Roles of developmentally regulated KIF2A alternative isoforms in cortical neuron migration and differentiation

Alternative splicing (AS) is one of the major molecular processes underlying increased transcriptome and proteome diversity and associated organismal complexity in metazoans (Dhananjaya et al., 2018; Su et al., 2018; Zhang et al., 2016; Zhang et al., 2014). High-throughput sequencing efforts have revealed that approximately 95% of all genes in the human body are alternatively spliced, generating more than 100,000 different mRNA isoforms, and the vast majority of isoform functions are completely unknown. AS is especially prevalent in the nervous system (Pan et al., 2008; Wang et al., 2008) and its impairment is often associated with neurological disease (Irimia et al., 2014; Scotti and Swanson, 2016). AS is dynamically regulated during brain development, where it modulates important processes including neural progenitor cell-to-neuron transition, migration, axon guidance and synaptogenesis (Raj and Blencowe, 2015; Su et al., 2018; Yeo et al., 2004). Despite important advances in our understanding of how AS is regulated in a tissue-specific manner to allow transcriptome and proteome diversification, little is known about how individual alternative isoforms function to regulate nervous system development. In particular, only a small fraction of neurodevelopmental studies have investigated functions of distinct isoforms, and only the most abundant isoform is typically studied. Here, we have focused on the separable functions of three different isoforms of the kinesin motor protein 2A (KIF2A) within the context of mouse cerebral cortex development.

KIF2A, a microtubule (MT)-depolymerizing kinesin, functions in mitotic spindle assembly (Ganem and Compton, 2004) and primary cilium disassembly in mitotic cells (Miyamoto et al., 2015; Zhang et al., 2019), and in axonal pruning in differentiating neurons (Homma et al., 2018; Maor-Nof et al., 2013). In humans, mutations in the motor domain of KIF2A are causative for malformations of cortical development, manifesting as lissencephaly, microcephaly, pachygyria and heterotopias (Broix et al., 2018; Cavallin et al., 2017; Poirier et al., 2013; Tian et al., 2016). Recent studies have suggested that Kif2a AS is regulated by neuron-specific RNA-binding proteins (Ince-Dunn et al., 2012) and its isoform abundance dynamically shifts during the neural progenitor-to-neuron transition during development of the cerebral cortex (Zhang et al., 2016). However, how individual KIF2A isoforms function during cortical development has not been studied previously. Here, we show that KIF2A is essential for proper dendritic arborization of cortical neurons and that all three KIF2A isoforms are interchangeable during this process. Moreover, we demonstrate that only two of the three KIF2A isoforms can functionally support radial neuronal migration in mice. We reveal isoform-specific interacting protein partners of KIF2A by a proximity-ligation-based proteomic study carried out in a neuronal cell line. In addition to a number of MT-binding proteins, our interactome screen also identifies mitochondrial proteins and nuclear RNA-binding proteins as either common or isoform-specific interactors, raising a potential role for KIF2A in mitochondrial assembly and RNA localization. Finally, we apply our proximity-based interactome mapping approach using KIF2A^H321D, which harbors a dominant-negative point mutation causative for malformations of cortical development in humans (Broix et al., 2018; Poirier et al., 2013). Intriguingly, we find significant rewiring of the protein interactome of KIF2A^H321D, with a remarkable loss of interactions with almost all mitochondrial proteins identified in wild-type KIF2A. Our results highlight the importance of investigating isoform-specific functions of gene products regulated by AS and reveal the utility of proximity based interactome mapping in identifying separable roles for either protein isoforms or mutants.

A previous study aimed at identifying the regulated targets of neuronal Elav-like RNA binding proteins (nElavl RNABPs) in the developing mouse cortex had identified Kif2a pre-mRNA as a top target (Ince-Dunn et al., 2012). In mice, the Kif2a gene consists of 20 coding exons, and nElavl proteins promote the inclusion of exon 18, which encodes for 37 amino acids rich in serine residues. While cloning these two major isoforms from the postnatal mouse cerebral cortex, we also identified a third isoform that includes exon 18, along with a shorter (by 20 amino acids) exon 5 generated as a result of alternative 5′ splice site selection (Fig. 1A). We numbered these isoforms as Kif2a.1 (exon 18+ and long exon 5), Kif2a.2 (exon 18− and long exon 5) and Kif2a.3 (exon 18+ and short exon 5) (Fig. 1A).

