Trans-synaptic EphB2–ephrin–B3 interaction regulates excitatory synapse density by inhibition of postsynaptic MAPK signaling

Abstract

Nervous system function requires tight control over the number of synapses individual neurons receive, but the underlying cellular and molecular mechanisms that regulate synapse number remain obscure. Here we present evidence that a trans-synaptic interaction between EphB2 in the presynaptic compartment and ephrin-B3 in the postsynaptic compartment regulates synapse density and the formation of dendritic spines. Observations in cultured cortical neurons demonstrate that synapse density scales with ephrin-B3 expression level and is controlled by ephrin-B3–dependent competitive cell–cell interactions. RNA interference and biochemical experiments support the model that ephrin-B3 regulates synapse density by directly binding to Erk1/2 to inhibit postsynaptic Ras/mitogen-activated protein kinase signaling. Together these findings define a mechanism that contributes to synapse maturation and controls the number of excitatory synaptic inputs received by individual neurons.

Neuronal activity is a key determinant of maintaining and controlling the number of synaptic connections (1, 2), but the molecular mechanisms likely to establish the normal density of synaptic contacts are still not well defined (3). Synapse density could be controlled by either secreted or cell surface molecules, but trans-cellular interactions are attractive because of their ability to coordinate events between cells. One family of synaptic adhesion molecules that is well suited to control synapse density is ephrin-Bs (eBs), a family of three (eB1–3) ligands for the EphB family of receptor tyrosine kinases (4). Recent work suggests that ephrins may negatively regulate Ras/MAPK activation in nonvertebrate systems during morphogenesis (5), and that MAPK signaling can negatively regulate presynaptic terminal maturation (6, 7). Here we show that eB3 controls synapse density and the formation of dendritic spines through a competitive postsynaptic mechanism relying on inhibition of MAPK signaling.

Results

Ephrin-B3 Expression Level Controls Synapse Density.

Early in development [5–10 days in vitro (DIV 5–10)], there are few dendritic spines and synapses that are rapidly added on the dendritic shaft. During this time, eB3 is enriched at excitatory synapses but is also localized to many extrasynaptic locations (Fig. 1A). As neurons mature (DIV 14–21), dendritic spines are formed and become the principal site of excitatory synapses (8); eB3 becomes largely restricted to spine and shaft excitatory synaptic contacts (Fig. 1A and Fig. S1A). The synaptic pattern of eB3 expression suggests that it may be involved in the formation and maturation of excitatory synapses.

Fig. 1.

Synapse density is proportional to eB3 expression. (A) DIV-8 (Upper) and DIV-21 (Lower) cortical neuron dendrites transfected with GFP at DIV 0 and stained with anti-eB3 (green), anti-PSD-95 (red), and anti-VGlut1 (blue) antibodies. Grayscale images show the pattern of eB3 staining. (B) Quantification of eB3 puncta density in neurons transfected with vector control (n = 27), eB3 shRNA#1 (n = 32), or eB3 shRNA#2 (n = 15). (C) DIV-9 neurons transfected with GFP plus indicated shRNA construct at DIV 0 and stained with anti-GFP (green), anti–PSD-95 (red), and anti-VGlut1 (blue) antibodies. Arrows indicate synaptic puncta identified by colocalized PSD-95 and VGlut1 immunostaining. (D) Quantification of colocalized PSD-95 and VGlut1 synaptic puncta density for vector control (n = 29), ephrin-B1 shRNA (n = 19), eB3 shRNA#1 (n = 31), or eB3 shRNA#2 (n = 16). (E) Quantification of colocalized PSD-95 and VGlut1 synaptic puncta density for vector control (n = 26), eB3 shRNA#1 (n = 23), or ephrin-B2 shRNA (n = 24). (F) DIV-9 cortical neuron dendrites transfected with GFP plus eB3 shRNA constructs and stained with anti-GFP (green), anti-eB3 (red), and anti-PSD-95 (blue) antibodies. Arrows and arrowheads indicate eB3 or PSD-95 puncta. (G and H) Plot of (G) eB3 puncta density or (H) eB3 staining intensity in arbitrary units vs. PSD-95 puncta density in DIV-9 neurons expressing GFP plus vector control [filled circles: (G) PSD-95 puncta density × eB3 puncta density, R² = 0.515, P < 0.0001; (H) PSD-95 puncta density × eB3 staining intensity, R² = 0.481, P < 0.0001] or eB3 shRNA#1 [open circles: (G) PSD-95 puncta density × eB3 puncta density, R² = 0.272, P < 0.002; (H) PSD-95 puncta density × eB3 staining intensity, R² = 0.354, P = 0.0002) stained with anti-eB3, anti–PSD-95, and anti-GFP antibodies. Plot is linear regression curve for all cells. (Scale bars, 3 μm.) Error bars indicate SEM. *P < 0.05, **P < 0.0001.

