The autism and schizophrenia-associated protein CYFIP1 regulates bilateral brain connectivity

Copy-number variants of the CYFIP1 gene in humans have been linked to Autism and Schizophrenia, two neuropsychiatric disorders characterized by defects in brain connectivity. CYFIP1 regulates molecular events underlying post-synaptic functions. Here, we show that CYFIP1 plays an important role in brain functional connectivity and callosal functions. In particular, we find that Cyfip1 heterozygous mice have reduced brain functional connectivity and defects in white matter architecture, typically relating to phenotypes found in patients with Autism, Schizophrenia and other neuropsychiatric disorders. In addition, Cyfip1 deficient mice present deficits in the callosal axons, namely reduced myelination, altered pre-synaptic function, and impaired bilateral-connectivity related behavior. Altogether, our results show that Cyfip1 haploinsufficiency compromises brain connectivity and function, which might explain its genetic association to neuropsychiatric disorders.

and regulated protein synthesis, are crucial for synaptic development and brain functioning, strengthening the hypothesis that alterations in the CYFIP1 gene can account for the clinical features of patients with ASD and SCZ.
Deletion of the Cyfip1 gene in mice leads to embryonic lethality 41,44,45 while heterozygous mice show aberrant behavior 44 , decreased dendritic complexity, increased number of immature spines 41,45 , as well as electrophysiological defects, such as enhanced mGluRdependent LTD 44 and vesicle release probability 46 , suggesting that Cyfip1 heterozygous mice are a suitable model for the neurological defects observed in BP1-BP2 haploinsufficient patients. Recently, it was shown in zebrafish that CYFIP1 regulates axonal growth of retinal ganglion cells (RGSs) 47 . The role of Cyfip1 in brain connectivity, however, remains not explored. Here, we show that Cyfip1 heterozygous mice (Cyfip1 +/-) have reduced functional connectivity and white-matter architecture defects. Cyfip1 haploinsufficiency leads to a decreased transmission of callosal synapses, reduced myelination in the corpus callosum, and affects motor coordination. Together, our findings suggest that CYFIP1 is likely the key gene that accounts for the functional connectivity and callosal defects observed in patients with the 15q11.2 deletion, and in other forms of neuropsychiatric disorders with reduced CYFIP1 levels and defects in functional connectivity.

Bilateral functional connectivity is impaired in Cyfip1 +/mice
Many neuropsychiatric disorders are characterized by impaired brain connectivity. To investigate whether and how Cyfip1 haploinsufficiency could affect brain networks in vivo, we analyzed functional connectivity (FC) in several brain regions using resting-state functional magnetic resonance imaging (rsfMRI). Cyfip1 +/adult mice (postnatal day 60, P60) showed decreased functional connectivity (FC) compared to wild-type (WT) (compare the lower left and upper right halves of the matrix, Figure 1a). Brain areas that present the most significant differences between WT and Cyfip1 +/mice are the cingulate cortex (Cg) and the thalamus (T) (Figure 1b). FC networks as measured by resting state fMRI are characterized by midline symmetry, and correlations between homotopic regions are particularly strong 48,49 . To analyze bilateral FC, seed-based analysis was performed for some of the areas (Figure 1c).
Remarkably, FC with the contralateral side of the seed region was particularly affected in the hippocampus, somatosensory and motor cortices (Figure 1d). In conclusion, our results show significant defects in functional connectivity, especially of the bilateral connections.

