CaMKII controls neuromodulation via neuropeptide gene expression and axonal targeting of neuropeptide vesicles

Ca2+/calmodulin-dependent kinase II (CaMKII) regulates synaptic plasticity in multiple ways, supposedly including the secretion of neuromodulators like brain-derived neurotrophic factor (BDNF). Here, we show that neuromodulator secretion is indeed reduced in mouse α- and βCaMKII-deficient (αβCaMKII double-knockout [DKO]) hippocampal neurons. However, this was not due to reduced secretion efficiency or neuromodulator vesicle transport but to 40% reduced neuromodulator levels at synapses and 50% reduced delivery of new neuromodulator vesicles to axons. αβCaMKII depletion drastically reduced neuromodulator expression. Blocking BDNF secretion or BDNF scavenging in wild-type neurons produced a similar reduction. Reduced neuromodulator expression in αβCaMKII DKO neurons was restored by active βCaMKII but not inactive βCaMKII or αCaMKII, and by CaMKII downstream effectors that promote cAMP-response element binding protein (CREB) phosphorylation. These data indicate that CaMKII regulates neuromodulation in a feedback loop coupling neuromodulator secretion to βCaMKII- and CREB-dependent neuromodulator expression and axonal targeting, but CaMKIIs are dispensable for the secretion process itself.


Introduction
Neuropeptides and neuromodulators are secreted via Ca 2+ -dependent fusion of dense-core vesicles (DCVs) found in axons and dendrites of most neurons in the brain. Upon binding to their receptors, the secreted molecules modulate synaptic function [1,2] and synaptic plasticity. One of the well-documented examples is how brain-derived neurotrophic factor (BDNF) facilitates long-term potentiation (LTP) and learning [3,4]. However, the mechanism by which molecules like BDNF are secreted from DCVs is still a poorly defined element. DCV exocytosis is strictly Ca 2+ -dependent and requires higher frequency stimulation than synaptic

CaMKII facilitates neuropeptide secretion
To understand the role of CaMKII in neuropeptide secretion, we used αCaMKII and βCaMKII conditional DKO mice [35]. Hippocampal neuron cultures were infected with Cre-recombinase (DKO) or an inactive form of Cre [36] (WT) on the day of plating (day in vitro, DIV0) (Fig 1A), producing complete knockout of both α and βCaMKII at DIV17 (Fig 1B). To investigate DCV fusion, we performed live-cell imaging of the DCV cargo BDNF-pHluorin [37] ( Fig  1C), which allowed us to probe neuropeptide exocytosis at single-vesicle resolution ( Fig 1D  and S1 Movie). Neurons were stimulated twice with 8 bursts of 1 second at 50 Hz to trigger DCV fusion [6], and individual BDNF-pHluorin fusion events ( Fig 1E) were counted. The number of BDNF-pHluorin fusion events was lower in DKO neurons compared with WT (WT = 515 ± 281.11; DKO = 282 ± 153.15 p = 0.0575; Fig 1G). This reduction indicates that neuromodulators, such as BDNF, require CaMKII for efficient secretion, as suggested in Drosophila NMJ [7] and by pharmacological inhibition in mouse [10]. The majority of DCVs fuse at synapses [38,39]. In Drosophila, CaMKII inhibition reduces the amount of DCV fusion at the NMJ synapse [7]. We investigated the localization of BDNF-pHluorin fusion events ( Fig 1H) and observed no differences in the percentage of BDNF-pHluorin fusion events that occurred at Synapsin-mCherry marked synapses (WT = 68.18% ± 4.42; DKO = 71.81% ± 3.54 p = 0.106; Fig 1I). Thus, CaMKII facilitates the secretion of neuromodulators both at synaptic as well as at extrasynaptic locations.
To assess the specificity for neuropeptide transmission, we used Synaptophysin-pHluorin (SypHy) [40]  As suggested in previous studies in αCaMKII KO [32] as well as α-and βCaMKII DKO [22,35], increase in SypHy signal during stimulation (S1 Fig), as well as its decay during the recovery period (S1 Fig), was comparable between DKO and WT neurons. We, therefore, concluded that CaMKII contributes to neuropeptide exocytosis specifically over synaptic transmission.

