During post-natal development, the formation and elimination of glutamatergic synapses are thought to be reflected in the growth and retraction of dendritic spines. In cortical pyramidal neurons, waves of new spine growth (spinogenesis) and synapse formation (synaptogenesis) occur at specific developmental stages, followed by pruning as the brain matures5
. Many signals have been proposed to trigger and regulate de novo
spine growth in a developing circuit including neurotrophins, neurotransmitters, and cell adhesion molecules6–9
. To uncover the triggers for and mechanisms of spinogenesis, we imaged dendrites of EGFP-expressing cortical layer 2/3 pyramidal neurons while releasing glutamate at a specific dendritic location by 2-photon laser induced-photolysis of MNI (4-methoxy-7-nitroindolinyl)-glutamate (). Analysis was performed in acute cortical brain slices from young mice (postnatal day (P) 8~12), a time period in which spinogenesis occurs in vivo10
De novo spine generation is induced by glutamate uncaging
Stimulation near edge of a dendrite with forty 0.5 ms laser pulses at 0.5 Hz in a Mg2+
-free extracellular solution induced growth of a new spine in ~14% of cases (, Supplemental Fig. 1
), showing the possibility of de novo
spinogenesis induced by glutamate exposure11
. Increasing stimulation frequency and laser pulse duration while maintaining the total number of stimuli at 40 increased the rate of spinogenesis such that at 5 Hz with 4 ms duration, a maximal success rate of ~50% was achieved (). Nascent spines arose from the dendrite where glutamate was released with high specificity () such that more than 70% of them grew within 1 μm of the uncaging spot () and 94% of them grew on the side of the dendrite exposed to glutamate.
In 128 of 132 examples of glutamate-induced spinogenesis, the spine was seen to emerge without a filopodial stage (See Supplementary Fig. 2a
for an exception). Instead, spine growth occurred incrementally but explosively such that the spine head volume increased from 10 to 90% of maximum within 11.8 ± 1.5 pulses of glutamate (5.9 ± 0.8 sec at 2 Hz stimulation) (, Supplementary Fig. 3
). The final sizes and lengths of the newborn spines were heterogeneous but not different from those of pre-existing neighboring spines (). The lifetime of newly formed spines was variable such that ~20% lasted less than 2 minutes but those that lasted 5 minutes were stable and remained for at least 30 minutes (Supplementary Fig. 4
). Thus, these newly formed spines either did not require continued exposure to glutamate for maintenance or they received glutamate from an alternative source such as an axonal bouton.
New spines grow rapidly and acquire morphology similar to pre-existing spines
Glutamate-induced spinogenesis was restricted within postnatal developmental such that its efficiency diminished by P14–15 and it failed to occur by P19–20 (Supplementary Fig. 5
). This was not due to decreased glutamate receptor activation in older animals since the uncaging-evoked EPSC was larger at P19~20 than at P10~12 (Supplementary Fig. 6
Previous ultrastructural studies have revealed a high frequency of dendrite shaft synapses in hippocampus in early postnatal life that decreases as spinogenesis occurs12
, leading to a model of synaptic development in which synapses are initially formed directly onto the dendritic shaft and a spine subsequently grows from this point with the synapse attached. On the other hand, rapid movement of a physically connected spine head and axonal bouton together through a complex neuropil is difficult to reconcile with the high density of crossing axons and dendrites13
. Our data demonstrate that glutamate uncaging-induced spinogenesis occurs with high spatial specificity and probability on the side of the dendrite exposed to glutamate. These findings place a lower limit of ~ 1/μm for the density of dendritic shaft synapses required to support this model of spinogenesis. To estimate the number of dendritic shaft synapses, Ca2+
imaging was performed in conditions in which the majority of synapses formed onto a stretch of dendrite were activated (~90%, Supplementary Fig. 7
). Under these conditions, we observed hotspots of Ca2+
influx in spineless stretches of dendrite at a density of 0.05/μm. This corresponds to a density of dendritic shaft synapses containing NMDARs that is approximately 20 fold less than necessary to explain the specificity and efficiency of glutamate-induced spinogenesis (see Supplementary Fig. 7
for further discussion).