We asked whether the expression level of eB3 is related to the density of excitatory synapses on individual neurons by plotting the density of endogenous PSD-95 puncta vs. two different measures of eB3 expression: the density of eB3 puncta and the overall intensity of eB3 immunofluorescence. We found that both measures of eB3 expression vary from neuron to neuron and are correlated with PSD-95 puncta density (Fig. 1 F–H). We next examined the relationship between expression level and synapse density of three different protein families (neuroligins, synCAMs, and EphB2) known to control synapse formation (4), but none of these three factors had any significant correlation between their intensity of expression and PSD-95 puncta density (SI Text and Fig. S2). Thus, unlike other known synaptogenic factors, the level of eB3 expression is positively correlated with the density of PSD-95 puncta.

To test whether the amount of eB3 expression in an individual neuron determines the number of synapses it receives, we artificially reduced the expression of eB3 using previously characterized short hairpin RNA (shRNA) (9) that had no effect on synapse density when expressed in neurons from ephrin-B3^−/− (eB3^−/−) mice (SI Text and Fig. S3 A and B). Cortical neurons were cotransfected with eB3 shRNA constructs and GFP at DIV 0 and stained for endogenous eB3 and PSD-95 at DIV 9. Because neurons normally express different amounts of eB3, knockdown (kd) results in a range of expression levels (Fig. 1G). We found that kd of eB3, but not eB1 or eB2, causes a significant decrease in synapse number (Fig. 1 C–E and Figs. S4B and S5 A–C). The effects of eB3 kd did not require neuronal activity (SI Text), and we found an overall decrease in both eB3 and PSD-95 puncta density relative to control (Fig. 1B). Yet, the linear correlation between eB3 expression and PSD-95 puncta density in eB3 kd neurons (Fig. 1 F–H) was statistically indistinguishable from wild-type (wt) neurons (P = 0.4659, eB3 staining intensity; P = 0.1241, eB3 puncta density; ANCOVA). These results demonstrate that eB3 expression and the density of synaptic specializations are correlated in both wt and kd neurons, and that the linear relationship in these two groups is the same. Therefore, the variability in synapse density in neurons is likely a result of differences in eB3 expression level.

To test whether eB3 expression affects the number of functional synapses that a neuron receives, we transfected DIV 0 cortical neurons with eB3 shRNA constructs and GFP, and recorded mini excitatory synaptic currents (mEPSCs) at DIV 9–10. Expression of eB1 shRNA resulted in mEPSC that occurred at normal (∼1-Hz) frequency but had smaller amplitudes (Fig. 2 A, C, and D). In contrast, kd of eB3 with two different shRNAs resulted in neurons with fewer and smaller mEPSCs (∼0.2 Hz; Fig. 2 A and C) without affecting mIPSC frequency (Fig. S6). Cotransfection of eB1 or eB3 shRNA with constructs encoding kd-insensitive ephrin-B1 or eB3 (HAeB1R or HAeB3R, respectively) resulted in rescue of the observed changes, indicating that these effects are specific to loss of the targeted eB molecule (Fig. 2 A, C, and D). We confirmed the effects on synapse density after kd and rescue using immunostaining for synaptic marker proteins (Fig. 1 C–E and Figs. S4B and S7). These results indicate that kd of eB3, but not eB1, reduces the density of functional excitatory synapses a neuron receives.