CaMKII affects the number of DCVs, not their fusion
To investigate the mechanism of reduced BDNF secretion, we measured the fraction of DCVs that fused upon stimulation [6]. Strikingly, the total number of pHluorin-labeled DCVs, quantified upon brief NH 4 Cl superfusion (Fig 2A), was reduced by 43% in DKO neurons compared with WT (WT = 7.72 × 10 3 ± 1.19 × 10 3 ; DKO = 4.44 × 10 3 ± 386; p = 0.0124; Fig 2B). The same effect is also observed as a leftward shift in the histogram of total DCVs/per cell (S1 Fig). The reduction in pHluorin-labeled DCVs was not due to a defect in acidification of the vesicles, because the baseline fluorescence was comparable in the 2 genotypes (WT = 1 ± 0.017; DKO = 0.961 ± 0.019; p = 0.1437 S2 Fig). Taking into account the lower number of pHluorinlabeled DCVs, the fraction of DCVs fusing in DKO neurons that was similar to WT (WT = 6.83% ± 2.4; DKO = 5.6% ± 2.2; p = 0.3575; Fig 2C and 2D), indicating that CaMKII does not control the DCV fusion process itself but regulates the number of available DCVs.
The reduced signal upon NH 4 Cl superfusion could be due to reduced loading of BDNF-pHluorin into vesicles. The increase in signal intensity during fusion was calculated to estimate the loading of BDNF-pHluorin into vesicles (S1  Taken together, our data demonstrate that CaMKII is a positive regulator of neuromodulation by promoting the number of available DCVs. Moreover, CaMKII is not involved in the fusion of SVs.

CaMKII DKO neurons have reduced number of DCVs per synapse
Because the total number of BDNF-pHluorin-labeled DCVs was significantly reduced ( Fig  2B), we tested whether CaMKII regulates the expression of the reporter by regulating the promoter activity of the synapsin promoter in our reporter construct. Like many other studies, we synapses. Traces show mean ± SEM (shaded area); boxplots with 95% CI whiskers, the white cross represents mean ± SEM, and the central bar is median. Columns and dots represent individual litters and neurons, respectively. The presented data can be found in S1 Data; original Western blot can be found in S1 Raw images. � p = 0.057. Scale bar = 25 μm (D); 5 μm (D insert); 10 μm (I). AP, action potential; BDNF, brain-derived neurotrophic factor; CaMKII, Ca2+/ calmodulin-dependent kinase II; CFP, cyan fluorescence protein; CI, confidence interval; DCV, dense-core  vesicle; DIV, day in vitro; DKO, double-knockout; WT, wild type. https://doi.org/10.1371/journal.pbio.3000826.g001 The fraction is calculated as the number of DCVs fusing per frame divided by the total number of BDNF-pHluorin puncta per neuron. (D) Fraction of BDNF-pHluorin-labeled DCV fusing during stimulation. Traces show mean ± SEM (shaded area); boxplots with 95% CI whiskers, the white cross represents mean ± SEM, and the central bar is median. Columns and dots represent individual litters and neurons, respectively. The presented data can be found in S1 Data. � p < 0.05. Scale bar = 25 μm (A). AP, action potential; BDNF, brain-derived neurotrophic factor; CaMKII, Ca2+/calmodulin-dependent kinase II; CI, confidence interval; DCV, dense-core vesicle; DIV, day in vitro; DKO, double-knockout; WT, wild type.
https://doi.org/10.1371/journal.pbio.3000826.g002 designed these reporters to be specific for neurons and do not express in the supporting glia in the cultures, using the widely used (human) synapsin promoter. However, the expression levels of all constructs driven by this synapsin promoter were similar (S2 Fig).
To further investigate the lower neuromodulator levels in CaMKII DKO neurons, the number of DCVs was assessed along the neuritic arborization and stainings for endogenous neuromodulators were quantified as a function of the distance from the soma. The number of DCVs along the neuritic arborization was lower in CaMKII-deficient neurites (S2 Fig), and the staining intensity was reduced in CaMKII DKO neurons at every point in the dendritic arborization (S3 Fig). These data support our previous conclusion that CaMKII-deficient neurons contain fewer DCVs and, consequently, less neuromodulators in their neuritis. This suggests that the reduction in neuromodulator in CaMKII DKO neurons is explained by defects upstream of neurite targeting.
