ABSTRACT
In the mouse olfactory bulb, sensory information detected by ∼1,000 types of olfactory sensory neurons (OSNs) is represented by the glomerular map. The second-order neurons, mitral and tufted cells, connect a single primary dendrite to one glomerulus. This forms discrete connectivity between the ∼1,000 types of input and output neurons. It has remained unknown how this discrete dendrite wiring is established during development. We found that genetically silencing neuronal activity in mitral cells, but not from OSNs, perturbs the dendrite pruning of mitral cells. In vivo calcium imaging of awake neonatal animals revealed two types of spontaneous neuronal activity in mitral/tufted cells, but not in OSNs. Pharmacological and knockout experiments revealed a role for glutamate and NMDARs. The genetic blockade of neurotransmission among mitral/tufted cells reduced spontaneous activity and perturbed dendrite wiring. Thus, spontaneous network activity generated within the olfactory bulb self-organizes the parallel discrete connections in the mouse olfactory system.
INTRODUCTION
Sensory information detected by sensory organs is represented by spatiotemporal patterns of activity in corresponding regions of the brain. In general, continuous physical information is spatially represented by continuous topographic maps, such as the retinotopic map for visual space and the tonotopic map for sound frequencies. The formation of continuous maps and their connectivity is known to be regulated by a combination of graded guidance cues and neuronal activity 1, 2. The coarse targeting of axons is controlled by the graded expression of guidance molecules such as ephrins and Eph receptors 3. Later on, spontaneous activity generated by the sensory organ (e.g., retina and cochlea) fine-tunes connections to produce a precise nearest-neighbour relationship from sensory to higher-order neurons 1, 4–8. In vivo imaging of the neonatal mouse brain demonstrated a propagating wave of activity originating from the retina into the entire visual system in the brain 9. Similarly, spontaneous activity generated in the developing cochlea propagates all the way to the auditory brain regions 10.
In contrast, the olfactory system has to detect a vast variety of odorants, which are mostly discrete in nature and convey distinct information. Odorants are detected by a large family of odorant receptors (∼50 in insects and ∼1,000 in mammals) which are expressed by olfactory sensory neurons (OSNs) 11. OSNs expressing the same type of odorant receptor converge their axons onto one or a few glomeruli in the olfactory bulb. As a result, olfactory information is represented by spatiotemporal patterns of activity in the glomerular map 12, 13. In mammals, sensory input into one glomerulus is relayed to 20-50 mitral and tufted cells, which project their axons to the olfactory cortices. Each mitral/tufted cell receives inputs from just one glomerulus, allowing for segregated odor information processing. A glomerulus is also innervated by various types of interneurons and centrifugal inputs, which perform dynamic odor tuning including gain control, lateral inhibition, pattern separation, and oscillatory activity for phase coding 14, 15. Thus, the segregated glomerular microcircuitry is an important platform for odor information processing and for subsequent behaviors.
Over the last two decades, the mechanisms of neuronal wiring specificity in the glomerular map have been intensively studied in Drosophila and mice. In the Drosophila olfactory system, the connectivity of OSN axons and second-order neurons is mostly genetically-determined with a variety of transcription factors and cell-surface molecules 16, 17. Notably, the wiring specificity in the Drosophila olfactory system is based on molecular matching between pre- and post-synaptic neurons 18, 19.
However, the wiring of the mouse olfactory system is more flexible in order to accommodate a much larger number of odorant receptors. For example, the ectopic expression of a new receptor leads to the formation of a new glomerular microcircuit in the olfactory bulb 20, refuting the possibility of a strictly genetically-determined regulation. It has been revealed that the axonal projections of OSNs are regulated by a combination of genetic programs, odorant receptor-derived cAMP signals, and neuronal activity, which together define the combinatorial expression of various axon guidance molecules 12, 21–26. Notably, the glomerular map is mostly formed by axon-axon interactions. However, these mechanisms are merely one side of the coin: The dendrite wiring of mitral/tufted cells also needs to be precisely controlled. It has been known that migration and the domain organization of mitral/tufted cells within the olfactory bulb is genetically-determined 27–29. Later on, the discrete dendrite wiring of each mitral cell to a glomerulus is established through a dendrite remodeling process. While mitral cells initially extend multiple dendrites, just one of them is strengthened to form a mature primary dendrite and the supernumerary ones are pruned during development 30. Activity-dependence has been an important issue, in relation to other sensory systems 31; however, previous efforts have all failed to find a contribution for OSN-derived neuronal activity in dendrite remodeling 32–34. Therefore, it has long been an enigma as to how the discrete wiring of mitral/tufted cell dendrites is established in the mammalian olfactory system 14. In the present study, we found that the dendrite pruning of mitral cells is in fact an activity-dependent process. Unlike other sensory systems, spontaneous activity is generated “within the olfactory bulb” and “self-organizes” the discrete connectivity of mitral cell dendrites.
RESULTS
Developmental remodeling of mitral cell dendrites
To understand mechanisms of dendrite wiring in mitral cells, we took a genetic approach using in utero electroporation 35–37. It has been known that mitral cells are generated between embryonic day (E) 10 and E13 29, 38. Therefore, we electroporated plasmid DNA at E12 to label a subset of mitral cells (Fig. 1b). To facilitate the reconstruction of fine neuronal morphology, we electroporated a Cre-dependent tdTomato plasmid and a small amount of Cre plasmid, which produced sparse and bright labeling (Fig. 1a). To avoid any biases in quantification, we obtained volumetric fluorescence image data from olfactory bulb tissues cleared with SeeDB or SeeDB2 39, 40, and analyzed mitral cells in the medial side of the olfactory bulb in an unbiased manner (Supplementary Fig. 1 and Supplementary Video 1). We only focused on labelled neurons in the mitral cell layer, and excluded tufted cells from our analyses.