Fig. 2.

Ephrin-B3 knockdown reduces mEPSC frequency. (A) mEPSCs from whole-cell patch-clamp recordings of neurons expressing GFP plus indicated constructs (calibration, 40 pA, 1 s) and individual events (Inset: calibration, 20 pA, 20 ms). (B) Schematic of eB3 mutants used. HAeB3R is HA-tagged full-length eB3 rendered insensitive to eB3 shRNA#1. mCer-eB3/B1 is mCerulean tagged to the extracellular domain of eB3 fused to the intracellular domain of ephrin-B1. HAeB1/B3 is HA tagged to the extracellular domain of ephrin-B1 fused to a knockdown-insensitive intracellular domain of eB3. HAeB3R_L293A is HAeB3R in which alanine has replaced leucine 293 in a juxtamembrane domain (green star). (C) Quantification of mEPSC frequency after transfection with vector control (n = 14), ephrin-B1 shRNA (n = 11), eB3 shRNA#1, 1.5 μg/well (n = 23), eB3 shRNA#2 (n = 6), eB3 shRNA#1 plus HAeB3R (n = 19), eB3 shRNA#1 plus mCer-eB3/B1 (n = 12), eB3 shRNA#1 plus HAeB1/B3 (n = 6), or eB3 shRNA#1 plus HAeB3R_L293A (n = 7). *P < 0.005 from control. (D) Quantification of mEPSC amplitude after transfection with vector control (n = 686), ephrin-B1 shRNA (n = 566), ephrin-B1 shRNA plus HAeB1R (n = 861), eB3 shRNA#1 (n = 424), or eB3 shRNA#1 plus HAeB3R (n = 2979). ***P < 0.001. (E) Quantification of mEPSC frequency after transfection of 0.25 μg (n = 9) or 1.5 μg (n = 5) eB3 shRNA#1 per well. (F) Quantification of mEPSC frequency after transfection of eB3 shRNA#1 plus 0.25 μg (n = 8) or 1.0 μg (n = 13) HAeB3R per well. *P < 0.05. Error bars indicate SEM.

Next, by varying the degree of kd in single neurons (9) (Fig. S1B), we tested whether different levels of eB3 kd might result in a graded effect on mEPSC frequency. Consistent with a role in controlling synapse density, transfecting increasing amounts of eB3 shRNA constructs into single neurons led to a progressive reduction in mEPSC frequency (Fig. 2E). Increasing eB3 expression by cotransfecting eB3 shRNA constructs with different amounts of the HA-tagged eB3 rescue construct (HAeB3R) or by overexpressing eB3 without kd led to mEPSC frequencies greater than normal (Fig. 2F; eB3 overexpression: 2.2 ± 0.7 Hz, n = 9). Taken together, these results provide evidence that the postsynaptic expression level of eB3 in an individual neuron is a key determinant of the density of synapses that neuron receives.

Ephrin-B3^−/− Mice Have Fewer Dendritic Spines but Normal Numbers of Synapses.

To test whether eB3 is required for normal dendritic spine density, we examined dendritic spines in cortical and hippocampal cultures following eB3 kd (SI Text and Figs. S1 C–F and S5 E and F) or in cortical brain slices from wt or eB3^−/− mice. In cell culture, we found a decrease in spine density and a significant decrease in the density of both spine and shaft synapses following eB3 kd (SI Text and Fig. S1 C–F). Pyramidal neurons from eB3^−/− cortical brain slices had grossly normal arbors. However, the overall density of dendritic protrusions was significantly reduced (Fig. 3 A and B) with a ∼60% reduction in dendritic spine density (Fig. 3C) and no change in the density of filopodia-like protrusions (Fig. 3D). To address whether cortical neurons from eB3^−/− mice have fewer synapses, we determined synapse density in layer 5 of cortical cryosections from wt and eB3^−/− mice by measuring overlap of pre- and postsynaptic markers (10). Consistent with previous reports (11–13), we found no significant difference in excitatory synaptic density between wt and eB3^−/− mice (Fig. 3 F and G).

