Modulation of excitatory synapse number by sparse knock-down of neuroligin-1
To determine if postsynaptic NL1 levels regulate synapse development in vivo, we induced RNA interference (RNAi) to knock-down NL1 in cortical neurons using in utero electroporation. Electroporation at embryonic stage 15.5, when progenitors for layer 2/3 cortical neurons are accessible, results in sparse transfection (up to ~20%) of layer 2/3 pyramidal neurons while allowing neurons to develop in vivo under largely normal network activity and connectivity ().
Sparse knock-down of NL1 but not NL1 global knockout reduces synapse number and spine density in cortical layer 2/3 pyramidal neurons
Plasmids encoding small-hairpin RNAs (shRNA) with sequence homology to mouse NL1 were designed and purchased (see methods). Constructs were tested in vitro for knock-down of an NL1-EGFP fusion protein in HEK293 cells (). The most effective construct, sh-NL1 #7, was used for the majority of subsequent experiments and is referred to as sh-NL1 below. This construct was effective in neurons, as transduction of dissociated cortical cultures with lentiviruses encoding the plasmid strongly reduced endogenous NL1 levels (). Reduction of NL1 expression did not significantly alter levels of the family members NL2 and NL3 ().
Examination of dendritic spines of sh-NL1 expressing neurons in acute slices prepared from postnatal day 17–21 in utero
electroporated mice revealed that spine length and head area were increased and spine density was reduced compared to control EGFP transfected neurons (Supplementary Fig. 1a
and ; sh-NL1 and control: 0.50±0.06, 0.91±0.04 spines/μm, 11–17 neurons, 25–27 dendrites, p<0.05). Cotransfection of sh-NL1 and human NL1 (hNL1), which contains sequence alterations in the region targeted by sh-NL1, suppressed the effects of NL1 knockdown (Supplementary Fig. 1a
and ; 0.93±0.05 spines/μm, 10 neurons, 22 dendrites, p>0.05 vs. control), indicating that spine changes in sh-NL1-expressing neurons were due to loss of NL1. Similar morphological changes were observed in biolistically transfected hippocampal CA1 pyramidal neurons in organotypic slice cultures (Supplementary Fig. 2
The frequency of miniature excitatory postsynaptic currents (mEPSCs) in NL1 knock-down neurons measured by whole-cell voltage-clamp was reduced without significant effect on their amplitude (; sh-NL1 and control: amplitude: 7.40±0.54 pA, n=8, 8.36±0.41 pA, n=10, p>0.05; frequency: 0.82±0.12 Hz, n=8, 1.40±0.17 Hz, n=10, p<0.05). These effects were prevented by cotransfection hNL1 (; amplitude: 8.71±0.42 pA, frequency: 1.81±0.20 Hz, n=10, p>0.05 vs. control), confirming the NL1-dependence of the effects on synapse number.
Nevertheless, similar effects were not observed in layer 2/3 pyramidal neurons of NL1 knock-out (KO) mice, which had no spine morphology or density changes compared to those in wildtype (WT) animals ( and Supplementary Fig. 1b
: 0.91±0.1 spines/μm, 9 neurons/19 dendrites, NL1−/−
: 0.88±0.05 spines/μm, 10 neurons/22 dendrites, p>0.05). Importantly, introduction of sh-NL1 into NL1−/−
neurons had no effect on the structure and density of spines (Supplementary Fig. 1b
and ; NL1−/−
+ sh-NL1: 0.87±0.05 spines/μm, 10 neurons/19 dendrites, p>0.05 for each vs. NL1+/+
), confirming that the effects of sh-NL1 in WT animals were due to the loss of NL1 and not to possible off-target effects of the shRNA. Similarly, no changes in mEPSC amplitude and frequency were observed (; NL1+/+
, and NL1−/−
+sh-NL1: frequency: 1.35±0.16 Hz, n=8, 1.28±0.13, n=9, 1.30±0.1, n=8; amplitude: 9.0±1.5 Hz, n=8, 8.8±1.7, n=9, 8.7±1.3, n=9). Thus reduction of NL1 levels in a sparse subset of cortical neurons alters synapse number whereas global knockout of the gene has no effect. Importantly, both sets of experiments were carried out in the same cell type and in the same in vivo
Neuroligin-1 modulates NMDAR-mediated currents and Ca2+ transients of individual postsynaptic terminals
A possible mechanism for NL1-dependent modulation of synapse number is by reduction of NMDARs, which regulate synapse structure and function via a variety of mechanisms and whose activation triggers activity-dependent synaptogenesis in developing layer 2/3 pyramidal neurons35