These results indicate that CaMKII is required to maintain normal neuropeptide levels in synapses, as well as some members of the regulated secretory machinery, hence indirectly affecting neuropeptide secretion. Boxplots with 95% CI whiskers, white cross shows mean ± SEM, central bar is median. Columns and dots represent individual litters and neurons, respectively. The presented data can be found in S1 Data. � p < 0.05, �� p < 0.01, ��� p < 0.001. Scale bar = 5 μm. CaMKII, Ca2+/calmodulin-dependent kinase II; CI, confidence interval; DCV, dense-core vesicle; DKO, double-

Fewer DCVs enter the axon of CaMKII DKO neurons, whereas DCV trafficking is unaffected
To evaluate whether the reduced neuropeptide levels at synapses is a consequence of a decreased loading of neuropeptides into DCVs or a decreased number of DCVs entering the axon, hippocampal neurons expressing BDNF-mCherry and the axon initial segment (AIS) marker voltage-gated sodium channel intracellular domain fused to blue fluorescent protein (BFP-NaV II-III ) [44] (Fig 4A) were photobleached at the location of the AIS to visualize DCV trafficking into the axon (Fig 4B and S2 Movie). The number of BDNF-labeled vesicles trafficking into the AIS in CaMKII DKO neurons was reduced by 42% (WT = 4.115 ± 0.257 DCV per minute; DKO 2.395 ± 0.171; p = 1.52 × 10 −7 ; Fig 4C), in line with the reduction in intensity levels of endogenous BDNF and CHGB at synapses. To test whether the decreased number of trafficking DCVs was specific for the axon, 2 dendritic regions proximal to the soma were photobleached ( Fig 4E). The targeting of DCVs into dendrites was similar in WT and DKO neurons ( Fig 4F). The reduction in axonal targeting but normal dendritic targeting together leads to decreased axonal preference of DCV targeting (WT = 68.9 ± 0.05%; DKO = 60.6 ± 0.06; p = 0.0033; Fig 4G). In addition, the number of BDNF-labeled DCVs traveling in a retrograde direction was increased by 35% (WT = 1. Extracellular, released BDNF is known to be transported to the soma when associated with tyrosine receptor B (TrkB) and subsequently internalized [45]. We observed that CaMKII In Drosophila NMJ, CaMKII regulates DCV mobilization and capture at synapses [7,8]. Mammalian DCVs stall during high-frequency stimulation [46], a process that might involve a CaMKII-dependent modification of the cargo-kinesins (KIFs) complex, like for the interaction of Mint1 and KIF17 [23]. To evaluate this, we used high-density hippocampal cultures with 5% of neurons expressing tetracycling inducible (TetON) membrane-bound green fluorescence protein (mGFP) and NPY-mCherry. This allowed for discrimination between axon and dendrites to study DCV trafficking in both compartments (Fig 4H and S3 Movie). DCV speed was measured during high-frequency stimulation of 16 repetitions of 1-second field stimulation at 50 Hz and compared with speed during baseline (30 seconds before stimulation). The speed of moving DCVs was reduced during stimulation ( Fig 4I) for approximately 40% of vesicles in both anterograde and retrograde direction in both genotypes (S5 Fig). The fraction of vesicles that stalled (speed reduced to zero) was 16%-18%, whereas 24%-29% increased their speed during stimulation, similar in both genotypes ( Fig 4J). Therefore, CaMKII promotes the knockout; MUNC18, mammalian uncoordinated 18; NPY, neuropeptide Y; n. s., not significant; SNAP25, synaptosomal-associated protein 25; SV, synaptic vesicle; Syn1, synapsin 1; VAMP2, vesicle-associated membrane protein 2; VGLUT1, vesicular glutamate transporter 1; WT.
In cultured hippocampal neurons, DCV fusion occurs predominantly in the axon [6]. To test whether this preference is altered in CaMKII DKO neurons, the location of DCV fusion ( Fig 4K) was quantified in the same sparse infected neurons ( Fig 4L). The total release fraction of NPY-mCherry-labeled DCVs was the same between the 2 groups ( Fig 4M). However, the percentage of fusion that occurred in axons of αβCaMKII-deficient neurons was reduced by 22% (WT = 85.09 ± 5.39%; DKO = 66.67 ± 13.25%; p = 0.0239; Fig 4N).
Together, these results show that CaMKII regulates the number of DCVs entering the axon and axonal preference of DCV fusion but not DCV trafficking properties.

CaMKII regulates neuromodulator expression
The reduction of axonal preference for DCVs in CaMKII DKO neurons might be due to misstargeting into dendrites or a general reduction in neuropeptide expression, thus reducing DCV number. Of the 4 BDNF transcripts [30,47], transcript II and IV are regulated by low neuronal activity in a synergetic activity between CaMKII and PKC, via pCREB [28]. CREB binding sites are also present in the promoters of CHGB [48] and secretogranin2 (SCG2) [49]. Therefore, we reasoned that in our culture system CaMKII regulates transcription of neuropeptide mRNAs (Fig 5A and 5B).
https://doi.org/10.1371/journal.pbio.3000826.g004 76% ( Fig 5E). CaMKII holoenzymes are multimers of α and β subunits and the action of the 2 subunits is interdependent, e.g., βCaMKII KO neurons have substantial impairments in αCaM-KII signaling, resembling αCaMKII KO [21]. Therefore, we overexpressed the individual subunits in the DKO background to test if either α-or βCaMKII has a dominant function in regulating neuropeptide levels in hippocampal neurons. αCaMKII did not restore neuropeptide levels, whereas βCaMKII restored expression levels to 70% and 95%, respectively, for CHGB and SCG2. Thus, βCaMKII regulates the transcription of neuromodulators, and the absence of the β gene is responsible for the reduction in the amount of neuromodulators produced.