Next, we quantified the relative mRNA abundance of the three Kif2a isoforms in the developing cerebral cortex using quantitative RT-PCR. Our results demonstrated that Kif2a.1 and Kif2a.2 were the predominant isoforms, manifesting higher embryonic expression levels compared with postnatal and adult stages. Kif2a.3 was generally the minor isoform, with expression peaking during late embryogenesis and once again during the second postnatal week (Fig. 1B). Taken together, our results indicate an overall trend in which expression of Kif2a isoforms is high during embryonic and early postnatal stages of corticogenesis, a time period in which neural progenitor proliferation, active neuronal migration and dendritogenesis ensues.

To understand whether the three KIF2A isoforms may have divergent functions we asked whether AS had an effect on their subcellular localization. For this purpose, we imaged all three KIF2A isoforms in Neuro2A cells and primary cortical neurons after expressing EGFP-KIF2A fusion proteins of individual isoforms. Our images revealed that all three KIF2A isoforms are localized predominantly to the cytoplasm and nucleus in both cell types (Fig. 2A; Fig. S1). Furthermore, all three isoforms similarly localized to the soma, axons, dendrites and growth cones in primary neurons (Fig. 2A; Fig. S2). Our results therefore suggest that AS of KIF2A does not influence its subcellular localization in Neuro2A cells and cultured cortical neurons.

Previous studies have demonstrated a role for KIF2A in pruning of axon collaterals of cortical neurons and sensory neurons in mice (Homma et al., 2003; Maor-Nof et al., 2013). As regulation of MT dynamics is also crucial for dendritic arbor growth and pruning (Urbanska et al., 2008), we next evaluated the role of KIF2A isoforms in dendrite development. Initially, we verified that KIF2A was indeed expressed specifically in neurons and not in glia of our primary cultures, by carrying out co-immunofluorescent stainings against KIF2A and the neuronal marker NEUROD2 or the glial marker GFAP (Fig. S3). Next, we knocked down Kif2a expression in primary cortical neurons by expressing a short hairpin RNA (shRNA) targeting a common sequence in all three isoforms (Fig. 2B). We co-expressed soluble EGFP for imaging neuronal processes and measured dendritic arborization by Sholl analysis, which is a quantitative measure of dendritic length and branching (Fig. 2D) (Ferreira et al., 2014; Schindelin et al., 2012). Our results revealed a significant increase in the number of primary dendrites, consistent with a diminished capacity for MT depolymerization and dendritic pruning in the absence of KIF2A (Fig. 2E). In order to test the ability of different KIF2A isoforms to carry out their role in dendrite arborization, we knocked down all Kif2a expression, and then returned one isoform at a time by re-introducing shRNA-resistant forms of individual isoforms (resKif2a.1, resKif2a.2 and resKif2a.3) (Fig. 2C,D). As above, we quantified dendritic development by Sholl analysis. Our analysis revealed that all three isoforms of KIF2A were able to suppress the growth of superfluous dendrite branches and could reverse the KIF2A loss-of-function phenotype equally well (Fig. 2F-H). Taken together, we conclude that KIF2A is required for normal dendrite development in cortical neurons and all three KIF2A isoforms may carry out this function.

