrane and cytoplasmic components as a general mechanism of cell-cell communication
Xinle Niu, Kshitiz Gupta, Joy T. Yang, Michael J. Shamblott, Andre Levchenko


Recent evidence from different research areas has revealed a novel mechanism of cell-cell communication by spontaneous intercellular transfer of cellular components (ICT). Here we studied this phenomenon by co-culturing different cells that contain distinct levels of proteins or markers for the plasma membrane or cytoplasm. We found that a variety of transmembrane proteins are transferable between multiple cell types. Membrane lipids also show a high efficiency of intercellular transfer. Size-dependent cytoplasmic transfer allows exchange of cytoplasmic macromolecules up to 40 kDa between somatic cells, and up to 2000 kDa between uncommitted human precursor cells and human umbilical vein endothelial cells. Protein transfer, lipid transfer and cytoplasmic component transfer can occur simultaneously and all require direct cell-cell contact. Analyses of the properties of ICT, together with a close examination of cell-cell interactions, suggest that the spontaneous ICT of different cellular components might have a common underlying process: transient local membrane fusions formed when neighboring cells undergo close cell-cell contact.


Living tissues are increasingly seen as exceptionally intricate, well-structured collections of cells embedded in dynamically redefined extracellular matrices. In order to robustly perform their functions, cells comprising these tissues have to be in constant communication, engaging in a complex cooperation in the face of a dynamically changing environment. Among the various means of cooperative interaction evolved by living cells, communication by direct physical intercellular transfer still ranks among the more exotic and poorly understood.

Intercellular protein transfer (IPT), although relatively recently identified, has been implicated in a variety of biological processes, many of which have clear relevance to important pathological states. For instance, a number of glycosyl phosphatidylinositol (GPI)-anchored proteins can be directly transferred onto the plasma membrane of neighboring cells (Ilangumaran et al., 1996; Kooyman et al., 1995; Liu et al., 2002) including CD55 (19 kDa) and CD59 (70 kDa), two GPI-anchored complement restriction factors. Two animal models demonstrated in vivo that normal functions of transferred human CD55 and CD59 were preserved on recipient cells. In the mouse model, human CD55 and CD59 transferred from mouse erythrocytes to mouse endothelial cells inhibited the complement activation in mouse heart vasculature when it was perfused with human plasma (Ilangumaran et al., 1996). Similar protection of the heart vasculature from complement activation by CD55 and CD59 transferred from erythrocytes to endothelial cells was observed in a transgenic pig model (McCurry et al., 1995). It was also found that GPI-anchored cellular prion protein PrPC (35 kDa) could be transferred in vitro from donor neuroblastoma cells onto erythroleukemia cells when they were co-cultured in the presence of phorbol myristate acetate (PMA), suggesting that transfer of GPI-anchored proteins might also have implications in the pathogenesis of prion diseases (Liu et al., 2002).

Intercellular transfer of membrane proteins is not limited to GPI-anchored proteins, which are only linked to the outer layer of plasma membrane with a single GPI linkage. It has been reported that CCR5, a 40 kDa co-receptor for HIV-1 infection, could be transferred on to CCR5-negative cells through IPT (Mack et al., 2000). Transferred CCR5 enabled the HIV-1 infection of CCR5-negative cells, demonstrating the functionality of transferred protein on recipient cells. Moreover, we have recently shown that P-glycoprotein (P-gp), a 170 kDa multidrug-resistance-mediating protein, could be transferred from P-gp+ drug-resistant cells to P-gp drug-sensitive cells in vivo within tumors and in vitro between a number of cell types upon co-culture (Levchenko et al., 2005). Thus, functional transfer of P-glycoprotein confers the multidrug-resistance phenotype on recipient cells both in vitro and in vivo.

Recently, it has also been reported that membrane proteins and cytoplasmic organelles could be transferred through nanotubes between PC12 cells and several other cell types (Rustom et al., 2004). The transferred proteins included GFP-labeled Class I MHC proteins and the farnesylation signal of Ha-Ras. Transfer of GPI-anchored GFP along the nanotubes, also observed in this study, suggested that transfer of GPI-anchored membrane protein might share some similarities with the transfer mediated by nanotubes (Onfelt et al., 2004).

The mechanism of IPT is still unclear, leading to the suggestion of several, potentially conflicting hypotheses about how IPT might occur. For instance, it has been proposed that GPI-anchored proteins are released from the donor cell membrane or extracted by plasma lipid-carrier proteins, with some of the GPI-anchored proteins subsequently being re-inserted into recipient cell membrane (Ikezawa, 2002). The mechanism of CCR5 transfer has been proposed to rely solely on secretion of membrane microparticles by donor cells followed by their re-integration into the membranes of the acceptor cells. Transfer of P-gp was shown to crucially depend on cell-cell contact by a mechanism other than one of those previously proposed. As discussed above, formation of nanotubules by an as yet unknown mechanism presents a possible explanation of IPT mediated by cell-cell-contact, although it seems unlikely that significant amounts of proteins can be transferred through the one to three nanotubes reported to connect interacting cells. The mechanism(s) of IPT is thus still, for the most part, a mystery, and it is not clear whether IPT is a general phenomenon or is restricted to relatively few transferred proteins in a limited number of cell types.

In this study we propose that IPT is indeed quite general and can occur through transient cell-cell fusion following direct cell-cell contact. We also suggest that this transient fusion might lead not only to IPT, but also to exchange of other cellular components, including components of the plasma membrane and the cytosol, constituting a more general intercellular molecular component exchange phenomenon we termed intercellular component transfer (ICT). We provide evidence that ICT might be enhanced when it occurs between stem and somatic cells.


