The calcium/calmodulin-dependent protein kinase 2 (CAMK2) family consists of four different isozymes, CAMK2A, CAMK2B, CAMK2G, and CAMK2D, transcribed from individual genes but with high homology as they all consist of the same domains. Of these domains, the linker domain between the regulatory and association domain shows substantial variation between members (Sloutsky and Stratton, 2021). All four isozymes are expressed in the brain, though at different expression levels (Tobimatsu and Fujisawa, 1989; Bayer et al., 1999; Murray et al., 2003). The role of CAMK2A and CAMK2B in brain functioning has been studied extensively using different mouse models, revealing their critical role in learning and synaptic plasticity (Silva et al., 1992a,b; Giese et al., 1998; Elgersma et al., 2002; van Woerden et al., 2009; Borgesius et al., 2011; Achterberg et al., 2014). This is accentuated by the discovery of mutations in the corresponding genes of patients suffering from neurodevelopmental disorders with intellectual disability (Küry et al., 2017; Akita et al., 2018; Chia et al., 2018; Rizzi et al., 2020; Heiman et al., 2021; Proietti Onori and van Woerden, 2021). Studying the role of CAMK2G in the brain has only recently become a focus of interest, amongst others due to the identification of patients with neurodevelopmental disorders carrying a mutation in the CAMK2G gene (de Ligt et al., 2012; Proietti Onori et al., 2018), highlighting the function of CAMK2G in normal brain functioning.

CAMK2G expression can be detected by embryonic day 11.5 (E11.5) in mice, and it is expressed throughout the brain in excitatory neurons, interneurons, and astrocytes (Tobimatsu and Fujisawa, 1989; Sakagami and Kondo, 1993; Bayer et al., 1999; Vallano et al., 2000; Batiuk et al., 2020; Huntley et al., 2020; Kozareva et al., 2021). The highest expression is observed in interneurons, and both global and interneuron-specific loss of CAMK2G results in impaired cognition and plasticity (Cohen et al., 2016, 2018; He et al., 2021, 2022). Different splice variants are expressed by the Camk2g gene, one of which contains a nuclear localization signal (NLS) in its variable domain. This specific isoform is shown to shuttle calcium-bound calmodulin to the nucleus in excitatory neurons, potentially playing a role in excitation–transcription coupling (Ma et al., 2014; Cohen et al., 2016). However, in interneurons, CAMK2G is not involved in calmodulin shuttling to the nucleus (Cohen et al., 2016). Additionally, knockdown of Camk2g was found to cause abnormal neurite growth in primary neurons, which could be normalized by re-expressing both a CAMK2G isoform with as well as without the NLS (Proietti Onori et al., 2018). Together, these results show that more research is required to get full insight into the role of CAMK2G in neurons and more broadly, in the brain.

Similar to mice, in humans CAMK2G is one of the main CAMK2 isozymes expressed during early neurodevelopment (Proietti Onori et al., 2018). The finding that a mutation in this gene causes a neurodevelopmental disorder (Proietti Onori et al., 2018) suggests that CAMK2G plays an important role in neurodevelopment. Patients carrying mutations in CAMK2G show, besides intellectual disability, a spectrum of phenotypic traits (Proietti Onori et al., 2018). To provide further insight into the role of CAMK2G in normal brain function, we obtained global Camk2g knockout mice and performed an extensive phenotypic characterization. We tested cognitive behavior but also other forms of behavior as patients with CAMK2G missense mutations suffer from severe intellectual disability and autism spectrum disorder but also impaired motor skills, hypotonia during infancy, and atypical body growth development (Proietti Onori et al., 2018). Our results indicate that in contrast to CAMK2A and CAMK2B, CAMK2G is not essential for hippocampus-dependent learning or synaptic plasticity, which contradicts previous Camk2g knockout mice studies (Cohen et al., 2018; He et al., 2021). Instead, we found that CAMK2G plays a role in motor and innate behavior, further supporting the role of CAMK2G in neurodevelopment.