Fig. 3.

Reduced spine density in eB3^−/− mice. (A) Cortical neuron dendrites in slice culture from P4-6 wt or eB3^−/− mice transfected with GFP. Arrows indicate spines; arrowheads indicate filopodia-like protrusions. (B–E) Quantification of protrusion density for (B) all protrusions, (C) spines, (D) filopodia-like protrusions, and (E) protrusions classified as “other” for wt (n = 16) or eB3^−/− (n = 20) neurons. (F) (Top) Images of layer 5 cortical cryosections from P6 wt or eB3^−/− mice immunostained with anti-VGlut1 (green) and anti-SynGAP (red) antibodies. (Scale bar, 10 μm.) (Bottom) Binary mask illustrating colocalized VGlut1 and SynGAP puncta in above images. (G) Quantification of colocalized VGlut1 and SynGAP puncta density in wt (n = 59 images) and eB3^−/− (n = 60 images) sections. (H) DIV-4–6 cortical neuron dendrites in slice culture from P4-6 wt or eB3^−/− mice transfected with tdTomato (red) and PSD-95-GFP (green) at DIV 2. Arrows indicate PSD-95-GFP puncta in dendritic protrusions. Arrowheads indicate PSD-95-GFP puncta on the dendritic shaft. (I and J) Quantification of puncta density for (I) PSD-95-GFP puncta found in protrusions and (J) total PSD-95-GFP puncta for wt (n = 12) or eB3^−/− (n = 39) neurons. (Scale bars, 3 μm.) Error bars indicate SEM. *P < 0.002.

The majority of excitatory synapses on cortical neurons are found on spines, yet neurons from eB3^−/− mice have fewer spines than wt littermates but normal numbers of excitatory synaptic specializations; this suggests that loss of eB3 leads to a relocalization of synaptic specializations. Although neurons from slices lacking eB3 had PSD-95-GFP puncta density that was similar to wt (Fig. 3 H–J), PSD-95-GFP puncta were found primarily along the dendritic shaft with few puncta found in dendritic protrusions, indicating that the reduced number of spines in eB3^−/− mice results in a redistribution of synaptic specializations to the dendritic shaft of cortical neurons.

Ephrin-B3 Controls Synapse Density Through Competitive Cell–Cell Interactions.

Neurons lacking eB3 have normal synapse numbers, but eB3 kd reduces excitatory synapse density. Based on these findings, one likely mechanism mediating eB3 dependent control of synapse density is competitive cell–cell interactions. In this model, a neuron expressing more eB3 would receive more synapses than a nearby neuron expressing less eB3, but without eB3 there would be no signal generated. To test whether the level of eB3 expressed might enable competition between neurons, we asked whether the ability of eB3^−/− neurons to make normal numbers of synapses along their dendrites (Fig. S3 A and B) could be disrupted by simply coculturing them with neurons expressing normal levels of eB3. To do this, we made heterogenotypic cultures containing mixtures of labeled and unlabeled cortical neurons from wt and eB3^−/− mice. Labeling was carried out in suspension using electroporation of GFP (Fig. S8). Although synapse density in the GFP expressing neurons at DIV 10 in wt\wt or eB3^−/−\ eB3^−/− cultures (Fig. 4, A and B) was similar, GFP-positive eB3^−/− neurons mixed with wt neurons had significantly fewer synapses (Fig. 4 A–C and Fig. S4C). Thus, eB3^−/− neurons have normal numbers of synapses when surrounded by other eB3^−/− neurons, but not when they are surrounded by wt neurons. Moreover, GFP-expressing wt mouse neurons had significantly higher synapse density when cultured with unlabeled eB3^−/− neurons (Fig. 4 A–C and Fig. S4C). Thus, changes in the relative level of eB3 expression can selectively increase or decrease the density of synapses a neuron receives. Because the normal synapse density that we observe in eB3^−/− neurons is not likely due to compensation by other synaptogenic molecules (SI Text and Fig. S3C) or culture conditions (SI Text and Fig. S3 A and B), our data suggest that cell-to-cell differences in eB3 expression level enable neuronal competition that controls overall synapse density.