Model of membrane-fusion-based IPT

In both cell culture and in vivo, many cell types are actively or passively motile, with cells forming frequent transient contacts while undergoing random or directed movement. Our previous analysis of P-gp IPT strongly suggested that cell-cell contact is of crucial importance for efficient protein transfer. Moreover, we indeed observe that cells in culture can move with respect to each other, forming and breaking contacts, which might lead to transient fusion of cellular membranes of two neighboring, cells (supplementary material Movies 1 and 2). We therefore hypothesized that IPT can occur during transient cell-cell contacts made by cells while they move with respect to each other, and that the IPT is mediated by transient membrane fusion events, allowing a limited time for protein molecules to migrate by lateral diffusion to an adjacent cell. This hypothesis can be expressed in the form of a simplified model relating the level of expression of the transferred protein on the surface of acceptor cells to the expression of the same protein on the surface of donor cells. This model (see Materials and Methods) yields the following formula: Math(1) where ϕA and ϕD are the surface densities of the transferred protein on the acceptor and donor cells respectively, β is a normalization factor reflecting the extent of the transient cell-cell fusion, LC is the average size (diameter) of a cell, μ is the lateral membrane mobility of the transferred protein, cD is the relative density of donor cells in the cell population, and γ is the rate of loss of the protein from the acceptor cells (through dilution due to cell division or protein degradation). If γ is much greater than the product: β LC μ cD, formula (Eqn 1) can be reduced to the following expression: Math(2) where EAD is defined as the efficiency of IPT. An indirect way of establishing the validity of this approximation is to evaluate how close ϕA and ϕD are to each other. If γ is relatively small, than ϕAD, whereas if γ is relatively large, ϕA can be much smaller than ϕD, which, as shown below, is frequently the case. Therefore the approximate formula (Eqn 2) might be applicable in most cases. From the above formulas, it is clear, that the efficiency of IPT is expected to increase with increased lateral membrane mobility of the protein of interest, which can be enhanced for smaller proteins or with increased membrane fluidity. In addition, IPT is expected to be augmented if the relative density of donor cells or the expression of the protein on donor cells increases. These predictions can be directly tested experimentally. Additionally, the proposed hypothesis directly implies further properties of IPT: it should be a general phenomenon, not confined to a small number of proteins or cell types; the transfer should be bidirectional, with cells capable of being simultaneously both donors and acceptors; in addition to IPT one also expects transfer of membrane lipids and cytosolic molecules: the intercellular component transfer (ICT). Finally, continuous IPT and ICT are expected to critically depend on the constant presence of donor cells, with transferred intracellular components otherwise being quickly lost from the acceptor cells.

Protein transfer is bidirectional

To test whether surface proteins on co-cultured cells of distinct types can be mutually exchanged – or IPT is bidirectional – we explored potential transfer of three proteins not previously shown to undergo IPT. Specifically, we studied Chinese hamster ovary (CHO) cells overexpressing human membrane proteins CD36, ICAM-1 and a GFP-conjugated integrin subunit α4-GFP. Although all these proteins can mediate cell-cell adhesion, their ligands are absent in CHO cells. We then used flow cytometry to investigate protein transfer, because this technique allows the identification of the levels of fluorescence-labeled cellular components of both donor and recipient cell populations in one experiment. The results suggested that these proteins could indeed be transferred to the wild-type parental CHO cells. To investigate whether cell-cell contact was indeed required for protein transfer, we cultured untransfected CHO cells separated from transfected CHO cells with a 0.45 μm filter. Flow cytometry analysis demonstrated that, as previously observed for P-gp transfer, the transfer required cell-cell contact and the continuous presence of the protein donor cells (Fig. 1A-C). A time-course observation of protein transfer in a co-culture of donor CHO-CD36+ cells and recipient wild-type CHO cells revealed that transfer of membrane protein can be observed within 1 day of co-culture and continues for more than 4-5 days.

Although the appearance of GFP-conjugated α4 on untransfected CHO cells demonstrated that it was physically transferred onto the recipient cells, there existed a very remote possibility that the CD36 and ICAM-1 detected on untransfected CHO cells using mouse antibodies for human CD36 and ICAM-1 might be endogenous CD36 or ICAM-1 induced in untransfected CHO cells during co-culture. To provide an additional test of transfer of these proteins, we co-cultured CHO-CD36+ cells and CHO-ICAM-1 cells with mouse NIH 3T3 fibroblasts, followed by immunostaining with mouse anti-human CD36 or ICAM-1 antibody that could not recognize endogenous mouse CD36 or ICAM-1 on the recipient NIH 3T3 cells. A pronounced increase of CD36 and ICAM-1 in the mouse recipient cells (supplementary material Fig. S1L,N) indicated that CD36 and ICAM-1 could indeed be transferred. We provided further evidence for transfer by directly immunostaining the co-culture of CHO-CD36+ and CHO-α4-GFP cells and quantifying the protein levels by fluorescent microscopy (Fig. 1F). These two cell types were differentiated in fluorescent images by their distinct levels of phycoerythrin (PE)-labeled CD36 and α4-GFP. The results were consistent with flow cytometry analysis.

Fig. 1.

Physical transfer of membrane proteins in CHO cells. (A-C) Untransfected CHO cells (solid black lines) were co-cultured respectively with labeled cells (dashed pink lines): CHO-CD36+ cells (A), CHO-ICAM1 cells (B) and CHO-α4-GFP cells (C). Cell populations were separated by a 0.45 μm filter (dotted black lines), or were grown in co-culture (green lines). (D) Time course of CD36 transfer from CHO-CD36 cells to untransfected CHO cells before co-culture (blue), and after 1 day (green), 2 days (red), 3 days (cyan), 4 days (purple) and 5 days (light green). (E) Average fluorescence intensity per unit cell determined by epifluorescence in co-culture of CHO-CD36+ cells and CHO-α4-GFP cells. Each bar represents the average epifluorescence intensity of at least 50 cells.

We next co-cultured CHO-CD36+, CHO-α4-GFP and CHO-ICAM-1 cell lines in a pair-wise fashion. We observed that the corresponding proteins could be mutually exchanged between the respective protein donor and acceptor cells leading to a stable increase in transferred protein concentrations in the acceptor cells (Fig. 2A-C), suggesting that IPT is indeed bidirectional. The efficiency of the transfer was found to be different for different proteins, with CD36 being transferred with approximately double the efficiency (E ∼4%) of ICAM-1 and α4-GFP (E ∼2% for both). The differential transfer efficiency was possibly due to differences in the lateral membrane mobility of these proteins, which have considerably different molecular masses: 88 kDa for CDF36; 180 kDa for ICAM-1 homodimer and 280 kDa for α4β1 integrin heterodimer (Pinco et al., 2002), in agreement with the mathematical model.

IPT depends on the culture conditions membrane fluidity

We next tested whether other properties of IPT were consistent with our model. We have shown in a previous study that the efficiency of P-gp transfer increased as the ratio of donor cells to recipient cells increased (Levchenko et al., 2005). Similar results were observed for ICT of CD36 (Fig. 3A), ICAM-1 (Fig. 3B), α4-GFP (supplementary material Fig. S1H) and EGFR (170 kDa) (supplementary material Fig. S1H,T), in agreement with the model predicting an increase in transfer efficiency in proportion to CD/CA. We also found an increase in the efficiency of CD36 and ICAM-1 transfer under conditions of higher initial cell density (Fig. 3C; supplementary material Fig. S3), which was also consistent with our model.