A typical mitral cell has a single primary dendrite (also known as an apical dendrite) connecting one glomerulus and several lateral dendrites (also known as basal dendrites) extending within the external plexiform layer. During development, mitral cells initially extend multiple dendrites toward the glomerular layer 30, 32, 38, 41. At postnatal day 1 (P1), the labelled mitral cells extend 8.4 ± 2.4 (mean ± SD) primary dendrites to the glomerular layer (Fig. 1c,d) and 85% of these cells did not have a tufted structure at the end of any dendrite (Fig. 1e,f). By P3, the number of primary dendrites has decreased to 4.0 ± 1.0 (mean ± SD, Fig. 1c,d) while the number of cells containing at least one dendrite with a ramified tufted structure increased to 67.1%, with 30.2% containing two or more (Fig. 1f). We also examined the formation of synapses using PSDΔ1.2-GFP, which labels mature glutamatergic synapses without affecting synapse formation 42, 43. Prominent PSDΔ1.2-GFP puncta began to appear in tufted structures from P3 (Fig. 1e). Finally, at P6 most mitral cells have a single primary dendrite extending to a single (1.4 ± 0.7, mean ± SD) glomerulus (Fig. 1d), with 79.7% of cells containing a single tufted dendrite. Day-to-day quantification indicated that the pruning of supernumerary dendrites occurs between P4-5 (Supplementary Fig. 1c). The formation of lateral dendrites also occurred during P3-6.
Together, these observations indicate that the remodeling of mitral cell dendrites occurs in a step-wise manner (Fig. 1g). During Phase I (≤P3), the number of dendrites is gradually reduced to ∼4 and a tufted structure with excitatory synapses begins to appear in some of them. During Phase II (P3-6), just one dendrite is further strengthened to have a ramified tufted structure, while the others are pruned. Several lateral dendrites are also formed during this stage. Thus, dendrite pruning and synapse elimination occurs during Phase II at least in a subset of mitral cells.
Neuronal activity in mitral cells is essential to establish a single primary dendrite
Neuronal activity often plays an important role in neuronal remodeling in the mammalian brain 44. However, previous attempts to determine a role for neuronal activity in mitral cell remodeling have been unsuccessful 32–34, 41. As all of these studies manipulated neuronal activity only in OSNs, we directly manipulated neuronal activity in mitral cells by overexpressing a gene coding for the inward rectifying potassium channel, Kir2.1 45. Overexpression of Kir2.1 efficiently inhibited the neuronal activity of mitral cells in a cell-autonomous fashion (Supplementary Fig. 2a). Kir2.1 overexpression did not affect the initial growth (P1) and remodeling of dendrites at the early stage (Phase I). However, during Phase II, dendrite pruning was specifically perturbed (Fig. 2a,b). In control neurons, 78.1% of mitral cells connected to a single glomerulus at P6; however, only 19.3% of Kir2.1-expressing neurons connected to a single glomerulus at this stage. Even by P28, half of the mitral cells still connected to multiple glomeruli (Fig. 2b). The formation of lateral dendrites was also precluded (Fig. 2c). In contrast, the formation of mitral cell axons was not affected at P6 (Supplementary Fig. 2b). Tufted structures and PSDΔ1.2-GFP puncta were still visible in the primary dendrites of Kir2.1-expressing mitral cells, suggesting synapse formation (Fig. 2e). Thus, the pruning of supernumerary dendrites and the formation of lateral dendrites, which normally occur during Phase II, were specifically perturbed by Kir2.1 overexpression. Defective dendrite pruning is due to the hyperpolarizing effects of Kir2.1, because a non-conducting mutant of Kir2.1 45 did not affect the pruning process (Supplementary Fig. 2c). Overexpression of the bacterial sodium channel NaChBac which increases the excitability of neurons did not affect the dendrite remodeling process (Supplementary Fig. 2d) 46, although radial migration of the cell body was affected as reported for cortical pyramidal neurons (Supplementary Fig. 2e) 47.
Activity-dependent dendrite pruning is not just a specific feature of mitral cells labelled at a particular timing (E12) by in utero electroporation. We also overexpressed Kir2.1 in mitral cells generated earlier (in utero electroporation at E10). While these neurons potentially develop earlier than E12-labelled ones 29, Kir2.1 overexpression similarly prevented their dendrite pruning (Supplementary Fig. 2f). Thus, activity-dependent dendrite pruning is a common feature between early-born and late-born mitral cells. Hereafter, we only studied mitral cells labelled at E12.
Neuronal activity derived from OSNs is dispensable
Could the neuronal activity required for pruning be derived from OSNs? Previous studies failed to show a role for neuronal activity from OSNs, but with important caveats 32–34, 41. This may be due to different labeling and imaging techniques, an incomplete blockade of neuronal activity, and/or retardation of animal growth. We therefore re-examined the role of neuronal activity in OSNs using our analysis pipeline (Supplementary Fig. 1a).
Firstly, we examined the role of sensory-evoked activity, i.e., odor and mechanosensory responses in OSNs 48, 49. Both of these sensory stimuli can be effectively eliminated by unilateral naris occlusion from P0. We found no change in primary dendrite pruning at P6 (Fig. 3a). Thus, sensory-evoked activity is dispensable for the normal development of mitral cell dendrites.
We next examined the role of activity (both evoked and spontaneous) generated by the olfactory signal transduction cascade involving cyclic nucleotide-gated (CNG) channels, as this has also been known to contribute to spontaneous activity 22, 50. We used the OMP-Cre knock-in line to knockout CNGA2 specifically in OSNs. In the conditional mutant, dendrite pruning was incomplete at P6, but recovered to normal levels at P14 (Fig. 3b), consistent with an earlier study using the CNGA2 straight knockout 32. As this anosmic mutant animal shows severe growth retardation, likely due to poor suckling (Fig. 3c), this may explain the poor dendrite pruning. To further test the role of any kind of activity produced in OSNs, we generated another mouse line expressing tetanus toxin light chain (TeNT), which cleaves VAMP2 and thus inhibits synaptic vesicle release 51, 52. In OMP-Cre;R26-CAG-LoxP-TeNT (designated OSN-TeNT) mice, anti-VAMP2 staining signals in OSNs were completely eliminated (Fig. 3d). OSN-TeNT mice showed defective dendrite pruning at P6, but normal dendrite morphology at P14 (Fig. 3e). Again, OSN-TeNT mice demonstrated a severe growth retardation (Fig. 3f). Therefore, we could not conclude whether this indicates a role for activity in OSNs, or just a developmental delay in this mouse line.