Fig. 4.

Competitive control of synapse density by ephrin-B3. (A) Hetergenotypic cultures of P0 wt and eB3^−/− littermate mice. Unlabeled neurons of one genotype were plated followed by a second group labeled by electroporation with GFP. Cultures were stained at DIV 10 with anti-GFP (green), anti-PSD-95 (red), and anti-VGlut1 (blue) antibodies. Arrows indicate synaptic puncta identified by colocalized PSD-95 and VGlut1 immunostaining. (B) Quantification of synaptic puncta density for GFP-labeled neurons in heterogenotypic cultures (wt/wt, n = 44, eB3^−/−/wt, n = 67, eB3^−/−/eB3^−/−, n = 50, and wt/eB3^−/−, n = 70). (C) Model of experimental design and results. (Scale bar, 3 μm.) Error bars indicate SEM. * P < 0.002, ** P < 0.0003.

EphB2 Is the Presynaptic Ligand for Postsynaptic Ephrin-B3.

Our heterogenotypic culture experiments suggest that eB3 can control synapse density through trans-cellular interactions. Consistent with this hypothesis, eB3 expressed in nonneuronal cells can induce presynaptic differentiation when cocultured with hippocampal neurons (12). We tested whether this was a unique property of eB3, or whether other eBs or ephrin-As could also induce presynaptic differentiation. In heterologous cell assays, we find that HEK293T cells expressing any of the three eB ligands were able to induce presynaptic marker accumulation to a similar level as EphB2 (10) (Fig. 5A and Fig. S9). However, when cocultured with neurons, ephrin-A1–expressing HEK293T cells failed to induce an increase in VGlut1 staining (Fig. 5A and Fig. S9), suggesting that the effects we observe are specific to eB family members and likely occur through interactions with presynaptic EphBs.

Fig. 5.

EphB2 is a presynaptic ligand for eB3. (A) Quantification of induction of VGlut1 puncta area in axons under HEK293T cells transfected with vector control (n = 78), HAeB1 (n = 70), HAeB2 (n = 63), HAeB3 (n = 68), or eA1 (n = 23). (B) DIV-10 axons from cortical neurons transfected with synaptophysin-GFP (green) and indicated shRNA constructs cocultured with HEK293T cells transfected with RFP or HAeB3 (red) for 16–18 h. Arrowheads indicate synaptophysin-GFP puncta colocalized with HEK293T cells. (Scale bar, 3 μm.) (C) Quantification of fold increase in synaptophysin–GFP density in axon segments underneath HEK293T cells compared with adjacent axon segments for neurons transfected with indicated shRNA constructs (cocultured with RFP-expressing HEK293T cells, HAeB3-expressing HEK293T cells): vector control (RFP: n = 111, HAeB3: n = 86), EphB2 shRNA#1 (RFP: n = 81, HAeB3: n = 85), EphB2 shRNA#2 (RFP: n = 41, HAeB3: n = 45), EphB2 shRNA#2 + EphB2R (RFP: n = 55, HAeB3: n = 58). Error bars indicate SEM. *P < 0.006.