Fig. 2.

Bidirectional transfer of membrane proteins between CHO-CD36+, CHO-ICAM1 and CHO-α4-GFP cells co-incubated a pair-wise fashion. (A) CD36 and α4-GFP transfer between CHO-CD36+ cells and CHO-α4-GFP cells. (B) ICAM-1 and α4-GFP transfer between CHO-ICAM1 cells and CHO-α4-GFP cells. (C) CD36 and ICAM-1 transfer between CHO-CD36+ cells and CHO-ICAM1 cells.

Consistent with the model predictions and our previous studies (Levchenko et al., 2005), the transfer was transient and reversible. The CHO-CD36 cells distributed by ATCC were derived from CHO cells by transfection with linearized cDNA encoding human CD36 followed by stable selection with G418. We found that the cells were actually a mixture of two distinct subpopulations: CHO-CD36+ with a higher level of CD36, and CHO-CD36 with a much lower level of CD36. After these two subpopulations were separated by fluorescence-activated cell sorting, the CD36 level on CHO-CD36 cells rapidly decreased to that of untransfected CHO cells. When the sorted CHO-CD36 cells were co-cultured again with the sorted CHO-CD36+ cells, the CD36 level on CHO-CD36 cells increased to a level close to that of the low-CD36 subpopulation in the original heterogeneous CHO-CD36 cell culture (Fig. 3D). These results indicate that constant contact between co-cultured cells is required to maintain a steady protein transfer.

Our model also predicts that an increase in the lateral membrane mobility might enhance the efficiency of transfer. Linoleic acid, an unsaturated omega-6 fatty acid known to increase membrane fluidity (Nano et al., 2003) was found to promote protein transfer. In its presence, the ICAM-1 levels on the recipient cells and the donor cells did not change, but the transfer of ICAM-1 was enhanced approximately twofold (Fig. 3E). Cholesterol is a rigid amphipathic molecule that has an important role in plasma membrane integrity and function (Simons and Ikonen, 2000). Cyclodextrin has been demonstrated to be an effective method to manipulate cellular content of cholesterol (Christian et al., 1997). In the presence of methyl-β-cyclodextrin, levels of CD36 in acceptor CHO cells increased, compared with that in the control (Fig. 3F). The cases above indicate that physical properties of the plasma membrane might contribute to the regulation of intercellular protein transfer.

Fig. 3.

Properties of membrane protein transfer. Donor cells (CHO-CD36+ and CHO-ICAM-1) are represented by dashed pink lines, and acceptor CHO cells represented by thin black lines. (A) CHO-CD36+ cells co-cultured with CHO cells at the seeding ratio of 1:3.6 (thick green line) or 1.2:1 (thick blue line). (B) CHO-ICAM1 cells co-cultured with CHO cells at the ratio of 1:1 (thick blue line) or 1:2 (thick green line). (C) CHO-CD36+ cells and CHO cells co-cultured in a 75 cm2 flask at the ratio of 1:1 with initial cell numbers of 1×106 (thick green line) and 2×106 (thick blue line). (D) Transfer of membrane protein is transient. 5 days after separation from CHO-CD36 cells (thick blue line) by FACS, CHO-CD36+ cells (dashed pink line) and CHO-CD36 cells (thin black line) were co-cultured again (thick green line) Black dotted line, untransfected CHO cells. (E) CHO-ICAM1 cells and CHO cells co-cultured in the presence of linoleic acid (25 μM) (thick blue line) and without linoleic acid (dotted pink line). (F) CHO-CD36 and CHO cells co-cultured without (thick red line) and in the presence of methyl-β-cyclodextrin (1 mM) (thick green line) for 2 days.

IPT can involve different proteins and occurs in and among different cell lines

The results presented so far have strongly suggested that IPT can involve different proteins within the same cell line. However, our model suggests that IPT can occur between different cell types and involve almost any surface protein, as long as transient membrane fusion occurs and the proteins have sufficient lateral membrane mobility. To explore the generality of IPT, we extended the co-culture experiments from the exclusive use of CHO cells to other cell types including human epidermal A431 carcinoma cells, which naturally express high levels of EGFR (170 kDa), human neuroblastoma cells BE(2)C/CHC(0.2) selected for high levels of P-glycoprotein (170 kDa), human ovary carcinoma cells SKOV3 naturally expressing high levels of ErBb2, and other cells including human umbilical vein endothelial cells (HUVECs), rat pheochromacytoma cells (PC12), and, more interestingly, goat mesenchymal stem cells (GMSCs) and an embryoid-body-derived (EBD) cell line (LVEC) (Shamblott et al., 2001). EBD cells are uncommitted human precursor cells established from embryonic germ cells and have been used as models of human stem cells in vitro and in vivo (Frimberger et al., 2005; Kim et al., 2005; Mueller et al., 2005). In this study, LVECs were used as an accepted surrogate for human stem cells for co-culture with other somatic cells, in part because they could be cultured in a monolayer without feeder cells.

Fig. 4.

Generality of membrane protein transfer. Human cells expressing high levels of certain membrane proteins [P-gp on BE(2)C/CHC(0.2) cells, ErbB2 on SKOV3 cells and EGFR A431 cells] and stably transfected CHO cells CHO-CD36+, CHO-ICAM1 and CHO-α4-GFP cells were co-cultured with each other or with other recipient cells (PC12 cells, NIH 3T3 cells, untransfected CHO cells, HUVECs, iHUVECs, LVECs and GMSCs). After each co-culture, cells were labeled with fluorescence-conjugated monoclonal antibodies for specific membrane proteins, followed by flow cytometry measurement. This diagram summarizes the results of the co-culture experiments in this study (see supplementary material Fig. S1 and Figs 1, 3 and 5 for the corresponding data). A solid line represents transfer of a specific protein in a co-culture experiment with its arrow pointing to the recipient cells, whereas a dashed line represents no transfer detected in a co-culture experiment. All primary antibodies were raised from mouse and only recognized the extracellular domain of the transmembrane proteins.