Aberrant white matter architecture in Cyfip1 +/mice
Bilateral connectivity mostly passes through the corpus callosum and abnormalities in this brain structure are a hallmark of neuropsychiatric disorders. To determine whether the observed functional deficits are due to structural axonal abnormalities in the corpus callosum, we performed diffusion tensor imaging (DTI) 50 in WT and Cyfip1 +/animals. DTI is one of the most widely used techniques to analyze white matter architecture and integrity. DTI yields several variables, of which the fractional anisotropy (FA) is the most widely used. High FA values are indicative of fiber-like structures and therefore, FA provides an estimate of axonal architecture 50 . In Figure 2a, the FA values across WT and Cyfip1 +/brains were color-coded, with higher values in red highlighting white matter tracts such as the CC (prominent in the first to fifth image). FA was generally reduced throughout the brain in the Cyfip1 +/mice (compare the upper and lower rows); except for the fimbria that showed slightly increased FA. The differential map of WT and Cyfip1 +/-FA values showed that the highest reduction in FA was detected in the CC, in particular in the genu, its anterior part. (Figure 2b; first image).
We therefore calculated the DTI parameters in the corpus callosum. No statistical differences were observed in the mean (MD), axial (AD), or radial diffusivity (RD), indicating that there are no gross defects in the white matter structure. In contrast, fractional anisotropy (FA) was significantly lower in the corpus callosum of Cyfip1 +/mice (Figure 2c), which might indicate changes in axonal thickness or myelination.
To determine which of the two properties is affected, we performed electron microscopy of the corpus callosum (CC). Representative micrographs from the anterior part of the CC were selected (Figure 3a), and myelinated axons were automatically identified and parameterized.
For each axon, the diameter and the myelin thickness were determined. In addition, the gratio, i.e., the ratio of the internal over the external axon diameter (Figure 3b) was calculated.
We observed no changes in the axonal diameter between the two genotypes, whereas the myelin thickness was reduced ( Figure 3c). Consequently, the g-ratio increased across all axon sizes (Figure 3c). In conclusion, Cyfip1 haploinsufficiency causes defects in functional connectivity, especially of the callosal projections that connect the two cortical hemispheres, which correlates with a reduced myelin thickness of the callosal fibers.

Cyfip1 haploinsufficiency leads to altered callosal presynaptic function
It is well established that neuronal activity can regulate callosal myelination 51 . To investigate whether Cyfip1 deficiency leads to defects in neuronal activity, we measured spontaneous activity in the somatosensory cortex of acute brain slices using microelectrode arrays (MEAs) ( Figure 4a). We found that both the spike and burst rate were significantly reduced in the

Cyfip1 +/mice show motor coordination defects
Callosal abnormalities have been associated with motor deficits, particularly with motor coordination, both in humans 6 and in mice 53,54 . We therefore used the accelerating rotarod test (Figure 6a) to investigate how the impaired bilateral connectivity and callosal transmission in Cyfip1 +/mice affect motor function. As shown in Figure 6b-c, Cyfip1 +/mice fall off significantly earlier than their WT littermates, indicating reduced motor coordination. Of note, this defect was ameliorated after repeated trials (Figure 6c), suggesting intact motor learning. To exclude that the abnormal coordination was due to reduced locomotion, general activity of Cyfip1 +/mice was measured in the open field. Cyfip1 +/animals moved normally (Figure 6d-f) indicating that locomotion was normal and that the defect on the rotarod is indeed due to impaired motor coordination. In sum, we gathered compelling electrophysiological, anatomical, functional and behavioral evidence, which all demonstrate defects in callosal connections of the adult Cyfip1 +/mouse brain.