. To determine if the functional properties of individual postsynaptic terminals are differentially affected by sparse vs. global manipulations of NL1, we used glutamate uncaging to examine AMPAR- and NMDAR-mediated currents and Ca2+
influx. Whole-cell recordings were obtained from layer 2/3 pyramidal neurons using intracellular solutions containing Alexa Fluor-594 (20 μM) to visualize morphology and a Ca2+
indicator, Fluo-5F (300 μM), to monitor intracellular Ca2+
. MNI-glutamate (5 mM) in the extracellular solution was photolysed to release glutamate by 2-photon excitation with 0.5 ms-long 720 nm laser pulses. To improve voltage-clamp and monitor single terminal AMPAR and NMDAR signals, voltage-gated potassium, sodium, and Ca2+
channels were blocked with a cocktail of antagonists (see methods). Uncaging glutamate near a visualized spine elicited uncaging-evoked AMPAR- and NMDAR-EPSCs (AMPAR-uEPSCs and NMDAR-uEPSCs) that were measured by holding cells at −60 and +40 mV, respectively (). Simultaneous measurement of green fluorescence was used to monitor Ca2+
transients in the active spine and neighboring dendrite () at −60 mV. Under these conditions Ca2+
enters the spine through NMDARs which are not fully blocked by extracellular Mg2+ 36, 37
Neuroligin-1 modulates NMDAR-mediated currents and Ca2+ signaling at individual postsynaptic terminals
Voltage-clamp recordings from sh-NL1-transfected neurons did not reveal significant differences in AMPAR-uEPSCs compared to control (; sh-NL1 and control: 14.0±1.6 pA, n=25; 14.2±1.7, n=20, p>0.05). However, at these same postsynaptic terminals NMDAR uEPSCs (sh-NL1 and control: 5.0±0.8 pA; n=25, 13.7±1.8 pA, n=20, p<0.05) and Ca2+
transients (sh-NL1 and control ΔG/Gsat
: 4.1±0.6%, n=25; 7.6±1.0%, n=20, p<0.05) were smaller in sh-NL1 expressing neurons (), consistent with reduced NMDAR content in individual spines. Both NMDAR-uEPSCs and Ca2+
influx were restored or increased beyond control levels by coexpression of sh-NL1 and hNL1 (, NMDAR-uEPSCs: 18.1±2.4 pA, n=21, p>0.05 vs. control; ΔG/Gsat
: 15.3±1.4%, n=21, p<0.05). Similarly, larger NMDAR-uEPSCs and Ca2+
transients were measured from spines of cells transfected with hNL1 alone (; NMDAR-uEPSCs: 26.0±2.7 pA, n=21, p<0.05; ΔG/Gsat
: 19.9±1.7%, n=21, p<0.05). This positive correlation between NL1 levels and NMDAR-mediated synaptic signals suggests that NL1 facilitates incorporation or retention of NMDARs in the postsynaptic terminal, consistent with previous studies31, 33, 38
The peak amplitude of uEPSCs measured at −60 mV was not modulated by NL1 levels, but its decay was slowed in hNL1-transfected neurons (). To determine if this prolongation resulted from alterations of AMPAR properties or if it represented an unusual contribution of NMDAR currents at resting potentials, we repeated recordings in the presence of the NMDAR antagonist CPP (10 μM). CPP application abolished the spine and dendrite Ca2+ signals as well as the slow component of uEPSC, confirming that all were due to NMDAR activation ().
Parallel analyses were carried out in NL1−/−
mice (), in which NMDAR/AMPAR current ratios have been previously reported to be reduced in hippocampal CA1 pyramidal neurons, amygdala principal neurons, and striatal medium spiny neurons 19, 24, 39
. Consistent with these reports, we found that glutamate uncaging evoked AMPAR-uEPSCs measured from individual spines of layer 2/3 pyramidal neurons were not different between NL1−/−
littermate mice, whereas NMDAR-uEPSCs were significantly smaller in NL1−/−
mice (; NL1+/+
: AMPAR-uEPSCs: 10.7±1.3 pA, n=24, 8.6 ± 1.2 pA, n=25, p>0.05, NMDAR-uEPSCs: 9.9±1.2 pA, n=24, 4.0±0.8 pA, n=25, p<0.05). NMDAR-mediated spine Ca2+
influx was also reduced (; NL1+/+
: 8.8±1.1%, n=24, 5.2±0.6 pA, n=25, p<0.05). Furthermore, introducing sh-NL1 into NL1−/−
KO neurons had no effect on NMDAR-uEPSCs and Ca2+
influx, indicating that, as expected for an NL1-dependent phenomenon, sh-NL1-mediated effects were occluded by constitutive loss of NL1 (; NL1−/−
+sh-NL1: AMPAR-uEPSCs, NMDAR-uEPSCs, and spine ΔG/Gsat
: 11.0±1.3 pA, 5.2±0.6 pA, and 4.9±0.5%, n=20, p>0.05 for each vs. NL1−/−
). Thus, the effects of NL1 loss on synaptic NMDARs are similar in the global knock-out and RNAi-induced sparse knock-down, indicating that the level of NL1 in each cell intrinsically regulates NMDAR-signaling but not excitatory synapse number.