The high success rate of spinogenesis induced by glutamate uncaging allows identification of the signaling pathways that couple activity to spine growth (). Previous analyses of spine generation induced by electrical stimulation indicate a requirement for NMDA-type glutamate receptors (NMDARs) in this process14,15
and at this age NMDARs are found throughout the dendrite (Supplementary Fig. 8
). Preventing NMDAR activation with the antagonist CPP nearly abolished spinogenesis whereas it was unaffected by inhibiting AMPA/kainate glutamate receptors with NBQX (). The voltage-gated sodium channel antagonist TTX also did not affect spinogenesis, discarding the possibility that postsynaptic action potentials are necessary. Addition of extracellular Mg2+
significantly decreased the success rate, suggesting that the degree of current flux through NMDARs plays a crucial role in triggering spine formation. Blocking either mGluR1 or mGluR5 using CPCCOEt or MPEP or with the less selective group I mGluR antagonist AIDA demonstrated that neither was strictly necessary for spinogenesis. Lastly, depleting Ca2+
stores with cyclopiazonic acid (CPA) or thapsigargin (TG) significantly inhibited spine formation. Thus, our data indicate that NMDARs with additional contributions from intracellular stores provide the coupling between glutamate and activation of intracellular pathways responsible for spinogenesis.
Molecular mechanisms of glutamate-induced spine formation
We further considered intracellular signaling pathways that might be activated in a spatially-delimited fashion by glutamate and could provide the spatial information necessary for local spine growth. Previous studies of long-term potentiation (LTP) and associated spine enlargement in older animals demonstrated a need for activation of CaMKII or signaling via neurotrophin receptor tyrosine kinases such as the BDNF receptor TrkB16–19
. However, we find that activity-dependent spinogenesis is unaffected by the kinase inhibitors KN-62, KN-93, and K252a, indicating independence from CaMKII and TrkB signaling ().
The cAMP-activated kinase PKA is required for LTP induction in younger neurons20
and gradients in its concentration can be maintained over micron length scales21
. Raising cAMP concentration by applying the adenylate cyclase activator forskolin was not sufficient to generate spines on its own (n=3, data not shown), but glutamate uncaging in its presence increased the spinogenesis success rate to ~80% (). In the presence of forskolin, multiple spines often grow per induction attempt (), including at sites more distant from the uncaging location (). In addition, the PKA inhibitor H-89 prevented new spine growth (), indicating that PKA activity is necessary but not sufficient for spinogenesis.
The small guanosine triphosphatase Ras is activated by Ca2+
influx through NMDARs and signals via mitogen-activated protein kinase (MAPK) to promote LTP induction17,22
. We found that the MAPK pathway was necessary for spinogenesis since the success rate was significantly reduced by blocking the upstream activator of MEK1/2 with U0126 and MEK1 with PD98059 (). In pyramidal neurons, activated Ras diffuses from active spines to neighboring spines and heterosynaptically facilitates plasticity17
. To examine if a similar phenomenon potentiates spinogenesis, we delivered a “priming” stimulus to a pre-existing spine (40 pulses at 2 Hz, ) and then a “test” stimulus to the dendritic shaft within 1~2 minutes. In contrast to previous studies of older neurons in hippocampus16–18
, we found no consistent increases in volume of the existing spine in response to this priming stimulus (), supporting that spinogenesis is not simply due to the growth of an undetectable preexisting spine. Nevertheless, the success rate of spine generation was enhanced by the priming stimulus in a MEK1/2 dependent manner such that the low level of spinogenesis evoked by a 0.5 Hz test stimulus was increased. In contrast, the priming stimulus did not increase the success rate when the test stimulus was delivered at 2Hz (), indicating that the priming protocol shifted the induction threshold for spinogenesis.