As summarized in Fig. 4, all the co-culture experiments that were tested clearly showed protein transfer, except the co-culture of A431 cells with NIH 3T3 cells, where no EGFR transfer was detected. In addition, no transfer of EGFR was observed in the presence of linoleic acid (supplementary material Fig. S4). All the antibodies we used recognized only the extracellular domains of the transmembrane proteins of interest. This indicated that the normal topology of transferred proteins was preserved on the plasma membrane of recipient cells. These results suggest that IPT is indeed a general phenomenon, occurring within and between different cell types and involving various proteins with different topology and molecular sizes.

One advantage of the generality of IPT is that it allows the investigation of potentially important heterotypic cell-cell interactions. To illustrate this, we analyzed CD36 transfer between human platelets and endothelial cells. Platelets express high amounts of CD36 and normally circulate in the bloodstream, coming in frequent contact with endothelial cells (Lou et al., 1997; Philippeaux et al., 1996; Philippeaux et al., 2000). Therefore, platelets might be natural donors of CD36, transferring the protein to endothelial cells lacking CD36 expression. We did indeed find that the CD36 levels on iHUVECs following co-culture were considerably higher than those in the control (supplementary material Fig. S5). We next tested whether enhancement of a specific molecular interaction controlling cell-cell adhesion could increase the IPT efficiency. We took advantage of the expression of LFA-1 on the surface of platelets, and the expression of ICAM-1 (the LFA-1 binding partner) on the surface of iHUVECs. It is well established that the expression of ICAM-1 can be further induced by pre-treatment of TNFα (Lou et al., 1997). We found that pretreatment of iHUVECs with TNFα considerably increased CD36 transfer from platelets to endothelial cells (supplementary material Fig. S5). This increase in IPT was reversed by adding an anti-LFA-1 antibody capable of blocking the interaction between LFA-1 and ICAM-1 (Bechard et al., 2001). These results suggest that the enhancement of cell-cell adhesion can increase the efficiency of IPT, potentially by facilitating the close juxtaposition of the plasma membranes of the two interacting cells, leading, in turn, to a higher probability of transient membrane fusion.

Transfer of plasma membrane lipids

Transient cell-cell fusion is expected to result not only in IPT, but also in intercellular transfer of membrane lipids. We therefore tested whether this was the case, using a lipid fluorescent dye PKH67. PKH67 is a green fluorochrome with long aliphatic carbon tails that stably incorporate into the lipid region of plasma membrane, allowing the dye to remain in the membrane for over 2 weeks or longer with little leakage and cytotoxic effects (Horan and Slezak, 1989; Waters et al., 2002). CHO-CD36+ cells were labeled with PKH67, and, after extensive washing, co-cultured with unlabeled CHO cells. A substantial transfer of PKH67 onto the unlabeled CHO cells was detected (Fig. 5A), suggesting that membrane lipids could indeed be transferred. The lipid transfer could be also directly visualized in the fluorescent images of PKH67-labeled SKOV3 cells (Fig. 5B,C), which showed high levels of PKH67 staining. As observed in the case of protein transfer, lipid transfer is also dependent on cell-cell contact. When PKH67-labeled CHO-CD36+ cells were separated from unlabeled parental CHO cells using a 0.45-μm-pore filter, lipid transfer was dramatically inhibited (Fig. 5A).

We further found that membrane lipids and membrane proteins can be co-transferred from donor cells to recipient cells. PKH67-labeled CHO-CD36+ cells were co-cultured with non-labeled CHO cells followed by immunostaining with a PE-conjugated antibody against CD36. Dual fluorescence plots for PE and PKH67 showed that lipids were transferred onto the recipient CHO cells as the donor CHO-CD36+ cells transferred CD36 onto the recipient CHO cells (Fig. 5C).

Exhibiting bidirectionality of lipid transfer, protein donor cells could also receive lipids from recipient cells as they transferred proteins onto recipient cells. When A431 cells were used as membrane protein donors, they acquired membrane lipids from the PKH67-labled CHO-CD36+ cells while transferring EFGR onto them (Fig. 5D). Again, when CHO-CD36+ cells were used as membrane protein donors, they acquired lipids from PKH67-labled A431 cells while transferring CD36 onto them (Fig. 5E). These results suggest an exchange of plasma membrane lipids between co-cultured cells as they exchange membrane proteins.

The lateral mobility of a membrane dye is expected to be much higher than that of a membrane protein, suggesting that the efficiency of the dye transfer (closely approximating membrane lipid transfer) should be much higher than the efficiency of IPT. The experiments reported in Fig. 5C-E allowed us to test this hypothesis. Indeed, we consistently found that the efficiency of PKH67 transfer was approximately ten times greater than that for CD36 between CHO cells, seven times more than that observed for EGFR between A431 and CHO cells and ten times greater than the efficiency of CD36 transfer between CHO and A431 cells.

Transfer of cytoplasmic components

In addition to IPT and membrane lipid transfer, transient cell-cell fusion is expected to result in transfer of diffusible cytoplasmic components, including proteins and small metabolites. To test this, we used cytosolic markers of different molecular sizes, as well as a marker for endosomes.

Calcein-AM (∼1 kDa) is a membrane-permeable molecule with low intrinsic fluorescence. Within a cell it is hydrolyzed by an endogenous esterase, yielding a highly negatively charged and thus membrane-impermeable metabolic product, calcein. Calcein can emit fluorescence in the green spectrum and is retained in the cytoplasm, making it a widely used cytosolic marker. We found that calcein was efficiently transferred from the calcein-AM-labeled CHO-CD36+ cells to the unlabeled CHO cells, with the transfer blocked if the labeled cells were separated from unlabeled cells by a filter with 0.45 μm pores during the co-culture (Fig. 6A). When the culture medium from the calcein-labeled CHO-CD36+ cells was collected, filtered (pore size of 0.2 μm) and used to culture the unlabeled cells, no calcein transfer was observed (Fig. 6B). These results demonstrated the dependence of ICT of small cytosolic molecules on direct cell-cell contact, which is similar to our findings for membrane protein transfer and lipid transfer.

To test whether larger cytoplasmic components could be transferred, we used quantum dots, now widely used for live cell labeling. These fluorescent nanocrystals become stably localized in endosomes upon cellular uptake (Jaiswal et al., 2003). We found no observable transfer of quantum dots in the co-culture of quantum-dot-labeled CHO-CD36+ cells with unlabeled CHO cells (Fig. 6C). This result suggested that, over the course of the experiment, no detectable transfer of endosomes occurs, which also rules out the possibility of extensive complete cell-cell fusion.