Camk2g knockout mice (EUCOMM; Camk2g^–/–) were bred in a C57BL/6J background. For all experiments, wild-type littermates (Camk2g^+/+) were used as control. Camk2g^–/– were viable and appeared healthy, but showed reduced body weight compared with Camk2g^+/+ (Supplementary Figure 1). To validate the Camk2g^–/– mice and assess where CAMK2G is expressed, we isolated 8 brain regions from Camk2g^–/– and Camk2g^+/+ mice and performed Western blots on forebrain, striatum, hypothalamus, diencephalon, cortex, hippocampus, cerebellum, and brainstem. Probing for CAMK2G revealed three bands, of which the upper two were specific and were absent in the Camk2g^–/– samples (Figure 1A), suggesting different isoforms of CAMK2G are expressed in the brain. In general, CAMK2G is expressed quite consistently throughout the brain, with a trend for the highest expression levels in the cortex (Figures 1A, B; for statistics see the legends). As expected, CAMK2G is absent in all brain regions of the Camk2g^–/– samples, while CAMK2A and CAMK2B expressions were unaltered compared to Camk2g^+/+ (Figures 1A, D). The lowest aspecific band of CAMK2G overlapped with CAMK2B (for raw blots see Supplementary Figure 2). Interestingly, the two CAMK2G isoforms showed a region-defined expression: in regions from the cerebrum, both isoforms were nearly equally expressed, but in the cerebellum and brainstem, expression of the larger isoform was markedly decreased (Figures 1A, C).

To confirm the presence of different CAMK2G isoforms in different brain areas, an online RNA-seq database (GEO accession number: GSE141252; Stoeger et al., 2019 bioRxiv) on the forebrain and cerebellum tissue of adult C57BL/6J mice was used to quantify exon-junctions of the Camk2g transcripts. Confirming literature (Takeuchi et al., 2000; Tombes et al., 2003; Kamata et al., 2006; Sloutsky and Stratton, 2021), we found splice variants solely in the linker domain of Camk2g, splicing out exon 13, 15, 16, and/or 19 (for numbering of exons see Supplementary Figure 3). Figure 1E illustrates exons in the linker domain with the amount of transcript reads between exons (for raw data plots, see Supplementary Figure 4). Intriguingly, exon 13 showed the largest difference between brain regions and is spliced out more frequently in the cerebellum than in the forebrain (Figure 1F), which would translate to a 1.1 kDa decrease in size on protein level, corresponding to our Western blot results (Figure 1A). Therefore, the observed different ratio in cerebrum versus cerebellum protein expression is likely due to region-specific CAMK2G isoform distribution.

Having established that our mouse model is indeed lacking CAMK2G expression, we set out to study the role of CAMK2G in learning and plasticity. For spatial memory, we made use of the Morris water maze. Although the Camk2g^–/– mice show slightly longer latencies to find the platform in the first 2 days compared with Camk2g^+/+ mice, all mice showed a reduction in their latency to find the platform over days, with no differences between genotypes over days (Figure 2A). Reduction in latency does not necessarily mean that the mouse shows spatial learning as mice can use different non-hippocampus-dependent strategies to find the platform. Hence, to assess hippocampus-dependent spatial memory, the platform was removed for a probe trial on day 5 and the time spent in the different quadrants of the water maze was analyzed. Both Camk2g^+/+ and Camk2g^–/– spent significantly more time in the target quadrant (TQ) compared with the other quadrants (Figures 2B, C), indicating intact, or at best very moderately impaired, spatial learning in our Camk2g^–/– mouse model. Additionally, no differences were found in the number of platform crosses (Figure 2D) or the swim speed (Figure 2E).

We next tested associative learning using contextual and cued fear conditioning. Mice were placed in a context, where after 150 s a tone was presented for 20 s, immediately followed by a foot shock (Figure 2F). No significant differences in freezing behavior were observed between the Camk2g^+/+ and Camk2g^–/– mice during baseline measurement (Figure 2G). Surprisingly, Camk2g^–/– showed significantly more freezing than Camk2g^+/+ in the context condition (Figure 2H). Also in the cued condition, Camk2g^–/– showed significantly more freezing than Camk2g^+/+; however, baseline freezing was already significantly different in the Camk2g^–/– mice compared with their wild-type littermates. To assess the learning effect, pre-tone freezing levels were subtracted from tone freezing levels, which still revealed significantly higher freezing levels in Camk2g^–/– than Camk2g^+/+ (Figure 2I). Combining the tone and shock strongly activates the amygdala, potentially masking the hippocampus-dependent contribution of this conditioning task. To explicitly test the hippocampal contribution to the phenotype, in a separate cohort we only tested contextual conditioning, omitting the cue. Again, no significant difference was observed in baseline freezing levels (Figure 2J). When placed back in the same context 24 h later, similar freezing levels were observed between Camk2g^+/+ and Camk2g^–/– (Figure 2K). Thus, our results suggest that CAMK2G is not necessary for hippocampus-dependent learning and memory, but potentially plays a role in amygdala-dependent associative fear conditioning.