Constitutive NL1 knock-out lowers NMDAR-uEPSCs and Ca2+ transients
Neuroligin-1 levels modulate glutamate-induced spinogenesis
Overexpression of NL1 in dissociated neuronal cultures influences synapse number in an activity-dependent manner19
, suggesting that NL1 regulates the selection of synapses after initial synapse formation. To determine if NL1 also regulates initial synapse formation, we utilized a glutamate uncaging protocol that triggers the rapid and de novo
formation of a spine and the establishment of a new synapse ()35
. This process requires activation of dendritic NMDA receptors, which are perturbed by changes in NL1 expression (see below).
NL1 regulates activity-dependent spinogenesis
In wild-type animals, sparse knock-down of NL1 in layer 2/3 pyramidal neurons by RNA interference lowered, whereas overexpression of NL1 increased, the success rate of new spine generation (). The magnitude of the effects depended on the frequency of stimulation such that NL1 overexpression enhanced the low probability of spinogenesis seen with low-frequency stimuli whereas down-regulation of NL1 decreased the high success rate triggered by higher frequency stimuli. In contrast, the same class of neurons in NL1 KO mice displayed normal activity-dependent spinogenesis. Furthermore, the normal synaptogenetic potential of neurons in NL1 KO mice occurs despite a ~50% reduction in dendritic NMDAR currents (; NL1+/+ and NL1−/−: NMDAR-uEPSCs: 10.0±2.1 pA, n=15, 5.9±0.9 pA, n=15, p<0.05), which was not different from that seen in shNL1 transfected neurons in WT animals (; 4.7±0.9 pA, n=16, p>0.05). Thus, sparse but not global manipulations of NL1 modulate the threshold of activity-dependent spinogenesis, likely explaining the parallel observations in synapse and spine number at later developmental stages.
Excitatory synapse number is regulated by relative levels of Neuroligin-1
The findings that synapse number and structure are altered in neurons with NL1 knock-down but not in neurons with NL1 KO may be explained by a transcellular competitive mechanism40, 41
. In this model, a cell expressing higher levels of NL1 relative to its neighbors has an advantage in forming synapses. For example, each cortical neuron might compete in NL1-dependent manner with surrounding neurons to establish proper connectivity with a limited number of presynaptic boutons. If correct, this mechanism explains why manipulations that eliminate NL1 from all neurons fail to recapitulate the perturbations seen in genetically mosaic tissue.
To test this model, we performed co-culture experiments in which NL1+/+ and NL1−/− neurons were mixed. We used in utero electroporation to transfect EGFP into layer 2/3 pyramidal neurons of NL1 KO mice. Cultures of cortical neurons were prepared by mixing, at specific ratios, cells dissociated from these manipulated NL1 KO mice and age-matched wild-type mice (). Consistent with the competition hypothesis, neurons in pure cultures of NL1−/− or NL1+/+ cells had similar spine densities (; NL1+/+ and NL1−/−:1.03±0.03 spines/μm, 37 fields of view, 1.01±0.05 spines/μm, 21 fields of view, p>0.05). However, when NL1−/− cells were mixed 1:1 with NL1+/+ neurons, spine density in the NL1−/− cells was reduced (; 0.58±0.05 spines/μm, 16 fields of view, p<0.05). Spine density was further reduced when the ratio of NL1−/− to NL1+/+ cells was reduced to 1:5 (; 0.39±0.04 spines/μm, 18 fields of view, p<0.05). Thus, the spine density of NL1−/− cortical layer 2/3 pyramidal neurons in culture depends on the fraction of co-cultured neurons that express NL1, indicating that transcellular interactions determine synapse number. These results are obtained without use of RNA-interference, demonstrating context-dependent defects in synapse numbers in neurons with constitutive genetic loss of NL1.