The contrast between the prominent transfer of smaller calcein and the relatively lesser transfer of quantum-dot-labeled organelles implies that the cytoplasmic component transfer may be size dependent. To further explore this dependence, we used Fluorescein-conjugated dextrans with molecular masses ranging from 3 kDa to 2000 kDa. These polysaccharides with low cell toxicity enter cells through phagocytotic and endocytotic pathways followed by gradual dispersion into the cytosol. supplementary material Fig. S2 shows an example of cytosolic distribution of a dextran molecule. After incubation with high concentrations of Fluorescein-conjugated dextran (10 kDa) for 2 hours at 37°C followed by extensive washing, about 30-40% of HUVECs showed a uniform distribution of dextran in the cytosol.

Fig. 5.

Intercellular transfer of membrane lipids. (A) PKH67-labeled CHO-CD36+ cells (dashed pink line) were co-cultured with unlabeled CHO cells (solid black line) in a monolayer (thick green line) or separated by a filter with 0.45 μm pores (thick black line). (B) PKH67-labeled SKOV3 cells were co-cultured with unlabeled SKOV3 cells (white arrows). (C) PKH67-labeled CHO-CD36+ cells were co-cultured with CHO cells. (D) PKH67-labeled CHO-CD36+ cells were co-cultured with A431 cells. (E) PKH67-labeled A431 cells were co-cultured with CHO-CD36+ cells. CD36 and EGFR levels in C-E were determined by flow cytometry.

CHO-CD36+ cells were labeled by dextrans of different molecular sizes followed by co-culture with unlabeled CHO cells (Fig. 6C-J). We found that smaller dextrans (3 kDa, 10 kDa, 40 kDa) displayed highly efficient ICT (Fig. 6C,E,G), whereas the larger ones (500 kDa, 2000 kDa) did not. The transfer of larger dextrans was not enhanced even at higher ratios of labeled cells to unlabeled cells (Fig. 6I,J), which otherwise promoted cytoplasmic component transfer (Fig. 6L). Similarly to calcein transfer, dextran transfer was dependent on cell-cell contact. Separation of labeled cells from unlabeled cells with a 0.45 μm filter in the co-culture, or growing unlabeled cells using 0.2-μm-filtered culture medium from labeled cells showed little transfer. It is important to note that the transfer of dextrans was not likely to occur through gap junctions, because they have been estimated to selectively permit the intercellular transport of cytoplasmic molecules with molecular mass not exceeding 1 kDa (Kumar and Gilula, 1996).

Fig. 6.

Analysis of intercellular transfer of cytoplasmic components by flow cytometry. (A-L) Unlabeled CHO cells (solid black line) and CHO-CD36+ cells (dashed pink line) labeled with calcein-AM, Fluorescein-conjugated dextrans of different molecular sizes and quantum dots, were co-cultured (thick green line), separated with a 0.45 μm filter (dotted blue line), and with higher ratios of labeled to unlabeled cells (thick blue lines). (M-N) Cytoplasmic macromolecules and membrane proteins were co-transferred onto recipient cells (M) or mutually transferred between co-cultured cells (N).

In another experiment, after the cells attached on the growth surface, the flasks of the co-cultured cells, the labeled cells and the unlabeled cells were gently shaken together on an orbital shaker so that the culture medium in each flask was always well mixed and homogeneous. The co-culture under shaking conditions showed similar transfer to that under normal static culture conditions (Fig. 6D). This ruled out the possibility that dextran might be released individually or within micro-membrane vesicles shed into the cell medium followed by integration into unlabeled cells.

We verified that CD36 and cytoplasmic dextran (10 kDa) could be co-transferred from dextran-labeled CHO-CD36+ cells onto recipient CHO cells (Fig. 6M). Conversely, after co-culture of dextran-labeled CHO cells with protein donor CHO-CD36+ cells, we found that the CHO-CD36+ cells received dextran from the dextran-labeled CHO cells when they transferred CD36 onto the CHO cells (Fig. 6N). This suggested a close relationship between protein transfer and cytoplasmic component transfer.

Stem cells can receive large cytoplasmic macromolecules

It was recently reported that embryonic-stem-cell-derived microvesicles reprogrammed hematopoietic progenitors by transfer of mRNA and proteins (Ratajczak et al., 2006). This raises the possibility of enhanced transfer of large cytoplasmic macromolecules, such as mRNA, between stem cells and surrounding somatic cells compared with the relatively inefficient transfer between somatic cells. We examined the properties of the transfer of cytoplasmic components in the context of uncommitted human precursor cells LVECs and HUVECs, the latter representing a potentially interesting human somatic cell type. Surprisingly, both 10 kDa and 2000 kDa dextrans were significantly transferred (Fig. 7C,D) from dextran-labeled HUVECs to LVECs after co-culture. A parallel experiment showed a considerable transfer of 10 kDa dextran from labeled HUVECs to NIH 3T3 cells but little transfer of 2000 kDa dextran (Fig. 7A,B), which was consistent with the findings in CHO cells.

Fig. 7.

Stem cells can accept large cytoplasmic macromolecules from somatic cells. (A-D) Primary HUVECs labeled with 10 kDa and 2000 kDa Fluorescein-conjugated dextrans (dashed line) were co-cultured with NIH 3T3 cells or uncommitted LVECs (thin black lines), and co-cultures (gray lines) studied by flow cytometry. (E,F) iHUVECs transfected stably with GFP were co-cultured with LVECs (E) and CHO-CD36+ cells (F).

Stem cells show a propensity of fusion with somatic cells (Cowan et al., 2005; Ogle et al., 2004; Ogle et al., 2005). To verify that cytoplasmic transfer between HUVECs and LVECs was not mediated by permanent cell fusion, we co-cultured LVECs with iHUVECs. The GFP gene in iHUVECs is transcriptionally coupled to the telomerase hTERT gene by an IRES element in the construct (Freedman and Folkman, 2004). The GFP protein is localized both in the nucleus and in the cytoplasm. The co-culture of iHUVECs with LVECs showed no detectable transfer of GFP (Fig. 7E), indicating that there was little or no whole cell fusion during the co-culture for 1-3 days. The mechanism by which LVECs receive large dextrans (2000 kDa), but not much smaller GFP (about 30 kDa), is currently not clear. There was no noticeable transfer of GFP from iHUVECs to CHO-CD36+ cells either (Fig. 7F), even though CD36 was readily transferred from CHO-CD36+ to iHUVECs during the same co-culture (supplementary material Fig. S1C). In a report of cytosolic and membrane transfer between PC12 cells through nanotubes, no transfer of cytosolic GFP was observed (Rustom et al., 2004). Among the possible explanations of this phenomenon is potential GFP multimerization or its association with cytosolic components that are not readily transferred.