Secreted BDNF and CaMKII are required for BDNF-induced neuromodulator expression
BDNF is known to enhance transcription of neuropeptides and neuromodulators [50][51][52], including the precursor of BDNF itself [53][54][55]. In addition, extracellular, secreted BDNF is known to activate CaMKII via its receptor TrkB [56]. We therefore hypothesized that BDNFdependent activation of CaMKII is required for a normal neuromodulator expression. To test this, BDNF was scavenged from the culture medium by adding 0.2 μg/ml of BDNF antibody (BDNF Ab) [57] at DIV 15 and DIV 16 and analyzed BDNF-induced BDNF expression. As a CI whiskers, white cross shows mean ± SEM, and central bar is the median. Columns and dots represent individual litters and neurons, respectively. The presented data can be found in S1 Data; original Western blot can be found in S1 Raw images. ATG, transcription starting codon; BDNF, brain-derived neurotrophic factor; CaMKII, Ca2+/calmodulin-dependent kinase II; CHGB, chromogranin B; CI, confidence interval; DCV, dense-core vesicle; DKO, double-knockout; LTP, long-term potentiation; SCG2, secretogranin2; WT, wild type.

βCaMKII activity regulates neuropeptide levels
Because βCaMKII is known to be the dominant paralog to enhance pCREB [27] and βCaMKII, but not αCaMKII, rescued the total amount of CHGB and SCG2 (Fig 5D and 5E), we hypothesized that βCaMKII is necessary and sufficient for the regulation of neuropeptide expression and normal synaptic neuromodulator levels.
To test this hypothesis, we expressed α-and βCaMKII mutants that cannot be phosphorylated at their inhibitory sites, shown to cause a reduction in the threshold for LTP (long-active [LA], TT305/306VA for αCaMKII, and TT306/307VA for βCaMKII) [65,66], or a kinasedead ([KD] K42R for αCaMKII and K43R for βCaMKII) mutation [67] (Fig 8A) in CaMKII DKO neurons and immunostained them for the DCVs marker CHGB (Fig 8B) and endogenous BDNF (Fig 8C).  Fig 8C). These levels were similar to WT cells. For CHGB, βCaMKII-WT rescued the immunoreactivity (Fig 8B), as suggested by the Western blot data (Fig 5D). The KD variant of βCaMKII did not rescue the reduced BDNF or CHGB immunoreactivity. These data confirm that βCaMKII has an important enzymatic function to regulate synaptic levels of neuropeptides and neurotrophic factors.
The dendritic length was also normalized to WT levels with overexpression of the constitutively active form of βCaMKII (S7 Fig), confirming its function in regulating dendritic arborization [41,42], while synapse density was increased in all genotypes, except for the αCaMKII-LA (S7 Fig), indicating that αCaMKII activity is required to limit synaptogenesis in primary hippocampal neurons.
Together, these data indicate that the kinase activity of βCaMKII is critical to regulating neuropeptide levels at synapses. Thus, βCaMKII serves as a central link for BDNF-induced neuromodulator production controlling the signaling cascade toward pCREB, whereas it is dispensable for the trafficking of neuromodulator containing DCVs and their fusion at synapses (Fig 8D).

Discussion
This study confirms that CaMKII regulates the secretion of neuromodulatory signals. However, the current data reveal that in mouse hippocampal neurons, the release machinery that drives this process operates at a typical efficiency in the absence of α-and βCaMKII. Instead, we discovered that fewer DCVs that contain these neuromodulators were targeted into the axon, leading to fewer DCVs per synapse and less neuromodulator secretion during/after stimulation. This reduced axonal targeting was specific for DCVs; targeting of SV markers and exocytic proteins was only marginally reduced. The reduction of axonal targeting was explained mainly by a reduction in neuromodulator expression. We show that CaMKII is a crucial component in a positive feedback loop activated by secreted BDNF that promotes neuromodulator expression via βCaMKII, CaMKIV and pCREB. Finally, the kinase activity of βCaMKII, but not αCaMKII, was essential for normal neuromodulator expression and targeting to the axon.