To clarify this issue, we tried to rescue the growth retardation by hand rearing new-born animals. OSN-TeNT mice are apparently anosmic and cannot suckle very well. Therefore, we fed powdered milk to OSN-TeNT mice manually. We fed pups every 2 hours from P1-6 in a warm chamber (Fig. 3g,h). To facilitate digestion, excretion and to prevent bloating, we also massaged their stomach after feeding. This hand rearing partially rescued the growth delay (Fig. 3i), and these OSN-TeNT mice weighted similarly (98.9 ± 11.1%, mean ± SD) to hand-reared controls, and 62.2 ± 12.9% (mean ± SD) of maternally-reared controls. The hand-rearing also largely rescued defective dendrite pruning seen in mother-reared OSN-TeNT animals. Now, the difference between hand-reared control and OSN-TeNT mice was small, if any, at P6 (Fig. 3j). Thus, neuronal activity derived from OSNs plays a minimal role for the normal development of mitral cell dendrites.
Awake neonatal mice show OSN-independent spontaneous activity in mitral cells
Our genetic experiments raised the possibility that spontaneous activity in the olfactory bulb is important for dendrite pruning. However, to the best of our knowledge, spontaneous activity in neonatal animals (<P6) has never been characterized in the olfactory system. To directly examine whether spontaneous activity exists in mitral cells in vivo, we performed in vivo two-photon Ca2+ imaging of the olfactory bulb in awake newborn mice (Fig. 4a). We used mitral/tufted cell-specific GCaMP3 and GCaMP6f mouse lines (transgenic Pcdh21-Cre;Ai38 and Thy1-GCaMP6f, designated M/T-GCaMP3/6f hereafter) to record neuronal activity in mitral/tufted cell dendrites (glomerular layer) 49, 53, 54. In vivo imaging of awake neonatal mice at P4 revealed spontaneous neuronal activity in the glomerular layer of the olfactory bulb (Fig. 4b). Spontaneous activity was not seen in ketamine-anesthetized pups (Supplementary Fig. 3a-e), which was similar to earlier studies in other sensory systems 9, 10, 55. We also performed Ca2+ imaging of OSNs using an OSN-specific GCaMP3 mouse line (transgenic OMP-tTA;TRE-GCaMP3, designated OSN-GCaMP3 hereafter) at P4. However, we did not observe spontaneous activity in OSN axons in the glomerular layer at this age (Fig. 4c). Furthermore, robust spontaneous activity was observed even after naris occlusion (Fig. 4d) and in OSN-TeNT mice (Supplementary Fig. 3f). Thus, the spontaneous activity in the olfactory bulb can be generated without synaptic inputs from OSNs.
The age-specific patterns of spontaneous activity in vivo
We performed in vivo Ca2+ imaging of mitral/tufted cells during the dendrite remodeling process, from P1 to P6. The pattern of spontaneous activity changed dramatically during development. The spontaneous activity was synchronized among many glomeruli at P1-2, but became de-synchronized afterward (Fig. 4e,f, Supplementary Fig. 4, and Supplementary Video 2,3). We determined the correlation index for each time bin (1 frame, 0.429 or 0.564 sec) based on percentages of co-active glomeruli, which were also normalized by the firing frequency (See Methods). Based on the correlation index, each event was classified into highly correlated (H-) events and more sporadic lowly correlated (L-) events 56. A high ratio of H-events was observed only at P1-3, and L-events dominated afterward (Fig. 4g). At all stages, the activity was synchronized within each glomerulus as is known in the adult; this is likely due to dense electrical coupling via gap junctions within a glomerulus 57, 58. Spike frequency and amplitude was variable among glomeruli and ages (Supplementary Fig. 4b,c).
We also examined the spatial distribution of correlated events. We determined the correlation between two glomeruli normalized by their individual firing frequency (Spike Timing Tiling Coefficient, see Methods) 59, and examined the spatial distribution of glomeruli with correlated activity. At P1-2, correlated glomeruli were found within 100-150 μm, but not at later stages (≥ P3) (Supplementary Fig. 4d). This again demonstrates two different types of spontaneous activity, synchronized at an early stage and de-synchronized at a later stage.
Spontaneous activity is generated in isolated olfactory bulb in vitro
To examine whether the spontaneous activity is generated without any inputs from other brain regions, we performed Ca2+ imaging of acute olfactory bulb slices in vitro, thus removing any external inputs. The olfactory bulb was excised from M/T-GCaMP3/6f mice and the dorsal halves of the bulb were continuously perfused with artificial cerebrospinal fluid (ACSF) (Fig. 5a). Similar age-specific patterns of spontaneous activity were observed in this preparation as in vivo, both in the glomerular layer (Fig. 5b,c) and in mitral cell somata (Supplementary Fig. 2a). Notably, the pattern of spontaneous activity was synchronized at an early stage, and de-synchronized at a later stage (Fig. 5b,c, Supplementary Video 4,5), as was seen in vivo. H-events were more frequent at P2 than at P6 (Fig. 5d). We determined cross-correlation values for activity in all glomeruli imaged at P2 and P6 in vitro. A high correlation between glomeruli was found for many pairs at P2, but not at P6 (Fig. 5e).
At P2, the H-events were in fact propagating waves (Fig. 5b, Supplementary Fig. 5a-e,h, Supplementary Video 4) and were broadly, but not strictly, reproducible as seen in the cross-correlogram (Supplementary Fig. 5a-c). The propagation speed was much slower than those of action potentials (Supplementary Fig. 5e,h), suggesting that it is a multi-synaptic propagation of bursting activity. The spike frequency and amplitude were variable among glomeruli (Supplementary Fig. 6f,g).