Integration of cell function in the context of a tissue has long been the focus of analysis of various biological organisms. As a mechanism, direct exchange of cellular components between neighboring eukaryotic cells remains poorly characterized and understood. Our analysis here suggests that this phenomenon is quite general, capable of occurring in multiple cell types and involving multiple proteins and other cellular components. Moreover, the totality of our results is consistent with the hypothesis that ICT involves transient fusion of the plasma membranes of two neighboring cells as a result of cell-cell contact. Among the various pieces of evidence for this mechanism presented in this study, perhaps the stronger ones are the experimental data suggesting that an increase in membrane fluidity and cell-cell adhesion can enhance the transfer efficiency. Indeed, an increase of the membrane fluidity either through cholesterol depletion or addition of an omega-6 fatty acid increased the efficiency of ICT. The efficiency of ICT from LFA-1-expressing platelets was also increased following upregulation of the LFA-1 adhesion partner, ICAM, on acceptor cells.

Our results also argue that transfer by membrane microparticles, although possible, might not be the chief mechanism of ICT. We could not detect ICT following incubation of cells with filtered medium conditioned by donor cells. Moreover, if membrane microparticles shed by donor cells are indeed a major route of ICT, one would expect that the transfer of proteins and lipids and cytosolic markers would occur in proportion to their relative content in the donor cells. However, we found that the efficiency of the transfer of the lipid dyes and cytosolic dextrans was considerably higher than that of the proteins in the same experiment.

The transient cell fusion hypothesis does not exclude the possibility of ICT occurring through nanotubules. However, direct observations of cells through time-lapse microscopy suggest that the usual depictions of protein transfer through a few long nanotubules connecting a few neighboring cells might be a transient point in a more dynamic cell interaction process. In particular, we generally observed that cells actively moving within relatively dense layers, both in cell culture and tissues, can have extensive opportunities for the formation of tight transient contacts and possibly membrane fusions. As cells move away, they can transiently form multiple short tethers transiently connecting them, which ultimately either merge into a few residual long nanotubules or break off. The residual nanotubules are also very transient, rapidly disintegrating if cells continue separation (supplementary material Movies 1 and 2).

During membrane adhesion, small portions of the plasma membranes of two contacting cells might be brought into sufficiently close contact, making the polar and hydrophilic head groups of phospholipids of the outer membrane layer face each other. This energetically unfavorable state might lead to transient fusion of the outer layers of the membranes, which could then disintegrate into separated membranes or evolve into a full local membrane fusion, forming a small pore, which are both lower energy states. The intermediate state during which only the outer membranes layers fused has been demonstrated in hemagglutinin (HA)-induced membrane fusion (Melikyan et al., 1997; Melikyan et al., 1995), and has been proposed in spontaneous whole cell fusion between stem cells and somatic cells. Low efficiency of transfer of large cytoplasmic macromolecules or big organelles between somatic cells indicates that the fusion occurs within very small areas and there is very limited whole cell fusion. Rather, there might be several such small fusion events at the interface of two juxtaposed membranes, because a few small fusions would not enable considerable transfer of membrane and cytoplasmic components. These sieve-like multi-fusions have been revealed during HA-induced membrane fusion between erythrocyte and fibroblasts (Melikyan et al., 1997; Sarkar et al., 1989) and in the co-culture of neutrophils with parasites Schistosoma mansoni (Caulfield et al., 1980). Through these small continuities of the membrane and the cytosol, cellular components can be transferred passively by diffusion.

Recent studies of cell fusion have focused on fusogenic proteins that irreversibly bring two membranes close enough to induce a stable membrane fusion, which might be followed by complete cell fusion (Chen and Olson, 2005; Oren-Suissa and Podbilewicz, 2007). The apparent lack of cell-cell fusogens that resemble class I viral fusion proteins and SNARE proteins in some observations of cell fusions indicates the possible involvement of other mechanisms (Tse et al., 1993). Consistent with this prediction, our results suggest a complementary, but not exclusive mechanism: transient membrane fusion initiated by reversible membrane juxtaposition followed by local membrane destabilization. This transient membrane juxtaposition might be induced by weaker but ubiquitous adhesion molecules on many cells types, such as the LFA-1–ICAM interaction (supplementary material Fig. S5), or by mechanical impinging of neighboring cells during locomotion and local deformation, and might not be maintained long enough to lead to complete cell-cell fusion, in contrast to some of the known fusogenic proteins.

The sizes of individual fusions might depend on the cell type. The transfer of cytoplasmic macromolecules as large as 2000 kDa between some stem cells and somatic cells might allow the stem cells to exchange many types of cellular macromolecules with somatic cells, including mRNAs and proteins. This could further induce subsequent changes in neighboring cells. As the duration of cell-cell co-incubation increases, small fusions might occasionally combine into larger ones, potentially resulting in a complete whole-cell fusion for certain cell types. This might explain the observations that stem cells can fuse with somatic cells after more than 10 days of co-culture in vitro (Cowan et al., 2005) or in vivo (Ogle et al., 2004; Ogle et al., 2005). Our data suggest that stem cells might start exchanging large cytosolic molecules or membrane proteins with somatic cells much earlier. Because complete whole-cell fusion is much less frequent, as suggested in Figs 6 and 7, the earlier and more frequent instances of ICT through transient and local membrane fusions might be an important factor for stem cell differentiation induced by surrounding somatic cells, as well as for somatic cell reprogramming by adjacent, not fully differentiated cells.

It was observed that human embryonic stem cells incorporate nonhuman sialic acids from the mouse feeder layer cells (Martin et al., 2005). In this study we demonstrated that both LVECs and GMSCs could receive transmembrane proteins from neighboring differentiated cells during co-culture (Fig. 4). Of particular significance is the transfer of EGFR from A431 cells to LVECs, as well as its increasing efficiency as the ratio of A431 cells to LVECs increased (supplementary material Fig. S1T). Stem cells in adults are usually surrounded by a larger number of fully differentiated somatic cells. These naturally occurring high cell ratios might remarkably augment ICT between stem cells and somatic cells. In addition, because of the powerful amplification along the pathways of signal transduction, even a temporary acquisition of growth factor receptors could initiate sufficient downstream signaling events for stem cells to make irreversible decisions.