Having assessed hippocampal learning, we set out to investigate hippocampal synaptic plasticity in Camk2g^–/–, which could potentially still reveal a mild phenotype. We performed extracellular field recordings in the CA1 to CA3 Shaffer collateral pathway and detected no differences in baseline synaptic transmission since the presynaptic fiber volley and the field excitatory postsynaptic potential (fEPSP) increased equally between Camk2g^+/+ and Camk2g^–/– mice with increasing stimulus strength (Figure 3A). We detected a significant decrease in the paired-pulse facilitation in Camk2g^–/– compared with Camk2g^+/+, but this was only apparent at the 10- and 50-ms intervals (Figure 3B). Induction of long-term potentiation (LTP) using a 100-Hz induction protocol resulted in LTP in both Camk2g^+/+ and Camk2g^–/–, with no significant difference neither after 60 min nor after 120 min recording between the two groups (Figure 3C and Supplementary Figure 5). As the classical 100-Hz stimulation is very strong, it is possible that a plasticity phenotype is masked, hence, milder LTP induction protocols were tested. No differences in LTP were found between Camk2g^+/+ and Camk2g^–/– in the 10-theta condition; however, Camk2g^–/– showed significantly impaired LTP compared to Camk2g^+/+ in the 3-theta induction condition (Figures 3D, E). Surprisingly, this difference was lost in the 2-theta condition (Figure 3F). These results suggest that if there is a role for CAMK2G in synaptic plasticity at the CA3-CA1 synapse, it is very mild.

To further study the effect of the absence of CAMK2G on behavior and brain function, we expanded our tests to include other forms of behavior such as motor and innate behavior. We tested locomotor activity in the open field and found a trend toward decreased activity in Camk2g^–/– compared with Camk2g^+/+, though not significant (Figure 4A). We further challenged their motor performance as well as their motor learning using the accelerating rotarod. Although both the Camk2g^–/– and Camk2g^+/+ mice showed motor learning over days, the Camk2g^–/– performed significantly poorer at all time points measured (Figure 4B). Also, on the balance beam, sensitive to balance and fine motor coordination, Camk2g^–/– consistently showed more foot slips crossing beams of varying widths (Figure 4C). Since CAMK2G patients show developmental hypotonia (Proietti Onori et al., 2018), we tested if reduced muscle strength could play a role in the reduced motor performance seen in Camk2g^–/– mice, but this was not the case as grip strength was equal between the groups (Figure 4D). Finally, we assessed some innate behaviors of the Camk2g^–/– using the nest building, marble burying, and forced swim test. Compared with Camk2g^+/+, Camk2g^–/– used significantly less material to build a nest over days (Figure 5A) and showed impaired burying behavior (Figure 5B). No significant differences were observed between Camk2g^–/– and Camk2g^+/+ mice in floating behavior in the forced swim test (Figure 5C).

With the identification of patients suffering from CAMK2G-associated neurodevelopmental disorder (de Ligt et al., 2012; Proietti Onori et al., 2018), the interest to study the role of CAMK2G in normal brain functioning significantly increased. Here, we set out to expand the knowledge on the role of CAMK2G in neuronal functioning through the characterization of a novel Camk2g^–/– mouse model. We provide evidence for differential expression of different CAMK2G isoforms depending on the brain area, where we speculate that this is due to differential splicing of exon 13. At the behavioral level, our results suggest that CAMK2G has a minor role in hippocampus-dependent spatial learning and memory, as well as synaptic plasticity, but does affect amygdala-dependent fear learning. Finally, we find clear phenotypes in fine motor performance and intrinsic behaviors, as measured using the nest-building and marble-burying assays.