EphB2 copurifies with both presynaptic active zones and the postsynaptic density fraction (14). To identify whether a presynaptic EphB receptor mediates eB-dependent presynaptic induction, we used a modified coculture assay (9) in which axons expressing shRNA targeting potential presynaptic receptors are cocultured with eB expressing HEK293T cells (9). We transfected neurons with constructs expressing shRNA targeting EphB2 (10, 15) along with a GFP-tagged presynaptic marker synaptophysin (syn-GFP) to label transfected axons, and at DIV 7 transfected neurons were cocultured with HEK293T cells expressing either HA-tagged eB3 (HAeB3) or RFP. At DIV 8, single axons at sites of contact with eB3-expressing HEK293T cells resulted in a significant ∼1.3-fold increase in the density of syn-GFP (Fig. 5 B and C). The expression of either of two independent EphB2 shRNA constructs in axons (15) blocked the increase in syn-GFP density (Fig. 5 B and C). These effects were rescued by coexpression of a kd-insensitive EphB2 (Fig. 5 B and C). Thus, EphB2 appears to be necessary for eB3-dependent presynaptic induction, suggesting that EphB2 is the presynaptic ligand for eB3.

Erk and Ephrin-B3 Interact to Control Synapse Density.

To determine how eB3 regulates synapse density, we used a whole-cell patch clamp to record mEPSCs in DIV 9–11 neurons and asked whether the intracellular domain of eB3 is required for changes in mEPSC frequency. Although the known signaling domains of eBs are conserved (12, 16), eB3 also contains a nonconserved, ∼60–amino acid juxtamembrane domain. We constructed two chimeric proteins consisting of the extracellular domains of either mCer-eB3 or HAeB1 fused to the intracellular domains of either eB1 or eB3 (Fig. 2B). Both chimeric molecules reached the cell surface when expressed in HEK293T cells. However, expression of mCer-eB3/B1 in neurons expressing eB3 shRNA failed to rescue mEPSC frequency, whereas coexpression of HAeB1/B3 was able to rescue mEPSC frequency (Fig. 2G). These results indicate that the regulation of synapse density by eB3 requires intracellular residues that are distinct from those found in eB1.

Sequence analysis revealed a putative Erk binding domain (D-domain) centered at leucine 293 (L293) in the juxtamembrane region of eB3. To test for an eB3-Erk interaction, we asked whether eB3 and Erk physically associate at synapses by immunoprecipitation from mouse brain synaptosomes. Consistent with previous reports (11) and our immunostaining data, eB3 was enriched in the synaptosomal fraction (Fig. 6A). In purified synaptosomes (17), we found that whereas both Erk1 and Erk2 were present at similar levels (Fig. 6A), only Erk2 effectively coimmunoprecipitated with eB3 (Fig. 6B). Thus, eB3 and Erk2 appear to associate at synapses in brain lysates.

Fig. 6.

Postsynaptic eB3 acts through Erk1/2 to control synapse density. (A) Various purification fractions from three P30 mouse brains. Blots are probed with antibodies recognizing eB3, NR1, GluR2, PSD-95, and Erk1/2 (n = 3). (B) EB3 immunoprecipitated from synaptosomes and resulting Western blots were probed with anti-Erk1/2 and anti-eB3 antibodies (n = 3). Lysates from the same preparations are shown in lower blots. (C) GST fusion proteins of wt (eB3intra) or L293A mutant (eB3 L293A) intracellular domains of eB3 were expressed in bacteria and used in GST pull-down assay with E19 rat brain lysates. Resulting elutates and input to the columns were analyzed by Western blotting with anti-Erk1/2 antibody. Fusion protein expression was analyzed on a Western blot with anti-GST antibody. (D) Quantification of relative Erk1 and Erk2 pull down as compared with total Erk1 and Erk2 input (n = 3). (E–G) Density of excitatory synapses in rat cortical neurons transfected with DN-MEK and either control or eB3shRNA constructs at DIV 10. (E) Quantification of the frequency of mEPSCs vector control as in Fig. 2, DN-MEK (n = 11), CA-MEK (n = 7), eB3 shRNA#1 plus DN-MEK (n = 12), eB3 shRNA#1 plus CA-MEK (n = 6). (F) Examples of staining for markers PSD-95 (red) and VGlut (blue) in transfected neurons. (G) Quantification of the synapse density. Vector control (n = 27), eB3 shRNA#1 (n = 39, P = 0.1212). (H) EB3^−/− and wt neurons transfected with GFP alone or GFP and DN-MEK cultured on wt mouse cortical neurons. GFP (green), PSD-95 (red), and vGlut1 (blue). Arrowheads indicate colocalized PSD-95 (red) and vGlut (blue) puncta in transfected neurons. (I) Quantification of synaptic density in heterogenotypic cultures (wt/wt, n = 18; eB3^−/−\wt, n = 30; eB3^−/−\wt+DN-MEK, n = 34). (J) Neurons transfected with GFP and vector control or eB3 shRNA at DIV 0 [anti-GFP (green), anti-Erk1/2 (red), and anti-VGlut1 (blue)]. (Scale bar, 5 μm.) Arrowheads indicate VGlut1 puncta. Straight arrows indicate cytoplasmic Erk1/2; curved arrows indicate nuclear Erk1/2. (K) Quantification of percent nuclear Erk1/2 after transfection with vector control (n = 6 transfections) or eB3 shRNA (n = 6 transfections). A total of 50–300 cells were scored for each experiment. (L) Dendrites of control or eB3 shRNA–expressing neurons stained for GFP (green), Erk1/2 (red), and vGlut1 (blue). Arrowheads indicate VGlut puncta in transfected neurons. (M) Quantification of ratio of synaptic Erk to dendritic Erk staining in control and eB3 shRNA-expressing neurons. Vector control (n = 22), eB3 shRNA #1 (n = 24). (J–M) Experiments were conducted following a 1-h treatment with blockers of neuronal activity (1 μM TTX), NMDA receptors (50 μM APV), and L-type calcium channels (50 μM nifedipine). (Scale bars, 3 μm except in J.) *P < 0.03, **P < 0.01, ***P < 0.0001. Error bars indicate SEM.