NMDARs are required for the dendrite pruning cell-autonomously
So far, we have excluded the contribution of OSN-derived and centrifugal inputs as the possible origins of the spontaneous activity. Leading to the question, how is spontaneous activity generated and propagated within the olfactory bulb? We investigated the possible origin of spontaneous activity using pharmacological manipulations of the isolated olfactory bulb. Spontaneous activity was completely eliminated by tetrodotoxin (TTX), demonstrating that the Ca2+ signals are indeed generated by action potentials (Fig. 6a). When we applied inhibitors to NMDA-and AMPA-type ionotropic glutamate receptors (APV and CNQX, respectively), both the amplitude and frequency of spontaneous activity were reduced (Fig. 6b). Inhibitors for gap junctions also reduced the frequency of spontaneous activity (Fig. 6c, Supplementary Fig. 6a).
In the developing brain, GABA can also act as an excitatory neurotransmitter for some types of neurons, with high intracellular Cl− levels 60. However, the application of GABA shuts off the spontaneous activity of mitral cells at P2, suggesting that GABA is inhibitory toward mitral cells (Supplementary Fig. 6b). We also analyzed dendrite morphology in mice deficient for NKCC1, a Na+-K+-2Cl− co-transporter essential for maintaining a high Cl− concentration inside cells 61. We did not see any defects in dendrite pruning in the NKCC1 mutant mouse line (Supplementary Fig. 6c), further excluding the role of excitatory action of GABA in dendrite pruning.
We therefore investigated a role for glutamatergic synaptic transmission. Specifically, we examined a possible role for the NMDA glutamate receptor (NMDAR) in dendrite pruning. We performed a single-cell knockout of Grin1, which encodes an essential subunit of NMDARs, NR1. We electroporated a Cre plasmid into a subset of mitral cells in floxed-Grin1 mice and examined their morphology. We found that dendrite pruning was significantly perturbed in NMDAR-deficient neurons both at P6 and P14 (Fig. 6d,e). At P6, more than half of mitral cells maintained connections to multiple glomeruli. The formation of lateral dendrites was also slightly affected in the NMDAR-deficient neurons (Fig. 6f). Thus, NMDARs regulate the pruning of mitral cell dendrites in a cell-autonomous fashion.
Neurotransmission among mitral/tufted cells is the critical source of activity
Where is the glutamate coming from? In the isolated olfactory bulb, mitral/tufted cells are the only possible source of glutamatergic neurotransmission. They are known to interact through dendro-dendritic neurotransmission 14. We therefore tested the possibility that glutamatergic neurotransmission among mitral/tufted cells is required for the dendrite pruning. To block the neurotransmission among mitral/tufted cells, we generated Pcdh21-Cre;R26-CAG-loxP-TeNT (designated M/T-TeNT) mice, in which TeNT is specifically expressed in mitral/tufted cells. Immunoreactivity for VAMP2 was reduced in the external plexiform layer of the olfactory bulb (Fig. 7a). Contrary to the OSN-TeNT mice, M/T-TeNT mice did not show growth defects (data not shown). As we used the transgenic Pcdh21-Cre line, a small fraction of mitral/tufted cells may have failed to express Cre, as has been seen in the adult 49. Nevertheless, we confirmed a reduction in spontaneous activity by using in vivo Ca2+ imaging of awake M/T-TeNT pups at P6. The frequency of glomerular spikes were significantly reduced, but not eliminated in M/T-TeNT mice (Fig. 7b,c). The glomerulus-specific L events were particularly reduced (Supplementary fig. 7). Morphological analysis revealed that M/T-TeNT mice show a significant reduction in the percentage of mitral cells innervating a single glomerulus, both at P6 and P14 (Fig. 7d). Thus, glutamatergic neurotransmission among mitral/tufted cells is important for generating spontaneous activity, and for the normal dendrite pruning in Phase II.
DISCUSSION
Step-wise regulation of dendrite remodeling in mitral cells
While the discrete connectivity of mitral cell dendrites to a single glomerulus has been known since Camillo Golgi’s seminal work more than 140 years ago 62, the mechanisms behind the formation of this highly precise connection has been a long-standing mystery. In the present study, we found that spontaneous neuronal activity plays a key role in establishing the precise dendritic connectivity. This is in stark contrast to the strictly genetically-defined wiring of the Drosophila projection neurons 16. Together with the “one neuron-one receptor” and the “one glomerulus-one receptor” rules for OSNs 12, this discrete dendrite wiring ensures segregated olfactory information processing, from OSNs, expressing a defined odorant receptor, to distinct sets of mitral/tufted cells innervating a specific glomerulus (Fig. 7e).
The dendrite remodeling process of mitral cells can be divided into two major steps, while they may be partially overlapping in time (Fig. 1g). In the first step (Phase I), mitral cells reduce their exuberant dendritic processes, while strengthening their surviving dendrites by forming ramified tufted structures and excitatory synapses within one or a few glomeruli. This step occurs independently of neuronal activity. Indeed, the overexpression of Kir2.1 had minimal impacts on dendritic morphology at this stage, as well as on the initial formation of tufted structures and excitatory synapses. Thus dendrite remodeling at Phase I is likely controlled by a genetic program and/or physical and chemical interactions with OSN axons. For example, the ablation of OSNs has been known to impair the formation of the apical tuft 27, 63. It has also been reported that OSN-dependent Notch signaling regulates dendritic complexity during late embryonic stages 36. Thus, OSN axons may serve as a scaffold for subsequent glomerular circuit formation. During Phase II, one dendrite is exclusively strengthened to form a thick tufted dendrite, whereas the supernumerary ones are pruned. Neuronal activity is essential for the fine tuning of the dendritic connectivity at this stage, by selecting one out from a few possible targets. As wild-type neurons do not always form multiple tufted dendrites during development (30.3 % at P3) (Fig. 1f), the elimination of previously established tufted dendrites may occur only in this minor population of neurons. However, the overexpression of Kir2.1 results in formation of multiple primary dendrites in majority of mitral cells (80.7 % connecting to multiple glomeruli at P6). This may indicate that neuronal activity is required for the pruning of supernumerary dendrites that could otherwise become mature primary dendrites at a later stage. Activity may also be important for the maintenance of synapses, because Kir2.1-expressing mitral cells tend to show a poor tufted structure at a later stage (P28, data not shown).