CaMKII was already implicated in neuromodulator secretion based on pharmacological perturbations in mouse [10] and Drosophila [7,8] and using genetic perturbation in C. elegans [9]. However, these studies reached different conclusions, based on different assays: pharmacological inhibition reduced the amount of DCV marker fluorescence lost from dendrites [10] or NMJ buttons [7] during stimulation, which was interpreted as reduced neuromodulator secretion. In contrast, mutant nematodes showed reduced DCV marker fluorescence in axons but a typical accumulation in coelomocytes, which was interpreted as abnormal DCV targeting but normal secretion [9]. Our study provides the first data that directly assess DCV fusion, at single vesicles' resolution, demonstrating that CaMKII is not required for the fusion process itself and thus agrees with the nematode study. Hence, pharmacological and genetic perturbations appear to result in different conclusions. The use of chemical inhibitors provides the most acute perturbations, but KN-93 is known to have side effects on the activity of other kinases and ion channel function [68].
One of the well-characterized actions of CaMKII is to couple excitation to transcription of neuronal plasticity genes, such as BDNF, via pCREB [25][26][27]69]. Pharmacological inhibition of CaMKII represses the chromatin state of BDNF promoter [70] and blocks long-term depression (LTD)-mediated BDNF transcription [28]. Likewise, CHGB [48] and SCG2 [49] genes contain a CREB binding site, and their expression in PC12 cells is activity dependent [71]. In this study, we observed that the mRNA levels of CHGB, SCG2, and BDNF transcript II were also reduced, resulting in a reduced protein level of these DCV cargos, whereas a less activity-sensitive transcript of BDNF (transcript IV) [28,30] was not reduced. Our data indicate that active βCaMKII expression restores CHGB and SCG2 protein levels (Fig 5D and 5E). βCaMKII selectively regulates excitation-translation coupling via CaMKIV-dependent pCREB [27] and, unlike αCaMKII, is expressed in both glutamatergic and GABAergic neurons [72]. Our study demonstrates that activating the signal cascade downstream of βCaMKII restores neuromodulator expression in a pCREB-dependent manner (Figs 7 and 8). Therefore, we conclude that βCaMKII activity, but not αCaMKII activity, induces expression of neuromodulators and other DVC cargo by activating a downstream pathway involving nuclear translocation of CaM, CaMKIV activation, and pCREB ( Fig 8J). endogenous BDNF; bottom, quantification of synaptic BDNF intensity levels at VGLUT1 positive synapses for DKO and DKO rescued with the indicated constructs normalized to WT. (D) Schematic representation of the mechanism for CaMKII-and BDNF-dependent gene expression. BDNF binds to TrkB receptors; BDNF-TrkB binding leads to activation of βCaMKII that triggers a downstream pathway leading to Ca/CaM to enter into the nucleus and activating CaMKIV, which in turn phosphorylate CREB at S133. Neuropeptides and neuromodulator are then loaded into DCVs and targeted to dendrites and synapses, where they will fuse with the plasma membrane releasing their content. This mechanism will create a positive feedback loop to regulate neuromodulators content. Boxplots with 95% CI whiskers, white cross shows mean ± SEM. Columns and dots represents individual litters and neurons, respectively. The presented data can be found in S1 Data. � p < 0.05, �� p < 0.01, ��� p < 0.001. Scale bar = 10 μm (B and D). Ca/CaM, Ca2+/Calmodulin; CaMKII, Ca2 +/calmodulin-dependent kinase II; CaMKIV, Ca2+/calmodulin-dependent kinase IV; CHGB, chromogranin B; CI, confidence interval; CREB, cAMP-response element binding protein; DCV, dense-core vesicle; DKO, double-knockout; KD, kinase-dead; LA, long-active; NLS-Nrgn, nuclear-localized neurogranin; SV, synaptic vesicle; Syn1, synapsin1; TrKB, tyrosine receptor kinase B; VAMP2, vesicleassociated membrane protein 2; VGLUT1, vesicular glutamate transporter 1; WT, wild type. https://doi.org/10.1371/journal.pbio.3000826.g008 Secreted neuromodulators like BDNF are known to activate CaMKII [56]. Our study suggests that this BDNF-induced CaMKII activity, especially βCaMKII activity, is part of a positive feedback loop to initiate a signal cascade that induces neuromodulator expression, e.g., to compensate for neuromodulators lost because of evoked secretion. Indeed, preventing BDNF secretion or scavenging of secreted BDNF in WT neurons produced a phenocopy of the situation in CaMKII DKO neurons (Fig 6). In addition, expression of components of the signal cascade downstream of βCaMKII activation restored the levels of BDNF in CaMKII DKO synapses (Fig 7). Thus, in addition with the well-established role of forming the holoenzyme together with αCaMKII and its targeting to the PSD [21,22,35,69,72], βCaMKII has a unique role not shared by αCaMKII. In mouse hippocampal neurons, βCaMKII acts in a positive feedback loop downstream of BDNF binding to Trk receptors and upstream of CaMKIV-dependent pCREB, which supports normal neuromodulator secretion (Fig 8).