Previous studies have demonstrated that KIF2A is required for cortical neuronal migration (Broix et al., 2018; Ding et al., 2019). Mutations in Kif2a affect neuronal positioning in the cerebral cortex and cause cortical malformations in humans (Broix et al., 2018; Ding et al., 2019; Poirier et al., 2013). Consequently, we investigated whether the role of KIF2A in cortical neuronal migration is isoform-dependent. During development, the neural progenitor cells divide in the ventricular zone (VZ) of the embryonic telencephalon. After exiting the cell cycle, the early neurons radially migrate through the subventricular and intermediate zones (SVZ-IZ) until they reach the most superficial layer of the developing cortical plate (CP) (Kriegstein and Noctor, 2004). In order to assess the role of different KIF2A isoforms in this process, we knocked down expression of all Kif2a isoforms, then re-introduced one isoform at a time by expressing shRNA-resistant Kif2a cDNAs in ventricular neural progenitors of embryonic day (E) 14.5 cerebral cortex by in utero electroporation. We also electroporated a plasmid expressing the tdTomato marker and quantified the extent of neuronal migration by quantifying the number of tdTomato-positive neurons as a function of distance away from the VZ (Fig. 3A,B). As expected, in control samples ∼45% of tdTomato-positive neurons had reached the CP, ∼50% were migrating along the SVZ-IZ region and ∼5% were localized to the VZ at 3 days post-electroporation (Fig. 3A). Consistent with previous observations (Broix et al., 2018; Ding et al., 2019), there was a dramatic reduction in the relative fraction of tdTomato-positive neurons localized to the CP in Kif2a knockdown samples, and deficient migration caused most of the neurons to accumulate in the SVZ-IZ (Fig. 3A). In cortices where Kif2a expression was rescued with either resKif2a.1 or resKif2a.2, neurons could exit the SVZ-IZ and reach the CP, at levels comparable with control samples (Fig. 3C,D,F,G). In contrast, resKif2a.3-expressing neurons exhibited a severe migration defect, and the number of neurons that could migrate to the cortical plate was significantly reduced (Fig. 3E,H). It is noteworthy to mention that KIF2A.3 differs from the other two isoforms by 20 amino acids encoded by exon 5, which corresponds to a disordered protein region predicted by the IUPred2A prediction tool (Dosztányi, 2018) (Fig. S4). Interestingly, tissue-specific splicing of disordered protein regions is increasingly acknowledged as a means to increase interaction networks across tissue (Buljan et al., 2013). Taken together, our results suggest that KIF2A.1 and KIF2A.2 isoforms, but not the KIF2A.3 isoform, can support neuronal migration in the developing cortex.

Experimental analysis and computational modeling have suggested that an important function of AS is to increase phenotypic complexity by expanding the interactome (Buljan et al., 2013; Ellis et al., 2012; Kandoi and Dickerson, 2019). Therefore, we next asked whether different KIF2A isoforms were indeed characterized by different protein interaction networks. Toward this goal, we performed a proximity-labeling method which allows the identification of potential interactors of a protein of interest by its expression as a promiscuous biotin-ligase (the BirA* gene) fusion in the cell type of choice, followed by streptavidin-based affinity purification and mass spectrometry analysis (BioID) (Roux et al., 2013). In order to harvest sufficient total biotinylated protein, we decided to use the easily transfectable Neuro2A neuronal cell line for our BioID experiments. Initially, we confirmed that our BirA*-KIF2A fusion proteins were functional. For this, we transiently transfected BirA*-KIF2A fusions for each of the three isoforms, supplemented the cells with 0.05 mM biotin and 1 mM ATP, and confirmed that biotinylated proteins did indeed co-localize with expressed BirA*-KIF2A fusion proteins (Fig. 4A). Next, we verified that biotinylation was dependent upon the external supply of biotin, and that the background signal in the absence of external biotin following streptavidin pulldown was very low (Fig. 4B,C).

Having validated the BioID approach, we scaled up our protocol and deployed tandem mass spectrometry for analysis of purified proteins after streptavidin pulldown. We analyzed three biological replicates for each isoform, and based on peptide spectra match (PSM) values, we filtered out proteins identified in only one of the replicates, those with false discovery rate (FDR)>0.05, and those with fold-change<1.5. Non-transfected Neuro2A cells were used as control for background subtraction. Following these cut-off criteria, we identified a total of 63 potential interactors for KIF2A.1, 37 for KIF2A.2 and 40 for KIF2A.3 (Fig. 5A,G; Table S1). As expected, KIF2A itself was among the highest ranking hits for each of the three isoforms. We identified proteins that were common among all three KIF2A isoforms, shared in all possible pairwise combinations, as well as those that were unique for KIF2A.1, KIF2A.2 or KIF2A.3 (Fig. 5A,G; Table S1). Next we carried out gene ontology (GO) and interactome analysis using the String Database (Cline et al., 2007; Doncheva et al., 2019; Raudvere et al., 2019; Szklarczyk et al., 2019) separately with interactors of each of the KIF2A isoforms (Fig. 5D-F). As expected, a number of MT-associated proteins which function at the spindle poles during mitotic division were identified among the interactomes of at least one of the isoforms (e.g. CKAP5, NUMA1, CEP170 and MAP4). Our analysis revealed the highest number of interactors for KIF2A.1. Intriguingly, these were enriched for mitochondrial proteins with functions in mitochondrial gene expression (e.g. TIMM44, PNPT1, LARS2, RARS2 and TSFM) and metabolism (e.g. GLS, ALDH2, OAT, SUCLG2 and SUCLA2), and nuclear RNA regulation (e.g. SRSF3, SRSF4, SRRM2 and HNRNPA1) (Fig. 5D,G). KIF2A.2 displayed the least number of interactors, many of which were shared with KIF2A.1, with similar functions in mitochondrial metabolism and nuclear RNA transport (Fig. 5E,G). KIF2A.3 displayed the highest relative fraction of unique interactors. Its interactome was enriched in functions relating to nuclear RNA transport and regulation. Unlike the other two isoforms, KIF2A.3 was largely devoid of genes functioning in the mitochondria (Fig. 5F,G). Intriguingly, all KIF2A isoforms also appeared to interact with different eukaryotic translation factors (EIF5, EIF4G2, EEF1B2 and EIF3E). Finally, in order to verify that our BioID approach was able to identify endogenous interactions of KIF2A, we analyzed interaction between KIF2A and reticulon 4 (RTN4; also known as NOGO-A), which was identified in all three KIF2A isoform samples. Our co-immunofluorescence staining with KIF2A and RTN4 antibodies displayed a partial co-localization (Fig. S5). We also confirmed interaction by proximity ligation assay (PLA), which is a sensitive and specific method for detecting protein interactions with spatial resolution (Debaize et al., 2017). Our PLA results revealed a significant interaction between KIF2A and RTN4, compared with negative controls in which only anti-KIF2A or anti-RTN4 primary antibodies were used for the PLA protocol (Fig. 5B,C). In sum, our data identify common and unique interactors of the three KIF2A isoforms.