DISCUSSION
Here we describe an impairment in callosal architecture and transmission in a mouse model of Cyfip1 haploinsufficiency. These callosal defects have an impact on functional connectivity and on motor coordination which might explain the genetic association of CYFIP1 with neuropsychiatric disorders.
We observed impaired bilateral functional connectivity, which correlates with several kinds of alterations in the collosal axons: in the microstructure as determined by DTI, which correlates with reduced myelin thickness ( Figure 2 and 3), and in pre-synaptic transmission ( Figure 5).
Regarding pre-synaptic function, we observe that trains of stimulation lead to short-term synaptic depression in WT but not Cyfip1 +/slices ( Figure 5). This could be caused by changes in neurotransmitter release probability 55 , which would be higher in WT slices, but other presynaptic alterations could also be the cause. Arguably, changes in synaptic transmission have an impact on spontaneous activity, as we observed (Figure 4), and neuronal activity modulates myelination of the callosal axons 51 . In this model, the primary defect would lie in the callosal axons. On the other hand, reduced myelination affects signal propagation along the axons and could therefore be the primary cause of the callosal defects. Further studies will be needed to elucidate this cause-consequence relation.
Even though the largest reduction in FA in the Cyfip1 +/mice was observed in the CC, other axonal tracts may also be affected. CYFIP1 is expressed across the whole brain 45,56 and therefore axonal defects in the Cyfip1 +/mice may be generalized. Indeed, both fractional anisotropy and functional connectivity are generally reduced across the whole brain in Cyfip1 +/mice. Additionally, defects in FC and white matter architecture have been reported across the brain in patients with ASD or SCZ 15,19,22,57 . Long-range connectivity problems between brain areas may also contribute, in addition to the callosal defects, to the observed deficits in motor coordination.
Importantly, patients with the 15q11.2 BP1-BP2 microdeletion syndrome present callosal defects. Notably, a high proportion of 15q11.2 patients (42%) show motor delay 28 , which may correlate with the reduced motor coordination we observed ( Figure 6). In addition, motor coordination deficits have been observed in patients with ASD 58,59 . Of note, the BP1-BP2 region contains four genes. Here, we show that the callosal defects are caused by the sole deletion of the Cyfip1 gene, giving further evidence that human CYFIP1 is the causative gene in the BP1-BP2 region.
Alterations in the callosal region are not only found in patients with 15q11.2 BP1-BP2 deletions and duplications [31][32][33] , they are also a hallmark of ASD 24 and SCZ 21 . Thus, variations in other chromosomal regions genetically associated with ASD or SCZ present similar defects in brain structure and connectivity as the Cyfip1 haploinsufficient mice. For instance, 22q11.2 deletion carriers, a chromosomal rearrangement that is associated with SCZ, present structural abnormalities in several brain regions 60,61 which are recapitulated in a mouse model for the disease (Df(16)A +/-) 62 . Similarly, studies in humans carrying the autismassociated 16p11.2 microdeletion showed reduced prefrontal connectivity and microstructural defects in the corpus callosum. These defects were also found in the mouse model for the disease, in which the callosal abnormalities were associated with increased axonal g-ratio 63 . Finally, genetic variations in the CNTNAP2 gene have been associated with ASD 64 . Absence of CNTNAP2 leads to defects in callosal transmission and cortical myelination 52 , as well as reduced functional connectivity and aberrant g-ratio of the callosal axons 65,66 . Taken together, these studies point out that neuropsychiatric disorders of different genetic origins present similar defects in brain connectivity and white matter architecture, especially in the corpus callosum, highlighting the importance of brain connectivity as a possible point of convergence for several neuropsychiatric disorders 67 . In addition, some of the areas that show reduced functional connectivity in the Cyfip1 +/mice are part of the default mode network (DMN), a brain network that reflects functional connectivity across different brain regions during conscious-inactive tasks 68,69 . Although the exact function of the DMN is still unknown, DMN defects have been found in patients with SCZ and ASDs 70-73 and have been suggested to be key for understanding the pathophysiology of these disorders 71,74 .
All together, the connectivity deficits caused by CYFIP1 deficiency could therefore have a major contribution in the pathogenesis of neuropsychiatric disorders.
In conclusion, here we found that Cyfip1 haploinsufficiency is directly responsible for key phenotypes of patients with ASD and SCZ, namely the alterations in anatomic and functional connectivity, thereby explaining the increased incidence of these pathologies in patients with 15q11.2 deletions. Importantly, this pathological mechanism might not be limited to neuropsychiatric disorders caused by CNVs in the 15q11.2 region. CYFIP1 has been found dysregulated at the protein level in SCZ independently of mutations in the CYFIP1 locus 75 , indicating that other genetic and/or environmental factors may converge on the regulation of CYFIP1 abundance or activity, rendering the protein a hub for the development of SCZ and perhaps also related disorders.