Episomal expression of DCV cargo did not restore a normal amount of DCVs and BDNF at synapses. Hence, the availability of excess template is not sufficient to restore normal DCV/ BDNF amounts. Although one such DCV cargo, e.g., BDNF/NPY, is overexpressed, expression of other DCV proteins that are also regulated by CaMKII, like CHGB, are most likely still reduced. This might prevent a normal DCV biogenesis rate, and the excess single overexpressed cargo might be subject to degradation.
Increasing pCREB restored normal levels of BDNF in DKO neurons but induced no further increase in WT neurons (Figs 6 and 7). This seems in contrast with previous observations that at least mRNA levels increase in WT neurons after BDNF application [53][54][55]. First, as in many other cases, mRNA levels may simply not predict protein levels, because of extensive posttranslational regulation. For instance, episomal expression of DCV cargo did not restore a normal amount of DCVs and BDNF at synapses, and such episomally encoded cargo might be subject to degradation (see previous) Second, the regulation of BDNF mRNA levels is complex and may differ under different experimental conditions. For instance, in immature cortical neuronal cultures, BDNF mRNA is differentially regulated via a CaMKII-dependent and a mitogen-activated protein kinase (MAPK)-dependent pathway [30,54,55]. The MAPK cascade, however, is not necessary to maintain normal BDNF mRNA transcription in unstimulated neurons [55].
Although the secretion of neuromodulatory signals is affected in the absence of α-and βCaMKII, secretion of SVs at single synapses was normal (S1A-S1F Fig) during high-frequency stimulation. These data are in line with previous studies in single null mutant neurons: at CA3 synapses, αCaMKII, or βCaMKII deficiency does not alter basal neurotransmission but abolishes LTP [21,31]. In contrast, overexpression of α-or βCaMKII was shown to have opposite effects on spontaneous fusion of SVs in brain slices. The frequency of miniature excitatory postsynaptic current (mini-EPSC) was reduced in αCaMKII overexpression, whereas βCaM-KII increased the frequency [72]. Such effects, mainly the increased frequency, may be explained by CaMKII's role in neuromodulation, as increased BDNF secretion was reported to increase mini-EPSC frequency [2]. Furthermore, we observed that dendritic and axonal length and synapse number were all increased in the absence of α-and βCaMKII (S3 Fig). Such changes in synapse number are also expected to affect overall spontaneous SV fusion frequency.
Both CaMKII isoforms are required for LTP in the hippocampus [21,65,73], and neuromodulators, especially BDNF, are too [74]. The 40% reduction in DCV fusion events per cell (Fig 1G) may help to link these 2 well-documented phenomena. However, the role of BDNF release in LTP is also demonstrated by acute pharmacological inhibition of CaMKII with KN-93, which strongly decreased the dendritic release of BDNF [10]. This acute effect is unlikely to be explained by transcriptional regulation, as we report here for reduced BDNF release in our system. Hence, LTP probably involves additional KN-93 or CaMKII-dependent processes.
In Drosophila NMJ synapses, acute inhibition of CaMKII with KN-93 does not alter basal DCV fusion, but it limits the mobilization of stationary vesicles [33] and the capture at synapses of DCVs transiting along the axon [34]. This joint effect on mobilization and capture was proposed as a mechanism to support efficient DCV supply and fusion at NMJs during intense stimulation [7,33]. Our data indicate that in mammalian neurons, the activity of CaMKII is dispensable for normal fusion efficiency. These mobilization and capture mechanisms may not be crucial in mammalian neurons, and they might be different between NMJ and CNS synapses or redundant mechanisms may have evolved to secure efficient fusion in the absence of α-and βCaMKII.
We observed that altered neuromodulation in the absence of CaMKII is explained by a robust and rather specific reduction of neuropeptides in synapses (Fig 3). A similar phenotype was observed in C. elegans motor neurons, where DCV cargo was also reduced in axons [9]. We observed that cargo loading in individual vesicles was unaltered (S1G-S1I Fig), indicating that the reduction of neuropeptide levels at synapses must be attributed to fewer DCVs and not too less cargo per vesicle, in line with observations in C. elegans motoneurons [9]. Interestingly, endogenous and exogenous DCV cargo was subject to the same reduction, suggesting that neuropeptide overexpression does not rescue the number of DCVs produced.