The expression levels of CAMK2G throughout the brain were assessed by isolating different brain regions of Camk2g^+/+ and Camk2g^–/– mice. Isolation of brain regions was validated by CAMK2A expression, which was highest in the hippocampus and the cortex and lowest in the cerebellum and the brainstem, and CAMK2B expression, also highest in the hippocampus and dominant in the cerebellum yet nearly absent in the brainstem (Cook et al., 2018). Total levels of CAMK2G did not differ much between the different brain areas. Interestingly, in brain areas where CAMK2B and CAMK2A showed little expression, such as the hypothalamus, CAMK2G appeared to be one of the more dominating CAMK2 isozymes. However, we have to be careful in drawing such conclusion as we did not use a pan-CAMK2 antibody to directly compare expression levels between CAMK2s. Even though the CAMK2G antibody also recognized CAMK2B (see raw blots in Supplementary Figure 2), which runs at the height of the lowest band, it is interesting to see that in the hypothalamus and the brainstem, the lowest band can still be seen in the WT but no longer in the knockout samples. As CAMK2B is hardly detectable in these brain areas and this lowest band disappears in Camk2g^–/– samples, these findings could suggest the expression of a third smaller isoform of CAMK2G. A more specific antibody that does not cross-react with CAMK2B would be necessary to substantiate this conclusion. For the upper two CAMK2G bands, we find evidence for differential expression of different CAMK2G isoforms. Based on RNA-seq data, the different isoforms correspond with differential splicing of exon 13. Our data suggest that the isoform containing this exon is expressed in the cerebrum but hardly in the cerebellum. Interestingly, the most common transcript found in the human hippocampus does not contain this exon as the number of transcripts with exon 13 detected was <1% (Sloutsky et al., 2020). Whether CAMK2G in the human brain shows region-specific isoforms or whether this differential expression of isoforms is species-specific remains to be studied. A homolog of exon 13 is expressed by CAMK2B, sharing 55% similarity with exon 13 in CAMK2G on the amino acid level, and encompasses the binding region for F-Actin (O’leary et al., 2006). Future research will have to show whether this exon in CAMK2G also mediates binding to F-Actin.

Our conclusion that CAMK2G has a minor role in hippocampal learning, memory, and cognition is in contrast with previous findings. The first study on a global Camk2g^–/– mouse model showed impaired long-term (24 h) memory, as well as reduced late-phase LTP in Camk2g^–/– (Cohen et al., 2018). Furthermore, they show that the deletion of CAMK2G solely in forebrain excitatory neurons is sufficient to cause the deficits seen in the global Camk2g^–/– mice (Cohen et al., 2018). Interestingly, the following study showed that CAMK2G is mainly expressed in interneurons and that the deletion of CAMK2G specifically in parvalbumin-positive interneurons is sufficient to induce deficits in synaptic plasticity in these interneurons as well as in behavioral long-term (24 h) memory (He et al., 2021). It is challenging to explain the differences in findings between the different labs. One of the possible explanations is the differences in mouse strains used. The mice we used for our experiments were kept in the C57BL/6J background, whereas the original mice used in the other studies were in a 29SvEv/C57BL6/CD1 mixed background (Backs et al., 2010; Cohen et al., 2018; He et al., 2021). Indeed, it has been shown for other mouse models that mouse strain can make a significant difference in behavioral phenotypes (Born et al., 2017; Sonzogni et al., 2018). However, to test whether this is indeed the case for CAMK2G, the different mouse strains would need to be assessed side by side.

It might not be surprising that CAMK2G only has a minor role in hippocampus-dependent learning and plasticity. CAMK2A and CAMK2B, the two most abundant CAMK2 isozymes expressed in the brain, have both been shown to be essential for hippocampal plasticity and learning (Silva et al., 1992a,b; Elgersma et al., 2002; Borgesius et al., 2011), and the deletion of Camk2a and Camk2b simultaneously results in the complete absence of NMDA-receptor dependent LTP (Kool et al., 2019). If CAMK2G were to play a role in these processes as well, it should (partially) compensate for the lack of CAMK2A and CAMK2B, which it does not. Especially considering that CAMK2G is one of the earliest CAMK2 isozymes to be expressed during development; it can be detected already as early as E11.5, whereas CAMK2B starts to be expressed at E14.5 and CAMK2A around P1 (Bayer et al., 1999).