Not all cellular components might be available for transfer. It was reported that activation of EGFR could increase its association with the cytoskeleton. This would impair the diffusive movement of EGFR on the membrane. A possible reason for little EGFR transfer from A431 cells to NIH 3T3 cells could be that some signaling molecules secreted by NIH 3T3 cells could partially activate EGFR, which would inhibit the lateral membrane mobility of EGFR and thus ICT of EGFR.

In addition to cellular locomotion and local deformation, ICT might occur under many other conditions in vivo and in vitro. ICT of surface proteins, such as antigens and receptors, was also observed at immune synapses (Carlin et al., 2001; Hudrisiers and Bongrand, 2002; Poupot and Fournie, 2003) and has been suggested as a commonplace and important phenomenon in the functioning of the immune system (Davis, 2007). Although some of the ICT of surface proteins at immune synapses are mediated by direct ligand-receptor recognition followed by internalization, spontaneous ICT of other membrane components and cytoplasmic components might also exist where immune cells undergo transient cell-cell contact.

Although mammalian cells have been thought to be largely discrete functional units, increasing evidence suggests that they frequently undergo transient exchange of cellular components with their neighbors. Our results point to ICT as a potentially powerful general mechanism of cell-cell communication and provide insight into the underlying mechanisms. For instance, ICT might quickly modify the behavior of large groups of neighboring cells, not requiring all of them to change their genetic makeup. This could blur the boundaries between distinct cell types within complex tissues and thus the associated beneficial or potentially harmful phenotypes. Further analysis of the mechanisms might enable control or utilization of this phenomenon.

Materials and Methods

Cell culture and co-culture

CHO-CD36 cells, CHO-ICAM1 cells, SKOV3 cells and A431 cells were obtained from the ATCC (Manassas, VA) and cultured as recommended. Two subpopulations of CHO-CD36 cells: CHO-CD36+ and CHO-CD36 cells were separated by clonal selection or two rounds of fluorescence-activated cell sorting (FACS) and maintained in the same medium with CHO-CD36 cells. SKOV3 cells were selected from original SKOV3 cells from ATCC for high ErbB2 expression by FACS except for results shown in supplementary material Fig. S1, where original SKOV3 cells were used. Untransfected CHO cells were maintained in DMEM with F12 and 10% FBS and 100 U/ml penicillin-streptomycin. CHO-α4-GFP cells were maintained in the same medium with 0.8 mg/ml G418. BE(2)C/CHC(0.2) cells were cultured as described (Levchenko et al., 2005). PC12 cells were maintained in DMEM, supplemented with 10% heated-inactivated FBS, 5% horse serum and 50 U/ml penicillin-streptomycin. HUVECs and their medium EGM bullet kit were purchased from Clonetics (San Diego, CA). iHUVECs were a gift from Deborah Freedman (Harvard University, Boston, MA) and were maintained as described (Freedman and Folkman, 2004). NIH 3T3 cells were maintained in low glucose (1 mg/ml) DMEM with 10% calf serum and 100 U/ml penicillin-streptomycin. LVECs were cultured as described (Shamblott et al., 2001). GMSCs were a gift from Jennifer Elisseeff (Johns Hopkins University, Baltimore, MD) and used at passage 4. GMSCs were cultured in DMEM with 10% FBS, 100 U/ml penicillin/streptomycin, and 2 mM L-glutamine.

Platelets prepared by two methods from peripheral blood of healthy donors were used and no noticeable difference between the two isolation methods was found. In the first method, fresh platelets were prepared by density gradient centrifugation and used on the same day as the blood was drawn. Heparin-anticoagulated peripheral blood from healthy donors was centrifuged through Ficoll-Hypague gradients (30 ml of blood on 15 ml of Ficoll at 300 g for 25 minutes). The buffy layer with platelets was collected and washed in PBS with 2% FBS. After centrifuging at 300 g for 15 minutes, supernatant was collected as a fresh platelet suspension. In the second method, Interstate Blood Bank (IBB, Memphis, TN) platelet-rich plasma (PRP) was prepared by centrifugation of fresh ACD-anticoagulated blood at 800 g for 4 minutes and then collection of plasma. After receipt from IBB on the following day, PRP were centrifuged at 1000 g for 8 minutes and resuspended in PBS with 0.025 M EDTA. Platelets were further purified by two rounds of centrifugation (700 g for 2 minutes), collection of supernatant, and collection of IBB platelet suspension. For both fresh platelet and IBB platelet isolation methods, the platelet suspensions were centrifuged (1000 g for 8 minutes), resuspended in culture medium and counted for co-culture with other cells.

For co-culture, two types of cells were seeded together and cultured for about 2 days until they were close to a confluent monolayer. The initial numbers of the two cell types were adjusted according to their growth rates so that after co-culture their numbers would be comparable. The same medium was used for each flask in a co-culture experiment, usually the medium for recipient cells, or combination of both media for recipient and donor cells to ensure both cell types grew well.

Chemicals and antibodies

Linoleic acid (L5900) was purchased from Sigma (Carlsbad, CA) and was added to the co-culture medium at a concentration of 25 μM at the time of cell seeding, and medium was changed each day. Methyl-β-cyclodextrin (4555) purchased from Sigma, was added at a concentration of 1 mM at the time of cell seeding, and medium was changed each day. Recombinant human TNFα and IL-1β was from PeproTech (Rocky Hill, NJ). Fluorescence-conjugated antibody solutions for human CD36 (clone NL07), ICAM-1 (clone HA58), and EGFR (clone EGFR.1), and unconjugated antibody solutions for human CD36 (clone NL07), and EGFR (clone EGFR1) were purchased from BD Biosciences (San Jose, CA) for flow cytometry and fluorescent imaging. Other monoclonal antibodies for P-glycoprotein (clone MRK 16, Kamiya Biomedical, Seattle, WA), ErbB2 (clone 2G11, Chemicon, Temecula, CA) and a PE-conjugated goat anti-mouse IgG2a antibody (SouthernBiotech, Birmingham, AL) were also used. FITC-conjugated goat anti-mouse secondary antibody was purchased from Invitrogen (Carlsbad, CA).