Origins of spontaneous activity
The origins of spontaneous activity used in sensory circuit formation have been extensively studied in visual, auditory, and somatosensory systems. For example, in the visual system, propagating spontaneous activity (known as retinal waves) generated in the retinal ganglion cells regulates the fine tuning of the retinotopic map as well as eye-specific segregation in lateral geniculate nucleus (LGN) 1, 4, 8. In particular, the early postnatal retinal wave (Stage II wave) is driven by acetylcholine derived from starburst amacrine cells. In the auditory system, spontaneous ATP release activates hair cells, and subsequently afferent nerve fibers, contributing to tonotopic map formation 5, 10, 64. In the somatosensory system, spontaneous activity produced in the periphery by unknown mechanisms contributes to the barrel map formation in the somatosensory cortex 55. In addition, there is an ongoing debate regarding the roles of spontaneous vs. evoked activity. Nevertheless, the central hypothesis in the field has been that the peripherally-derived neuronal activity instructs higher sensory circuits.
In the current study, however, we found that the spontaneous activity generated within the mitral/tufted cell network establishes the discrete wiring of mitral cell dendrites. Dendro-dendritic glutamatergic synaptic transmission and/or glutamate spillover among mitral/tufted cells may be a major source of spontaneous activity that is required for dendrite pruning. Interestingly, the developmental changes of spontaneous activity from a synchronized to de-synchronized pattern (Fig. 4, 5) is also known for retinal waves 65. In the visual system, there are three types of retinal waves with different roles 1. It would be interesting to test in the future whether the different types of spontaneous activity in the olfactory bulb contributes to different aspects of circuit development in the olfactory and higher circuits 66, 67.
How do mitral cells choose just one glomerulus?
During the dendrite remodeling of mitral cells, one of the dendrites is strengthened to form strong synaptic connections, whereas the supernumerary ones are pruned. In particular, the pruning of primary dendrites is mediated by neuronal activity. We also found that this process is mediated by NMDARs. NMDARs are known to mediate the coincidence detection of pre- and post-synaptic activity. This mechanism is hypothesized to mediate the Hebbian-based remodeling of neurites, where “cells that fire together wire together” and “out of synch, lose your link” 68. In the visual system, it has been suggested that the eye-specific patterns of spontaneous activity facilitates the eye-specific segregation of visual circuits 69. Therefore, it is tempting to speculate that the developmental de-synchronization of the spontaneous activity may facilitate the dendrite pruning of mitral cells. It is an important ongoing challenge in the field to establish whether and how specific patterns of activity instruct the wiring specificity in sensory circuits.
Competition and synapse elimination is a wide-spread phenomenon in the developing brain. For example, it has been known that axons from different neurons compete for one target in the neuromuscular junction and cerebellar climbing fiber-Purkinje synapses. However, the mechanisms for the competition and selective synapse elimination remain elusive 70, 71. In mitral cells, competition occurs within a neuron, selecting one “winner” dendrite and eliminating all the other “losers”. Neuronal activity mediates the elimination of the “losers”. It will be important to determine which molecules mediate the intracellular competition.
In the visual, auditory, and somatosensory circuits, spontaneous activity generated in the periphery propagates all the way to the higher circuits to form coordinated connections as well as organized columnar structures. In contrast, the olfactory bulb sculpts parallel segregated olfactory connections based on its own intrinsic network activity, rather than the OSN-derived activity. An important lesson from our finding is that the autonomous formation of highly organized circuitry is another important principle for the activity-dependent development in sensory systems.
AUTHOR CONTRIBUTIONS
S.F. performed in utero electroporation, morphological analysis, hand rearing, and Ca2+ imaging of olfactory bulb slices. M.N.L. performed hand rearing and Ca2+ imaging of olfactory bulb in vivo. R.S. performed in utero electroporation at E10. Y.M. and T.S. established in utero electroporation of mitral cells. R. K. and K.K. generated floxed Cnga2. T.I. supervised the project. S.F., M.N.L. and T.I. wrote the manuscript. S.F. and M.N.L. contributed equally to this work.
METHODS
Mice
All animal experiments were approved by the Institutional Animal Care and Use Committee (IACUC) of the RIKEN Kobe Branch and Kyushu University. OMP-tTA BAC transgenic mice (OMP-tTA Tg; CDB0506T) and R26-TRE-GCaMP3 BAC transgenic mice (TRE-GCaMP3 Tg; CDB0505T) were generated in a previous study 49. Pcdh21-Cre Tg (RBRC02189) 72, R26-CAG-LoxP-TeNT knock-in (RBRC05154) 73, floxed-Cnga2 74, floxed-Grin1 (JAX #005246) 75, OMP-GFP knock-in (JAX #006667) 76, OMP-tTA knock-in (JAX #17754) 52, TRE-TeNT Tg (JAX # 010713) 77, OMP-Cre knock-in (JAX #006668) 78, a Cre-dependent GCaMP3 knock-in line Ai38 (JAX #014538) 54, Thy1-GCaMP6f Tg (line GP5.11) (JAX #024339) 53, Nkcc1 KO (MMRRC #034262) 79 have been described previously. ICR mice were used for in utero electroporation in Figure 1 and 2, and C57BL6N was used for Figure S1D. Grin1fl/fl, OMP-Cre;Cnga2fl/o, OMP-Cre;R26-CAG-LoxP-TeNT, Pcdh21-Cre;R26-CAG-LoxP-TeNT, Pcdh21-Cre;Ai38 in C57BL6 (C57BL6/N or C57BL/6J) or C57BL6 × 129 (129P2 or 129S2) mixed background were bred to ICR mice for in utero electroporation (mixed at various ratios). Nkcc1 KO backcrossed to FVB/N was bred to ICR mice for in utero electroporation (mixed at various ratios). Since the Cnga2 gene is X-linked, male hemizygous mutant mice (OMP-Cre;Cnga2fl/o) were analyzed for Cnga2. For OSN-TeNT, we initially analyzed OMP-tTA knock-in mice 52 crossed with TRE-TeNT Tg 77; however, TeNT was not expressed in all OSNs and a substantial amount of VAMP2 immunoreactivity remained in the OSN axon terminals (data not shown). We therefore used OMP-Cre;R26-CAG-LoxP-TeNT (both are knock-in), in this case VAMP2 immunoreactivity completely disappeared in OSN axons (Figure 3D). When mixed genetic backgrounds were used for quantification, samples from littermates were used for control experiments. Mice used for Ca2+ imaging (OMP-tTA;TRE-GCaMP3, Pcdh21-Cre;Ai38, Thy1-GCaMP6f, Thy1-GCaMP6f;OMP-Cre;R26-CAG-LoxP-TeNT, Thy1-GCaMP6f;Pcdh21-Cre;R26-CAG-LoxP-TeNT) were in C57BL/6 or C57BL6 × 129 mixed background. Both males and females were used for our experiments.