With the exception of the NMDAR and PKA antagonists, none of the pharmacological manipulations that altered spinogenesis rates affected dendritic currents or Ca2+
transients (Supplementary Fig. 8
). As expected, CPP largely abolished the Ca2+
transient and the prolonged phase of the currents. Consistent with a facilitation of Ca2+
influx through NMDARs by PKA23,24
, H-89 lowered dendritic Ca2+
transients by ~20% (Supplementary Table 1
). However, this effect is insufficient to explain the near abolishment of spinogenesis since the rate of spinogenesis evoked by 0.5 ms uncaging pulses in control conditions was higher than that evoked by 4 ms pulses in H-89, despite eliciting smaller dendritic Ca2+
transients (Supplementary Fig. 8
and Table 1
We examined whether synaptic activity generated new spines in cortical tissue from young mice via similar pathways. In normal extracellular Mg2+
, high-frequency (100 pulses at 100 Hz, delivered twice separated by 10 seconds) but not low-frequency (10 Hz) electrical tetani rapidly triggered new spine growth (). Blockade of NMDARs or PKA prevented this synaptically-evoked spine growth () as well as spontaneous new spine growth that occurred when NMDAR activation was increased by removing extracellular Mg2+
and blocking inhibitory GABAA
receptors (Supplementary Fig. 9
). Hence, multiple experimental paradigms demonstrate that activity-dependent spinogenesis in developing cortex requires NMDAR- and PKA-dependent signaling.
In order to determine if nascent spines detect synaptically-released glutamate and are functionally incorporated into the circuit, we generated new spines by high frequency stimulation (HFS) and examined their synaptic responses using optical quantal analysis of synaptic properties. A whole-cell recording was obtained and the probability and amplitude of synaptically-evoked NMDAR-mediated Ca2+
transients in the spine head were monitored (). Using the Ca2+
sensitive green fluorophore Fluo-5F, we detected stimulus-evoked Ca2+
transients in the heads of newly-grown spines (), demonstrating that they sense synaptic activity within a neural circuit within 30 min after growth. Similar results were obtained in 5 of 7 new spines and in 11 of 16 preexisting spines. Analysis of the spines in which an evoked Ca2+
transient could be detected indicated that nascent spines displayed smaller and less frequent synaptically-evoked Ca2+
transients than pre-existing spines (). Similar results were obtained for new spines that grew in response to glutamate uncaging and probed using a glass stimulating electrode placed near the spine (Supplementary Fig. 10
Functional characterization of new spines
The AMPAR and NMDAR content of new spines was characterized using glutamate uncaging following HFS-induced spinogenesis. In all cases, we detected fast uncaging-evoked postsynaptic currents (uEPSCs) at a holding potential of -60 mV and large prolonged uEPSCs at +40 mV (), consistent with AMPAR- and NMDAR-mediated currents previously characterized in these cells25
. Similarly, uncaging-evoked Ca2+
transients were clearly visible at -60 mV () but were larger at +40 mV (not shown), consistent with the known properties of NMDAR-mediated Ca2+
influx. Despite similar-sized AMPAR-uEPSCs in new and preexisting spines, NMDAR-uEPSCs and Ca2+
influx were significantly lower in nascent spines compared to pre-existing spines (). Therefore, the smaller Ca2+
transients measured in nascent spines by synaptic activation likely reflects a lower number of postsynaptic NMDARs.
In this study, we established a protocol for the reliable and spatiotemporally precise induction of spinogenesis. These experiments demonstrate that glutamate is sufficient to trigger rapid spine formation and suggest that neurons use glutamate release to establish circuit wiring. Thus, these data support the hypothesis that axonal growth and glutamate release may be the triggering event in synapse formation such that axonal bouton localization is an important early step for precise neuronal circuit formation10,26
. Given the involvement of NMDARs, Ca2+
stores, cAMP, PKA, and MAPK in activity-dependent spinogenesis, it is likely that many neuromodulators that regulate these molecules may influence the capacity or threshold for new spine formation. For instance, activation of dopaminergic, serotonergic, or adrenergic receptors that signal via Gαs
may facilitate spinogenesis, whereas receptors that activate Gαi
-coupled signaling may function as inhibitory signals.
Lastly, we provide experimental evidence that spines that grow de novo
in developing cortical tissue become rapidly functionally integrated into the circuit such that they sense synaptically-released glutamate via AMPARs and NMDARs. Whether these nascent spines are rapidly physically associated with a presynaptic bouton and display the ultrastructural correlates of a synapse are unknown27–30
. Our results indicate that spines can grow de novo
without the need for a filopodial and likely without a dendritic-shaft synapse stage. In total, this study demonstrates that synaptic activity can rapidly modify neuronal connectivity with high accuracy by generating new circuit elements.