Flow cytometry and fluorescence-activated cell sorting

Cells were incubated in the dark with monoclonal antibodies at concentrations recommended by their manufacturers at 4°C for 30 minutes, during which time cells were gently vortexed every 10 minutes. In the case of unconjugated antibodies, cells were washed twice and incubated in fluorescent-conjugated secondary antibodies for 20 minutes. After washing three times, cells were resuspended in cold PBS for flow cytometry measurement. Cells labeled with membrane dye or cytoplasmic markers were resuspended in cold PBS or basal culture medium, followed by immediate examination. Cells were filtered through 40 μm cell strainers (BD Bioscience, San Jose, CA) before fluorescence measurement with a FACSCalibur system or sorting by a FACSVantage SE system (BD Bioscience, San Jose, CA). At least 10,000 cells in each sample for histograms or 2500 cells for dot plots were analyzed. All curves in flow cytometry histograms were smoothed by Cellquest software.

Cell imaging

Cells were cultured in customized glass-bottom dishes (MatTek, Ashland, MA) with Fisher growth glass (Fisher Scientific, Hampton, NH), washed with warm PBS, fixed with 2% paraformaldehyde in PBS for 15 minutes at room temperature, washed twice followed by blocking with 1% BSA in PBS at room temperature for 1 hour. After removal of blocking solution, 200 μl labeling solution was added in the 15 mm well in a glass-bottom microwell dish and incubated for 45 minutes at room temperature. Labeling solution was made by adding 50 μl PE-conjugated anti-human CD36 antibody to 150 μl PBS. After incubation, cells were washed three times in PBS, and were kept in 0.5% paraformaldehyde before taking images using a Nikon Eclipse TE2000-U microscope with a Spot RT monochrome camera (Diagnostic Instruments, Sterling Heights, MI) at room temperature. The objective was ×40 or ×10 and the camera software was Spot v3.5. Fluorescence per unit cell area in the original images was measured using MetaMorph version 6 (Universal Imaging, Downingtown, PA). For live-cell imaging, cells were cultured and imaged under the same conditions except the control of temperature and humidity as well as supply of 5% CO2. Videos were generated by Spot software from sequential images.

Live-cell labeling

Qtracker 585 cell-labeling kit (Invitrogen) and membrane dye PKH67 (Sigma) were used following the manufacturers' step-by-step protocols. Low PKH67 concentrations were used in dual fluorescence experiments for better fluorescence compensation. Lysine-fixable dextrans conjugated with Fluorescein was dissolved in culture medium without serum at a concentration of 5-10 mg/ml. Cells were harvested and resuspended in dextran solution at a concentration of about 107 cells/ml, incubated at 37°C for 2 hours in a cell culture incubator, washed at least three times before co-culture with other cells. During the incubation, cells were mixed every 20 minutes to avoid cell-cell adhesion. Calcein-AM (Invitrogen) stock solution (4 mM in DMSO) was diluted in basal culture medium without serum. A final concentration of Calcein-AM of 1-5 μM was used. The final concentration of DMSO in labeling solution was about 0.1% or lower and no innocuous effects on the cell types used in this study were observed. Cells were resuspended in this labeling solution for 15-20 minutes at room temperature and washed three times in complete medium before co-culture with other cells.

Mathematical analysis of intracellular protein transfer

In setting up the model we made the following assumptions. First, we assumed that cells can be represented as essentially two-dimensional objects. Second, we assumed that IPT can only take place during formation of a transient cell-cell contact, which can be seen in a simplified fashion as a `collision' between two moving cells. Third, we assumed that prior to such collisions, two interacting cells move in relatively straight trajectories with the same speed V, the characteristic speed of migration for a given cell type. These assumptions make it possible to follow the model development usually used in molecular interaction theory, deriving the rate of chemical reactions. Instead of molecular concentrations, one can use here cell densities; the molecular sizes would correspond to the cell diameter. We can therefore assume that within a finite time period Δt a donor cell sweeps the total area equal to LcVΔt, with the probability of encountering an acceptor cell equal to this area multiplied by the density of the acceptor cells: LcVΔtcA. Since this logic can be extended to all the donor cells in the limited experimental space, the total number of donor-acceptor cell interactions will be also proportional to the density of the donor cells, cD: LcVΔtcAcD. The amount of protein transferred in each such cell interaction will depend on the efficiency of transient cell fusion μ, the difference in the protein expression levels in donor versus acceptor cells (ϕDA), and the duration of the cell-cell interaction, which is, in turn, inversely proportional to the cell speed V. The amount of the transferred protein over the same finite amount of time Δt is then proportional to the number of donor-acceptor cell interactions multiplied by the amount of the transferred protein per cell interaction:ΔTLcVΔtcAcDμDA)(1/V), where β is a proportionality coefficient. Taking the limit of Δt → ∞, which justifies the assumption of approximately straight cell movement at a constant speed, we arrive at the following expression of acquisition of the transferred protein by acceptor cells: Math(3) Note that the cell speed is no longer present in the equation, having been canceled. Since we also assume that the transferred protein is not produced within acceptor cells, but can be lost either through degradation or dilution in the process of cell division, equation 3 needs to be expanded to describe the loss of the transferred protein as follows: Math(4) At the steady state, equation 4 is equal to zero, yielding the final formula: Math(5) We note that this formula leads to reasonable predictions for several extreme cases. For instance, if cD = 0, then ϕA = 0; if cD is extremely large, then ϕA < ϕD; if γ = 0, then ϕA < ϕD and if γ is very large, ϕA < 0. We also note that although for simplicity we have postulated the parameter to describe the efficiency or extent of the transient cell fusion, this parameter will also depend on the lateral mobility of the protein with the membrane, because, if the lateral mobility was zero, no IPT would take place. Finally, we note that the implicit assumption made here is that the level of the protein of interest in the donor cells does not appreciably change because of the transfer process, a simplifying assumption that is nevertheless supported by data from all the experiments presented.


  • We thank Jennifer Elisseeff, David Sullivan, Monique Stins, Leslie Loew, Paul Roepe, Charles Flexner, Laurence Schramm and members of the Levchenko laboratory for helpful discussions and sharing of materials. The research was supported in part by the Johns Hopkins Malaria Research Institute Pilot Grant.

  • Supplementary material available online at http://jcs.biologists.org/cgi/content/full/122/5/600/DC1

  • Accepted October 24, 2008.


View Abstract