To test whether the putative Erk-D domain that we identified mediates the association between eB3 and Erk, we conducted a GST-pulldown assay from E19 rat brain. GST fusions of eB3 cytoplasmic domain effectively pulled down Erk, whereas a GST fusion with a point mutation in the putative Erk binding domain showed a significant reduction pull-down (Fig. 6 C and D). Consistent with our in vivo results, quantification of the relative pull-down revealed that Erk2 interacts with the intracellular domain of eB3 more strongly than Erk1 (Fig. 6 C and D).

We next asked whether the putative Erk-D domain in eB3 is necessary for eB3-dependent control of synapse density. We generated a full-length HAeB3R construct in which L293 was converted to alanine (HAeB3_L293A, Fig. 2B) that reached the surface of HEK293T cells and had no effect on mEPSC frequency when expressed in neurons alone. However, unlike coexpression of a full-length eB3 construct insensitive to the shRNA, coexpression of HAeB3R_L293A with eB3 shRNA failed to rescue the decrease in mEPSC frequency or synapse density measured by colocalization of synaptic marker proteins (Fig. 2B and Fig. S7). These findings suggest that the eB3–Erk interaction is required for eB3 regulation of synapse density.

To test whether eB3 dependent regulation of synapse number relies on activation or inhibition of the MAPK pathway in cultured cortical neurons, we activated or inhibited Erk1/2 signaling by expressing a dominant negative (DN) or constitutively active (CA) MEK (DN-MEK or CA-MEK) (18). Expression of either CA-MEK or DN-MEK alone in neurons had no significant effect on mEPSC frequency (Fig. 6E). In neurons expressing shRNAs targeting eB3, expression of CA-MEK had no effect on mEPSC frequency, but expression of DN-MEK rescued the decrease in mEPSC frequency and reduction in synapse density (Fig. 6 E–G and Fig. S4D). To test whether Erk has a similar function in our heterogenotypic neuronal cell culture system, we coexpressed DN-MEK and GFP in neurons cultured from eB3^−/− mice and mixed these cells with neurons from wt littermates. Remarkably, expression of DN-MEK in neurons from eB3^−/− mice was sufficient to rescue defects in synapse density normally found in eB3^−/−/wt heterogenotypic cultures (Fig. 6 H and I), suggesting that the Ras/MAPK pathway is a negative regulator of synapse density and that eB3 negatively regulates this pathway.