Animal care
Animal housing and care was conducted according to the institutional guidelines that are in
RsfMRI imaging procedures were performed as previously described 76,77 . In brief, a combination of medetomidine (Domitor, Pfizer, Karlsruhe, Germany) and isoflurane was used to sedate the animals. After positioning of the animal in the scanner, medetomidine was administered subcutaneously as a bolus injection (0.3 mg/kg), after which the isoflurane level was immediately decreased to 1%. Five minutes before the rsfMRI acquisition, isoflurane was decreased to 0.4%. RsfMRI scans were consistently acquired 40 min after the bolus injection, during which the isoflurane level was maintained at 0.4%. After the imaging procedures, the effects of medetomidine were counteracted by subcutaneously injecting 0.1 mg/kg atipamezole (Antisedan, Pfizer, Karlsruhe, Germany).
Regarding DTI measurements, after the handling procedures under isoflurane (2.5%), isoflurane levels were decreased to 1.5% and maintained throughout the scanning procedure. The physiological status of all animals was monitored throughout the imaging procedure. A pressure sensitive pad (MR-compatible Small Animal Monitoring and Gating system, SA Instruments, Inc.) was used to monitor breathing rate and a rectal thermistor with feedback controlled warm air circuitry (MR-compatible Small Animal Heating System, SA Instruments, Inc.) was used to maintain body temperature at (37.0 ± 0.5) °C

Imaging procedures
RsfMRI procedures were performed on a 9.4T Biospec MRI system (Bruker BioSpin, Germany) with the Paravision 5.1 software (www.bruker.com). Images were acquired using a standard Bruker cross coil set-up with a quadrature volume coil and a quadrature surface coil for mice. Three orthogonal multi-slice Turbo RARE T2-weighted images were acquired to render slice-positioning uniform (repetition time 2000 ms, effective echo time 33 ms, 16 slices of 0.5 mm). Field maps were acquired for each animal to assess field homogeneity, followed by local shimming, which corrects for the measured inhomogeneity in a rectangular VOI within the brain. Resting-state signals were measured using a T2*-weighted single shot DTI was performed on a 7T Pharmascan system (Bruker BioSpin, Germany). Images were acquired using a Bruker cross coil set-up with a transmit quadrature volume coil and a receive-only surface array for mice. Three orthogonal multi-slice Turbo RARE T2-weighted images were acquired to render slice-positioning uniform (repetition time 2500 ms, effective echo time 33 ms, 18 slices of 0.5 mm). Field maps were acquired for each animal to assess field homogeneity, followed by local shimming, which corrects for the measured inhomogeneity in a rectangular VOI within the brain. DTI images were acquired using a multislice two-shot spin-echo EPI sequence (repetition time 5500 ms, echo time 23.23 ms, 18 slices of 0.5 mm, b=800 s/mm², 60 DW direction). The field of view was (20 x 20) mm² and the matrix size (96 x96).

Image processing
Pre-processing of the rsfMRI and DTI data, including realignment, normalization and smoothing (for the rsfMRI data), was performed using SPM8 software (Statistical Parametric Mapping, http://www.fil.ion.ucl.ac.uk). First, all images within each session were realigned to the first image. This was done using a least-squares approach and a 6-parameter (rigid body) spatial transformation. For the analyses of the rsfMRI data, motion parameters resulting from the realignment were included as covariates to correct for possible movement that occurred during the scanning procedure. Second, all datasets were normalized to a study-specific EPI template. The normalization steps consisted of a global 12-parameter affine transformation followed by the estimation of the nonlinear deformations. Finally, in plane smoothing was done for the rsfMRI data using a Gaussian kernel with full width at half maximum of twice the voxel size (0.31 X 0.62 mm²). All rsfMRI data were filtered between 0.01-0.1 Hz using the REST toolbox (REST1.7, http://resting-fmri.sourceforge.net). Next, a seed-based analysis was performed identifying all functional connections of the specific region to other voxels in the brain. Seed-based analyses were performed by first computing individual z-transformed FC-maps of the respective region using the REST toolbox, after which mean statistical FC-maps were calculated for each group in SPM8.

Regions-of-interest (ROIs
Each individual DTI FA-map was warped to a reference anatomical image. Then, average FA-maps for WT and Cyfip1 +/mice were computed and a differential map (Δ(WT-Cyfip1 +/-)) was calculated. Finally, DTI parameters (i.e. radial diffusivity, axial diffusivity, mean diffusivity and fractional anisotropy) were computed with MATLAB and statistical analyses were performed using two sample t-tests to compare group differences in the corpus callosum.