Plasmid constructions
Cre-dependent pCA-LNL-GFP and Flp-dependent pCA-FNF-DsRed are kind gifts from C. Cepko (addgene #13770 and #13771)80. NaChBac is a kind gift from D. Clapham 46. Kir2.1 (NM_008425.4) was PCR amplified from P10 mouse OB cDNA. Mutant Kir2.1 (mutKir2.1)45 was generated using PCR-mediated mutagenesis. TdTomato (a gift from R. Tsien) was PCR-amplified and subcloned into pCA-LNL or pCA-FNF vector. Cre (addgene #14797, a gift from C. Cepko), iCre (a gift from R. Sprengel), and FLPo (addgene #13793, a gift from P. Soriano) were PCR-amplified and subcloned into pCAG vector. FLAG-tag was fused at the C-termini of Kir2.1, mutKir2.1, and NaChBac and subcloned into pCA-LNL vector. Dlg4 coding for PSD95 was PCR amplified from P10 mouse OB cDNA, and pCA-FNF-PSDΔ1.2-GFP was generated as described previously 42, 43. We used PSDΔ1.2-GFP because the overexpression of full length PSD95 perturbed dendrite morphology in mitral cells (data not shown). Plasmids newly generated in this study will be deposited to Addgene (#125573-125581).
In utero electroporation
For sparse fluorescent labeling, pCA-LNL-tdTomato (1 μg/μL) and pCAG-Cre (2-10 ng/μL) were used. To label mutant neurons sparsely, pCA-FNF-tdTomato (1 μg/μL), pCAG-Flpo (2-10 ng/μL) were used. To generate single-cell Grin1 KO mutants, pCA-FNF-tdTomato (1 μg/μL), pCAG-Flpo (2-10 ng/μL), and pCAG-iCre (1 μg/μL) were used in homozygous floxed Grin1 mice. In utero electroporation was performed as described previously 37. To label mitral cells at E12, 2 μL of plasmid solutions were injected into the lateral ventricle and electric pulses (a single 10-ms poration pulse at 72 V, followed by five 50-ms driving pulses at 32 V with 950-ms intervals) were delivered along the anterior-posterior axis of the brain with forceps-type electrodes (3-mm diameter, LF650P3, BEX) and a CUY21EX electroporator (BEX). In utero electroporation for E10 embryos was performed as reported previously 81.
Naris occlusion
P0 mice were anesthetized on ice and a unilateral naris occlusion was quickly performed by cauterizing with soldering iron. Naris closure was confirmed under a stereomicroscope.
Hand rearing
Hand rearing of newborn mice was performed as described previously (https://www.youtube.com/watch?v=sNX2byHbppM) with some modifications. Mice were kept in a boxed chamber (AsOne, #AS-600SC) maintained at 28-30°C and a humidity of 50%. A heater (Natsume, KN-475-3-40) and a water bath (Sansyo, SWS-181D) were used to maintain the temperature and humidity, respectively. Up to 3 pups were hand reared at the same time. Hand rearing was performed from P1 to P6. Pups were fed with Powdered Dog Milk at the recommended dilutions (Esbilac, PetAG, warmed to 37°C, ∼50μL/g) with a paint brush every ∼2 hours (done in shifts day and night). Their head and body were kept clean with a damp cotton bud. To facilitate digestion, excretion, and to prevent bloating, their abdomen was massaged with a cotton bud just after feeding. If the pups appeared dehydrated, a subcutaneous injection of saline was administered. A detailed step-by-step protocol will be published in Bio-protocols.
Mouse brain samples
Mice were deeply anesthetized with an overdose i.p. injection of pentobarbital (Dainippon Sumitomo Pharma). Anesthetized mice were perfused with 4% paraformaldehyde (PFA) in phosphate buffered saline (PBS). Excised brain samples were fixed with 4% PFA in PBS at 4°C overnight. Samples were then embedded in 4% low-melting agarose (Thermo Fisher, #16520-100) to prepare brain slices. For immunohistochemical analysis, samples were cryoprotected with 30% sucrose at 4°C overnight and embedded in OCT compound (Sakura).
Immunohistochemistry
Frozen OBs were cut into 16-μm-thick coronal sections using a cryostat (Leica). Sections were blocked with 5% normal donkey serum in PBS for 30 minutes and then incubated with mouse anti-VAMP2 (1:500, Synaptic Systems, 104211) in 5% normal donkey serum at 4°C overnight. After washing 3 times in PBS, sections were incubated with Alexa 555-conjugated donkey anti-mouse IgG (1:500, ThermoFisher, A-31570) for 1 hour at room temperature. Nuclei were stained with DAPI. Images were taken by a confocal microscope (Olympus, FV1000).