Extracellular signaling can cause the translocation of Erk to the nucleus and activation of downstream effector molecules. Therefore, we tested whether eB3 kd would alter Erk localization. Although, in control-transfected neurons, Erk1/2 was excluded from the nucleus, kd of eB3 resulted in a significant increase in the percentage of neurons with nuclear Erk1/2 localization (Fig. 6 J and L). In addition, eB3 kd reduced the amount of Erk staining colocalized with synaptic markers (Fig. 6 K and M). These results indicate that the eB3–Erk interaction is important for the subcellular localization of Erk1/2 and suggest that eB3 may regulate synapse number in part by preventing translocation of Erk to the nucleus or by retaining Erk at synapses.

Discussion

Our data suggest that control of synapse density and perhaps neuronal excitability may not be left only to activity-dependent homeostatic mechanisms, but are enabled by activity-independent competitive mechanisms as well. Analogous to morphogen protein gradients in embryonic development, eB3 regulation of synapse density depends on the amount of eB3 expressed rather than simply whether or not it is expressed. How might eB3 expression influence synaptic density? The heterogenotypic cultures provide evidence for a model in which the relative levels of eB3 expressed between neighboring neurons mediate competition for a presynaptic ligand, EphB2, enabling neurons expressing higher levels of eB3 to form more synapses. Future studies will be needed to determine how eB3 expression is controlled and whether mismatches in EphB2–eB3 interactions between two neurons determine synapse density.

EB3 kd reduces both spine and synapse density, whereas neurons from eB3^−/− mice only have fewer spines. The lack of an effect on synapses in eB3^−/− mice is consistent with our in vitro findings and suggests that eB3 has two distinct functions: noncompetitive control of spine formation and competitive control of excitatory synapse density. Notably, expression of DN-MEK rescues defects in heterogenotypic cultures of eB3^−/− and wt neurons, indicating eB3-dependent control of synapse density relies on Erk signaling. In eB3^−/− mice, the loss of spines and the corresponding increase in excitatory shaft synapses in eB3 mice is intriguing and should enable elucidation of the in vivo role of dendritic spines.

Our experiments establish eB3 as a member of a small group of transmembrane proteins that can interact with Erk (19, 20) and provide evidence for a biological function of the interaction. The Erk2–eB3 interaction appears to negatively regulate Erk signaling possibly by retaining Erk in the dendrites, or by virtue of eB3’s selective interaction with Erk2. One model suggested by our findings is that Erk1 and Erk2 act to oppose each other. However, more work is needed to explore Erk-dependent control of synapse density. In conclusion, we provide evidence that eB3 controls synapse density through a competitive cell–cell interaction with EphB and by the negative regulation of MAPK signaling.

Materials and Methods

Cortical Cell Culture and Transfection.

Primary dissociated rat cortical neurons and transfections were conducted as described previously (10, 15). Primary dissociated cortical neurons from eB3^−/− mice and wt littermate controls were prepared from postnatal day 1 (P1) mice. Organotypic slice cultures were prepared and transfected as previously described (10).

Immunocytochemistry.

Cells and tissue were fixed and stained as described previously (10, 15). Antibodies used were: chicken anti-GFP (1:2,500; Upstate), rabbit anti-Erk1/2 (1:500; Promega), rabbit anti-eB3 (1:50; specificity confirmed by observing loss of staining after incubation with acetone-precipitated proteins from wt but not eB3^−/− brain tissue; Zymed), mouse anti–PSD-95 (1:250; Affinity BioReagents), rabbit anti-SynGAP (1:1,000; Affinity BioReagents), and guinea pig anti-VGlut1 (1:5,000; Chemicon).

Electrophysiology.

Whole-cell recordings were made from rat cortical neurons transfected in suspension as described previously (10).

Heterologous cell coculture and shRNA constructs were previously described (9, 10, 15). All data were collected from a minimum of three independent experiments, and data from genetically engineered mice were collected from a minimum of three animals per condition.

Additional details are provided in SI Materials and Methods.