We observed that 50% fewer DCVs enter the axon of DKO neurons, whereas the same number entered into dendrites (Fig 4A-4G). This observation cannot be explained by an altered neuromodulator expression and probably involves CaMKII targets other than CREB. The targeting of vesicles to specific compartments involves post-translational modification of both cytoskeletal [75] and motor proteins [76]. For instance, a mutation in the CaMKII substrate cyclin-dependent kinase 5 [77] leads to the loss of neuropeptides targeting to the axon with a compensatory increase in dendritic targeting [78]. However, we did not observe such compensation, indicating that the altered targeting observed in the current study differs from what was previously reported. Therefore, CaMKII regulates neuromodulation by promoting DCV targeting to axons, with a mechanism that appears to be independent from dendritic targeting.
We also observed an increased number of retrograde moving vesicles in the axon (S3 Fig). This phenomenon could be a consequence of decreased synaptic capture of DCVs, which in Drosophila NMJ depends on CaMKII [8]. In NMJ boutons, electrical activity causes a rapid reduction of GFP-labeled neuropeptide signal [34]; this signal slowly recovers (25% recovery in 5 minutes) by capturing DCVs transiting along the axon in a process mediated by CaMKII activity [8]. However, we did not observe differences in the trafficking properties of DCVs in DKO axons, neither in baseline transport nor after high-frequency stimulation, where most DCVs reduce speed and arrest [46,79]. In dendrites, CaMKII is required to disrupt the interaction between KIF17 and Mint1, allowing cargo unloading from the motor complex [23], and calmodulin was recently reported to regulate DCV-kinesin loading and unloading in dendritic spines [80]. These considerations strengthen the idea that in addition to regulating neuromodulator expression, CaMKII also regulates DCV targeting to axons.
In conclusion, the current study provides direct evidence that CaMKII is dispensable for efficient neuromodulator exocytosis in mouse hippocampal neurons and that βCaMKII regulates neuromodulation by promoting CaMKIV-dependent neuropeptide expression downstream of BDNF signaling. Hence, CaMKII is crucial in a feedback loop coupling neuromodulator secretion to neuromodulators expression and subsequent DCV targeting into the axon.

Ethical statement
All animals were housed and bred according to institutional and Dutch Animal Ethical Committee regulations (DEC-FGA 11-03).

Laboratory animals and primary neuron cultures
The generation of floxed Camk2a and Camk2b mice was previously described (Kool 2019). Postnatal day 1 (P1) pups were humanely killed, and hippocampi were dissected in Hank's balanced salt solution (HBSS Sigma) with 10 mM HEPES (Life Technology), digested in 0.025% trypsin (Life Technologies) for 20 minutes at 37˚C and dissociated with fire-polished Pasteur pipettes. Dissociated neurons were resuspended in neurobasal supplemented with 2% B-27, 18 mM HEPES, 0.25% Glutamax, 0.1% penicillin/streptomycin (Life Technologies) and plated at a density of 1,300 neurons/well on astrocyte micro-island [6,81,82] in 12-well plates; for highdensity cultures 25,000 neurons/well were plated on pregrown glia cells. Astrocyte microisland were generated by plating 6,000 rat glial cells per agarose-coated 18-mm glass coverslip, stamped with 0.1 mg/ml poly-D-lysine (Sigma) and 0.7 mg/ml rat tail collagen (BD Biosciences). For Western blots and qRT-PCR, neurons were plated at a density of 250,000 neurons/ well on plates coated with a solution of 0.0005% Poly-L-ornithine, 2.5 μg/ml Laminin (Sigma). Neuronal cultures were kept in supplemented neurobasal at 37˚C and 5% CO 2 .
All plasmids were sequenced-verified and subcloned into lentiviral vectors to produce viral particles [84].
Neuronal cultures were infected with Cre-recombinase 5 hours after plating to generate CaMK2-DKO or an inactive Cre [36] as control. Rescue plasmids were added to the same mix. For experiments in Fig 1 and Fig 2, viruses encoding the BDNF-pHluorin and Synapsin-mRFP were added to the neuronal culture 6 days before the readout. For experiments in Fig  4A-4G, viruses encoding BDNF-mCherry and Nav II-III -YFP were added to the neuronal culture 3 days prior to live-cell imaging; experiments in Fig 4H-4N are described next for sparse labeling. For experiments in Fig 7A-7C, 0.2 μg/ml of BDNF antibody (DSHB) or 100 ng/ml or recombinant BDNF (Bio Connect) were added twice to the culture every 24 hours from DIV15, and cells were fixed 48 hours after the first administration. For experiments in Fig 7D-7F, viruses encoding for TeNT-IRES2-mCherry were added 6 days before fixation. For experiments in Fig 8, viruses encoding HA-CaMKIV, HA-NLS-Nrgn, and HA-CREB-Y134F were added to the culture 48 hours before fixation. Proper concentration of viruses was tested in WT cultures.