Optical clearing and imaging
Mouse brain slices (2 mm-thick for SeeDB, 500 μm-thick for SeeDB2) were cut using a microslicer (Dosaka EM). Slices were then cleared with SeeDB 39, 82 or SeeDB2 40 as described previously. Step-by-step protocol and technical tips have been described in SeeDB Resources (https://sites.google.com/site/seedbresources/). SeeDB2 samples were stained with DAPI. Cleared tissues were mounted on glass slides with 2-mm or 500 μm silicone rubber spacer 82. Samples were imaged with an upright two-photon microscope (Olympus, FV1000MPE) with a motorized stage (Sigma-Koki). A commercialized 25x objective lens (Olympus, XLPLN25XSVMP, NA 1.0, WD 4.0 mm) were used to image SeeDB samples. InSight DeepSee Dual (Spectra-Physics) was used for two-photon excitation of tdTomato (1040 nm). SeeDB2 samples were imaged with inverted confocal microscopes: Olympus FV1000 with 405 nm, 488 nm, and 543 nm laser lines, or Leica SP8 with 405 nm and 552 nm laser lines.
In vivo Ca2+ imaging of the olfactory bulb
Mouse lines
Transgenic OMP-tTA;TRE-GCaMP3(OSN-GCaMP3), Pcdh21-Cre;Ai38(M/T-GCaMP3), or Thy1-GCaMP6f (GP5.11) (M/T-GCaMP6f) were used for in vivo Ca2+ imaging. For the imaging of mutant mouse strains they were crossed with the Thy1-GCaMP6f line (M/T-GCaMP6f). It should be noted that robust odor-evoked responses (ΔF/F > 100%) were detected in the adult olfactory bulb using these mouse lines 49. Odor responses were also seen in the neonates (data not shown).
Anesthesia
Neonatal mice were anesthetized with age and weight specific doses and concentrations of ketamine/xylazine that were administrated by intra-peritoneal injections. 10mg/ml ketamine and 20mg/ml xylazine and H2O were mixed and injected as below.
Ratios (ketamine/xylazine/H2O in mL) and doses (/g body weight):
P0: 10:0.5:21, 0.025 mL/g; P2: 10:0.5:21, 0.02 mL/g; P4: 10:0.5:6, 0.015mL/g; P6-16: 10:0.5:6, 0.01 mL/g; P20-30: 10:0.5:6, 0.02 mL/g.
Surgery
Once anesthetized, a small craniotomy was made above the OB, and filled with Kwik-Sil (WPI). Then a 2mm circular coverslip (Matsunami) was placed above and sealed with superglue. Dental cement (Shofu) was then used to attach a custom-made bar (Narishige), allowing us to head fix with a custom head holding device (Narishige, MAG-2) and image once the mice had recovered.
Ca2+ imaging of olfactory bulb slices in vitro
Pcdh21-Cre;Ai38(M/T-GCaMP3), or Thy1-GCaMP6f (GP5.11) (M/T-GCaMP6f) were used for in vitro Ca2+ imaging of olfactory bulb slices. Mouse pups were anesthetized on ice and sacrificed by decapitation. The olfactory bulb was quickly excised into ACSF (125 mM NaCl, 3 mM KCl, 1.25 mM NaH2PO4, 2 mM CaCl2, 1 mM MgCl2, 25 mM NaHCO3, and 25 mM glucose). OB slices were placed in a custom-made chamber. The chamber was placed under an upright two-photon microscope (Olympus, FV1000MPE) with a motorized stage (Sigma-Koki) and a water-immersion 25x objective lens (Olympus, XLPLN25XWMP, NA = 1.05, WD = 2.0). Before starting an imaging session, olfactory bulb slices are perfused with oxygenated ACSF (bubbled with 95% O2 and 5% CO2) at least for 1 hour at room temperature (25°C). During imaging sessions, olfactory bulb slices were continuously perfused with oxygenated ACSF (0.85-1.07 mL/min) at room temperature.
Pharmacology
50 μM APV (Sigma, A5282), 25 μM CNQX (Sigma, C239), 300 μM Carbenoxolone (Sigma, C4790), 1mM 1-octanol (Sigma, 95446), 100 μM GABA (Sigma, A5835), 1 μM TTX (Abcam, ab120055), were used for pharmacological experiments for olfactory bulb slices.
Quantification of dendrites
Only the medial side of the olfactory bulb was analyzed. We only analyzed neurons located in the mitral cell layer. Therefore, external tufted cells and granule cells, sometimes labelled by in utero electroporation, were excluded from our analysis. We only quantified neurons where the entire dendritic field was included in the 3D image blocks. In our SeeDB protocol, glomeruli could be identified by autofluorescence signals. All primary dendrites were manually traced with Neurolucida software (MBF Bioscience). Feeding and quantification of dendrites were blinded in the hand rearing experiments. In all the remaining experiments, quantification was not blinded; however, to exclude any biases, all labelled neurons that meet the above criteria were quantified in an unbiased manner. Littermates were used as controls. Criteria for quantification are described in Figure 2B.
Analysis of in vivo Ca2+ imaging data
Motion and Drift Correction
Motion correction and drift correction were performed by using TurboReg (translation only, http://bigwww.epfl.ch/thevenaz/turboreg) then SPM (www.fil.ion.ucl.ac.uk/spm). Noise removal and subsequent ΔF/F movies were performed using AFID (a MATLAB plugin created by Dr. A. Lowe, https://github.com/mleiwe/AFID) 83.
Spike Detection
Glomeruli were manually selected and ΔF/F values were calculated for each glomerulus. To detect glomerular spikes, a normal distribution was fitted to the lowest 80% of data points. Glomerular spikes were defined as points that were 3 standard deviations above the mean value calculated from a 10 second window around the time point in question.
Correlation Index
In order to measure the correlation of glomeruli a new measurement was created that measured the number of glomeruli active per frame (Gn) as a percentage of the mean number of active glomeruli per frame (). This normalization was necessary in order to control for differences in the number of glomeruli imaged as well as differences in the firing frequency of the network.