Electron microscopy
Adult mice from both genotypes were perfused, via the heart, with 100 ml of a buffered mix of 2.5% glutaraldehyde and 2.0% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4), and then left for 2h before the brain was removed, embedded in 5% agarose, and sagittal vibratome sections cut at 80 µm thickness through the midline. These sections were then post-fixed in potassium ferrocyanide (1.5%) and osmium (2%), then stained with thiocarbohydrazide (1%) followed by osmium tetroxide (2%). They were then stained overnight in uranyl acetate (1%), washed in distilled water at 50°C before being stained with lead aspartate at the same temperature. They were finally dehydrated in increasing concentrations of alcohol and then embedded in spurs resin and hardened at 65°C for 24h between glass slides. The regions containing the corpus callosum were trimmed from the rest of the section using a razor blade and glued to a blank resin block. One micrometer thick sections were then cut from the block face and mounted onto silicon wafers of 1 cm diameter.
The sections were imaged inside a scanning electron microscope (Zeiss Merlin, Zeiss NTS) at a voltage of 2 kV and image pixel size of 7 nm. Backscattered electrons were collected with a Gatan backscattered electron detector with a pixel dwell time of 1 µs. Multiple images of the corpus callosum were collected and tiled using the TrakEM2 plugin 78 in the FIJI software (www.fiji.sc).
A custom-made MATLAB code was used to identify and parameterize myelinated axons within the EM images. Inner (axon) and outer (axon + myelin) areas were segmented and their corresponding circular equivalent diameters calculated. The g-ratio was calculated as the ratio between the internal and external diameter (d/D) (Figure 3b).

Multi Electrode Array (MEA) recordings
Cortical slices were prepared from adult male mice to study the effects of Cyfip1 haploinsufficiency in spontaneous activity. Briefly, the mice were anesthetized with isoflurane and the brains were quickly removed and placed in bubbled ice-cold 95% O 2 /5% CO 2 -

Patch clamp recordings
Animals were decapitated and the brain was quickly extracted. Slicing was done in bubbled ice-cold 95% O 2 /5% CO 2  Rotarod. Motor coordination was tested on the accelerating rotarod as previously described (Ugo Basile Model 7500, Gemonio, Italy) 79 . Briefly, mice were first trained for 2 min on the rotarod at constant speed (4 rpm). They were subsequently tested on four 5-min trials interleaved with 10 min rest. During the test trials, mice were placed on a rotating rod that accelerated from 4 to 40 rpm in 5 min and the latency to fall off the rod was recorded.

Statistics
Statistical analyses were performed with SPSS and Matlab for the MRI and DTI experiments, and with GraphPad Prism or R for EM, electrophysiology, and behavioral experiments. The statistical tests used are listed in the respective Figure legends. In brief, unpaired Student's ttest or non-parametric Mann-Whitney U-test were used for comparisons between the two groups. Multiple-t-test with Holm-Sidak multiple comparison correction was used for the rsfMRI experiments. Two-way ANOVA with Holm-Sidak's multiple comparison test was used for analysis of the g-ratio across different axonal sizes. For cumulative frequency distribution, Kolmogorov-Smirnov test was used. Paired pulse ratios and latency to fall were compared with Two-way Repeated Measures ANOVA. For all analysis, P values < 0.05 were considered significant. Results were presented as mean ± standard error of the mean (SEM).    (c) Graphs show average axonal diameter, myelin thickness and g-ratio. The histograms on the right show the g-ratio of axons with different diameters (n >10000 axons for each genotype, n=3 mice for each genotype, mean ± SEM; Mann-Whitney test; myelin thickness p<0.0001, g-ratio p<0.0001; Two-way ANOVA with Holm-Sidak's multiple comparison test, gratio p<0.0001). showing the spike events in WT and Cyfip1 +/cortical slice recordings. The x-axis corresponds to the recording time and the y-axis to the electrode ID. (d) Quantification of the spike rate (left) and burst rate (right, both on a logarithmic scale) in WT and Cyfip1 +/brain slices at P60 (WT n=12 slices, 6 mice, and Cyfip1 +/-n=28 slices, 13 mice) (mean ± SEM; two sample t-test; p=0.032 for spike rate and p=0.027 for burst rate). Center, cumulative frequency distribution of the spike rate for WT and Cyfip1 +/mice (Kolmogorov-Smirnov test, p<0.001).