On the other hand, it is possible that CAMK2G affects different intracellular pathways than CAMK2A and/or CAMK2B, having a unique role in plasticity. For example, CAMK2G plays a role in the excitation–transcription coupling in excitatory neurons, shuttling calmodulin to the nucleus upon excitation (Ma et al., 2014; Cohen et al., 2016). This specific excitation–transcription shuttling could explain why the previous study in Camk2g^–/– mice only found a difference in late-phase LTP and not in the early phase (Cohen et al., 2018). However, when testing late-LTP in our mouse model, we did not find a deficit (Supplementary Figure 5). The discrepancy in findings could be due to different induction protocols. We used a 1 × 100 Hz for 1 s induction protocol, whereas Cohen et al. (2018) used a stronger induction protocol: 3 × 100 Hz for 1 s with 5 min interstimulus interval. Whether these differences in induction protocols could result in differences in excitation–transcription coupling remains to be studied. The difference in mouse strains used is likely a better explanation for the discrepancy in findings between labs. Indeed, strain differences have been shown both for hippocampal learning and LTP (Nguyen et al., 2000). Finally, CAMK2G is not the only CAMK2 isozyme containing a nuclear localization signal (NLS), as this has also been shown for a CAMK2A isoform found in the midbrain/diencephalon regions (Brocke et al., 1995), and for CAMK2D (Srinivasan et al., 1994; Shioda et al., 2015). However, the exact role of the nuclear translocation of these isozymes remains to be elucidated.

There are several piles of evidence that suggest a role for CAMK2G in neurodevelopment: (1) the expression onset of CAMK2G is already very early in neurodevelopment, and at that developmental timepoint, CAMK2G and CAMK2D are the most abundant CAMK2 isozymes expressed in the brain (Proietti Onori et al., 2018); (2) CAMK2G is necessary for proper neurite formation during neuronal development in vitro (Proietti Onori et al., 2018; Xi et al., 2019), and (3) a pathogenic mutation in CAMK2G has been shown to cause a neurodevelopmental disorder in children (de Ligt et al., 2012; Proietti Onori et al., 2018). Our findings that CAMK2G plays an important role in motor and innate behavior further support the neurodevelopmental role for CAMK2G. Indeed, the behavioral tests assessed here, especially marble burying and rotarod, have been shown by us and others to have critical developmental periods, for example, in Ube3a and Shank3 knockout mice (Silva-Santos et al., 2015; Mei et al., 2016; Rotaru et al., 2020; Sonzogni et al., 2020). Not only does adult Ube3a or Shank3 gene reinstatement fail to rescue the marble burying or rotarod phenotypes but adult deletion of Ube3a also fails to induce phenotypes in these behavioral assays (Sonzogni et al., 2019). Another evidence for a neurodevelopmental origin for motor behavior comes from the inducible Camk2b knockout mouse. Deletion of Camk2b from the germline causes severe impairment in rotarod performance, whereas in mice where deletion of Camk2b is induced in adulthood, this phenotype is much milder (Kool et al., 2016). This is in stark contrast to cognitive behavior as, for example, tested in a conditional Camk2a mouse model. Deletion of Camk2a in adult mice is as detrimental for hippocampal learning and plasticity as germline deletion (Achterberg et al., 2014), and adult reinstatement rescues all cognitive behavior (Rigter et al., 2022). One important aspect here is the onset of expression. Whereas CAMK2G, CAMK2B, and UBE3A already have their expression onset prenatally, CAMK2A starts to be expressed postnatally; hence, early neurodevelopment is unaltered (Bayer et al., 1999; Judson et al., 2014).

Overall, our results reinforce previous literature suggesting that CAMK2G plays a critical role in normal brain functioning, albeit not in hippocampus-dependent memory, and further support the role of CAMK2G in neurodevelopment.

The original contributions presented in this study are included in the article/Supplementary material, further inquiries can be directed to the corresponding author.

The animal study was reviewed and approved in accordance with the European Commission Council Directive 2010/63/EU (CCD project license AVD101002017893), by an Independent Review Board (IRB) of the Erasmus MC.

PR performed the experiments, analyzed the data, and wrote the manuscript. CK performed the experiments and was responsible for the mouse breedings and genotyping. GW designed the study and wrote the manuscript. All authors contributed to the article and approved the submitted version.