H- and L-events
H/L events were determined by calculating the number of glomeruli that were active in each frame. In Figures 4e,f,g, 7 and Supplementary Figure 4a, 7b, if the correlation index was above 5%, and contained two or more glomeruli it was classed as an H-event.
Otherwise, if there were events present it was classed as an L-event.
Spike Timing Tiling Coefficient (STTC) analysis
Calculations were performed as described previously 59. Correlations were counted within a 2 second window (i.e. ΔT=2 seconds).
Custom written MATLAB codes were used for all in vivo and in vitro analyses and are available at GitHub (https://github.com/mleiwe/SpontActivityOfDevMitralCells).
Analysis of slice Ca2+ imaging data
Ca2+ imaging data were processed by using ImageJ (NIH) and MATLAB (Mathworks). Glomeruli were manually selected as polygonal region of interests (ROIs) on ImageJ. ROI information is converted to region mask using poly2mask function. Average fluorescence intensity within each ROI is defined as F. F0 is defined as the average of 10-30 percentile values of fluorescence intensity in each glomerulus. ΔF/F values were calculated for each glomerulus. Motion correction and drift correction were not performed for in vitro imaging data analysis. Baselines were corrected by polynomial curve fitting. Raster plots were plotted by detecting peak points located in more than 4 standard deviations above the baselines. For calculation of standard deviations, top 25 percentile data were excluded. Correlation Index, H- and L-events were determined as described above. Cross correlation was calculated from ΔF/F. The value was normalized so that the autocorrelations at zero lag equal 1. Cmax and Tlatency were defined as shown in Supplementary Figure 5b. Cross correlation matrix shows Cmax values of all combinations of two glomeruli. Average frequency and amplitude were calculated from the number of peak points and ΔF/F value at each peak point, respectively.
GCaMP images and movies
GCaMP fluorescence images shown in Figure 4, 5, 7, and Supplementary Figure 3, 7 are maximum intensity projection images throughout the imaging session (also background subtracted) showing glomeruli. To prepare serial GCaMP images and ΔF/F images, spatial median filter (diameter, 3 pixels) was applied to reduce the shot noise. For ΔF/F images, F0 values were created from the lowest 80% of pixel values. ΔF/F images are shown in pseudo-color (inferno). Unprocessed raw image data will be deposited to SSBD database (http://ssbd.qbic.riken.jp/set/2019xxxx/).
Statistical analysis
Excel, Prism7, and MATLAB were used for statistical analysis. Sample sizes were not pre-determined. However, post-hoc power analyses determined that sufficient samples were collected. All the statistical data including the sample size and the statistical test used in each figure panel are described in Supplementary Table 1. The number of neurons is described within figures, and all the numerical data including the numbers of animals are in Supplementary Table 2. Number of animals used for Ca2+ imaging was at least three and indicated within figures. χ2-tests were used in Figure 2b,c, 3a,b,e,j, 6e,f, 7d, and Supplementary Figure 2c,d,f, 6c. One-way ANOVA with Tukey post hoc tests were used in Figure 4. Wilcoxon signed-rank test was used in Figure 6a-c, and Supplementary Figure 6a. Kruskal-Wallis test and post hoc Dunn’s multiple comparison test was used in Figure 1d. Welch’s t-test was used in Figure 3c, 3f, 5d, and 7a. Welch’s t-tests, or Mann Whitney U tests were performed for imaging data in Figures 7c and Supplementary Figure 7b. In box plots, the middle bands indicate the median; boxes indicate the first and third quartiles; and the whiskers indicate the minimum and maximum values. Data inclusion/exclusion criteria are described in figure legends.
Data and code availability
Raw image data used in this study will be deposited to Systems Science of Biological Dynamics (SSBD) database (http://ssbd.qbic.riken.jp/set/2019xxxx/). Neurolucida tracing data will be deposited to NeuroMorpho.Org (http://neuromorpho.org/). Numerical data for all graphs are included in Supplementary Table 2. Program codes have been deposited to GitHub (https://github.co.jp/) as detailed above. Requests for additional program codes and data generated and/or analyzed during the current study should be directed to and will be fulfilled on reasonable request by the Lead Contact, Takeshi Imai (t-imai{at}med.kyushu-u.ac.jp).
ACKNOWLEDGMENTS
We thank M. Yokoi (Pcdh21-Cre), I. Imayoshi (R26-CAG-LoxP-TeNT), P. Mombaerts (OMP-GFP, OMP-Cre), S. Tonegawa (floxed Grin1, TRE-TeNT), H. Zeng (Ai38), K. Svoboda (Thy1-GCaMP6f), C. Ron Yu (OMP-tTA knock-in), G.E. Shull (NKCC1 KO) for mouse strains, D. Clapham (NaChBac), C. Cepko (pCA-LNL-tdTomato), P. Soriano (FLPo) for plasmids, A.S. Lowe for image processing software, I.D. Thompson for the neonatal anesthesia protocol, and T. Miyata for instruction of in utero electroporation at E10. We are also grateful to R. Iwata and M-T. Ke for valuable advices on imaging and analysis, and M. Nomura, Y. Sakashita, Y. Taniyama, and E. Yamashita for technical assistance. This work was supported by grants from the PRESTO program of the Japan Science and Technology Agency (JST) (T.I.), the JSPS KAKENHI (23680038, 15H05572, 15K14336, 16K14568, 16H06456, and 17H06261 to T.I., 15K14327 and 17K14944 to S.F., 17K14946 to M.N.L.), Sumitomo Foundation (T.I.), Nakajima Foundation (T.I.), and RIKEN CDB intramural grant (T.I.). Imaging experiments were supported by the RIKEN Kobe Light Microscopy Facility and animal experiments were supported by the Laboratory for Animal Resources and Genetic Engineering at the RIKEN Center for Life Science Technologies.
REFERENCES
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.
- 7.
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.↵
- 21.↵
- 22.↵
- 23.
- 24.
- 25.
- 26.↵
- 27.↵
- 28.
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.↵
- 46.↵
- 47.↵
- 48.↵
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