Spine density of NL1−/− neurons in vitro is affected by presence of neighboring NL1+/+ neurons
Levels of endogenous neuroligin-1 expression vary across cortical neurons
To understand whether neuron-to-neuron variability in NL1 expression in vivo could support the competitive model presented above, we measured mRNA levels by fluorescence in situ hybridization (ISH) across cortical neurons (). NL1 mRNA was detected throughout all cortical layers without layer-specific expression (). To determine the degree of variation of NL1 expression in cortical layer 2/3, we performed two-color fluorescence ISH of NL1 and GAPDH (). Levels of endogenous NL1 mRNA expression show large cell-to-cell variation compared to GAPDH (), resulting in a larger coefficient of variation (GAPDH and NL1: 26.1±1.6%, n=10 fields and 37.7±4.3%, n=10, p<0.05).
Variable NL1 mRNA in across cortical neurons
Gradients of NL1 levels across cortical neurons regulate spine number
To directly test in vivo
if transcellular gradients of NL1 expression level regulate synapse number, we examined a variety of conditions where NL1 expression is higher or lower in a cell few relative to neighbors (Supplementary Fig. 4
). First, we examined whether spine density is regulated in a dose-dependent manner by NL1. We took advantage of a specific sh-NL1, sh-NL1#9, which partially reduced NL1 levels (). When transfected into neurons of WT animals, sh-NL1#9 reduced spine density only slightly (0.7±0.03 spines/μm, 27 dendrites, p<0.05 vs. wildtype). In contrast a second very efficient shRNA, sh-NL1#4, greatly reduced spine density (0.4±0.04 spines/μm, 15 dendrites, p<0.05 vs. wildtype).
Second, if the magnitude of transcellular gradients in NL1 levels determines the density of excitatory synapses, then spine density in a sparse subset of neurons that over-express NL1 in NL1−/− mice should be higher than in neurons that over-express NL1 in wild-type mice. Indeed, spine density in hNL1 transfected neurons in NL1−/− mice was very high (; 1.9±0.1 spines/μm, 10 neurons, 17 dendrites, p<0.05 vs. hNL1 in WT, 1.4±0.1 spines/μm, 11 neurons, 22 dendrites). Increased density of spines was accompanied by increased mEPSC frequency, indicating that more functional synapses had been made (; hNL1 in NL1−/−, 3.67±0.35 Hz, n=8, p<0.05 vs. hNL1 in WT, 2.19±0.43 Hz, n=6).
Relative levels of NL1 determine spine number in vivo via intercellular interactions
Furthermore, across many manipulations the magnitude of changes in spine density and mEPSC frequency induced by manipulations of NL1 depend on the NL1 content of neighboring neurons. The effects of NL1 knock-down are reduced in the NL1+/− hemizygote mice and completely absent in NL1−/− mice (; sh-NL1 in WT, NL1+/−, and NL1−/−: percent change in spine density: −42±6%, −30±3%, −1±5% compared to the matched background; percent change in mEPSC frequency: −58±12%, n=8, −27±12%, n=8, 1±11%, n=9). Conversely, the effects of overexpression of NL1 are more dramatic when NL1 levels are reduced in neighboring cells (; hNL1 in WT, NL1+/−, and NL1−/−: percent change in spine density: 47±5%, 68±6%, 102±9% compared to the matched background; percent change in mEPSC frequency: 79±43%, n=6, 155±30%, n=10, 187±28%, n=8). The changes in spine density and synapse number seen with perturbation of NL1 levels are not due to variability in these parameters across animals, as they are observed when WT neurons, sh-NL1 expressing, and hNL1 overexpressing neurons in the same slice are compared () (spine density: sh-NL1, 0.53±0.06 spines/μm, n=12 neurons/22 dendrites; control, 1.06±0.05, n=9/15; hNL1, 1.40±0.05, n=8/19. mEPSC frequency: sh-NL1, 0.37 ± 0.17 Hz, n=5; control, 1.47±0.18, n=7; hNL1, 2.18±0.30, n=5 neurons. mEPSC amplitude: sh-NL1, 9.3±0.8 pA; control, 10.8±0.5; hNL1, 9.4±0.4 pA). Thus, spine density and mEPSC frequency in each cell are determined not by the cell’s absolute level of NL1 but the difference in its levels of NL1 relative to neighboring neurons.