For sparse labeling with lentivirus particles, neurons were incubated in supplemented neurobasal with adequate amount of viral particles for 2 hours at 37˚C, 5% CO 2 , washed in Dulbecco's modified Eagle medium (DMEM Life Technologies) containing 10% fetal calf serum (FCS), resuspended in supplemented neurobasal and plated 5,000 neurons/well on top of previously plated high-density culture, resulting in a 30,000 neuron/well cultures with 15% infected neurons/well. Expression was started 6 days before experiment by adding 2 g/ml doxycycline hyclate (Sigma).
For SypHy experiments, the imaging protocol included 30 seconds of baseline recording, electrical field stimulation using a A-385 stimulus isolator (WPI) controlled by a Master 8 (AMPI), delivering 1-millisecond, 30-mA pulses for 5 seconds at 40 Hz, followed by 1 minute of recovery time and a final 5 seconds' perfusion with modified Tyrode's containing NH 4 Cl (2 mM CaCl 2 , 2.5 mM KCl, 119 mM NaCl, 2 mM MgCl 2 , 30 mM glucose, 25 mM HEPES, 50 mM NH 4 Cl [pH 7.4]) delivered by gravity flow through a capillary placed above the cell.
For DCV fusion, neurons were imaged for 30 seconds as baseline, stimulated with electrical field stimulation for 8 pulses of 1 second at 50 Hz separated by 0.5 seconds, allowed to rest for 30 seconds after which the stimulation was repeated. After an additional 30 seconds of recovery Tyrode's containing NH 4 Cl were superfused for 5 seconds.
For trafficking experiments, labeled neurons were identified, and the field of view was adjusted to fit the majority of the axon. After 30 seconds of baseline, electrical field stimulation was applied with 16 bursts of 1 second at 50 Hz separated by 0.5 seconds. For fluorescence recovery after photobleaching (FRAP) experiments, laser intensity and pulse duration for bleaching were optimized to reach >90% fluorescence decrease of mCherry [82], images were acquired for 180 seconds after an initial 5 seconds of baseline.

Image analysis
For SV fusion analysis, individual synapses were identified and treated as single region of interest (ROI) with a custom-made ImageJ algorithm (National Institute of Health); briefly, synapses were identify based on their increase in signal during NH 4 application, if upon NH 4 an ROI had a ΔF/F 0 < 4 � StD(F 0 ) + F 0 , where StD represents the standard deviation of the signal, the ROI was discarded. Individual traces were analyzed with a custom-made MATLAB (Mathworks, www.mathworks.com) script where synapses were quantified as active if the maximum ΔF/F 0 upon stimulation was � 3 � StD(F 0 ); active synapses were pooled per cell. SypHy fusion fraction was calculated as the ΔF stimulation / ΔF NH4 .
For DCV fusion, 2×2 pixel ROIs were placed semi-automatically, using a custom-made script in ImageJ, on all pHluorin region that appeared during the electrical stimulation. Individual traces were validated using a custom-made MATLAB script; only regions that showed an increase in ΔF/F 0 > 3 � StD(F 0 ) + F 0 and with a rise time <1 seconds were considered as positive fusion events. The total number of DCVs was calculated based on the NH 4 Cl response of individual recording with a custom-made ImageJ algorithm. Because of the overlap of DCVs in individual puncta, the number of vesicles was corrected normalizing the puncta intensity by the mode of the first percentile of the intensity distribution per cell.
For neuronal morphology and synaptic quantification, maximum intensity projections of confocal images were analyzed with a custom-made ImageJ algorithm, dendrites and axons based on Ridge detection, and their length was calculated based on the skeleton analysis in ImageJ; synapses were identified based on their intensity and dimension. For signal intensity, the intensity of each individual neuron was normalized to the average intensity of the WT condition in that biological replica.
For Western blot analysis, the band at the proper molecular weight was considered as an ROI. Intensity density was used to normalize the signal using the density of actin as reference. The adjusted signal was normalized to the signal of the WT condition.
For visualization purposes, brightness and contrast of representative examples were adjusted in a linear scale using the WT as reference. Saturation was always <1% of the pixels and 0% of the pixels were set to undersaturation during adjustment.

Statistics
Anderson-Darling test was used to test normality distribution of the data, and Levene's test for homogeneity of variances. When the assumptions for parametric tests were met, t test and 1-way ANOVA (post hoc Dunn-Sidak or Fisher) were used to test significant differences in the mean of the population. In case data were nonparametric, Mann-Whitney U and Kruskal-Wallis tests were used to test the significance of the median of the populations. All statistics were done in MATLAB, and their values are reported in S1 Data.