Having established a system to uncover the protein interaction network of KIF2A, we next decided to turn our attention to KIF2A mutations that cause developmental cortical malformations in humans (KIF2A^S317N and KIF2A^H321D; Broix et al., 2018; Poirier et al., 2013). Initially, we generated identical disease-associated point mutations at conserved residues in the backbone of mouse Kif2a.1 cDNA. To evaluate how these two KIF2A.1 mutants localize subcellularly in neuronal cells, we expressed EGFP-KIF2A^S317N and EGFP-KIF2A^H321D in Neuro2A cells and mouse primary cortical neurons. Consistent with previous observations (Broix et al., 2018; Poirier et al., 2013), both KIF2A^S317N and KIF2A^H321D mutants were sequestered along polymerized MTs and did not display the typical cytoplasmic and nuclear distribution pattern we observed for wild-type KIF2A.1 in both cell types (Figs 6A and 7B).

Next, we asked whether these dominant mutations affected dendrite development in cortical neurons. For this purpose, we co-transfected E14.5 mouse primary cortical neurons with a non-silencing shRNA or shKif2a along with either one of the shRNA-resistant Kif2a mutants (resKif2a^S317N or resKif2a^H321D) and carried out Sholl analysis. We confirmed that these mutations did not cause protein instability and were expressed at levels comparable with wild-type KIF2A (Fig. S6). We observed significantly increased dendrite arborization in neurons when Kif2a was silenced (Fig. 6B,C). Furthermore, co-expressing shRNA-resistant forms of either Kif2a^S317N or Kif2a^H321D did not rescue this defect, whereas wild-type Kif2a.1 did so efficiently (Fig. 6B-F). Interestingly, if we expressed Kif2a^S317N or Kif2a^H321D in otherwise wild-type neurons without silencing endogenous Kif2a, we did not observe any effect on dendrite arborization (Fig. S7). Therefore, our results suggest that disease-causing dominant mutations in Kif2a cause dendrite arborization defects in primary cortical neurons, and this effect depends on the expression level of endogenous wild-type KIF2A.

Next, we analyzed results from our previous BioID proximity labeling experiments in which KIF2A^H321D (along with the three isoforms) was used as a bait. We compared proteins identified in KIF2A^H321D and KIF2A.1 samples as the mutation was created within this isoform. Our comparison revealed widespread rewiring of the interactome upon introduction of the H321D mutation. We identified 52 proteins that were unique to KIF2A.1, 31 unique to KIF2A^H321D and 11 that were common to both (Fig. 4A-C; Fig. 7A,E; Table S1). Next, we generated a GO and interactome map for KIF2A^H321D and compared it to that of KIF2A.1, using the String database (Cline et al., 2007; Doncheva et al., 2019; Raudvere et al., 2019; Szklarczyk et al., 2019). The relative fraction of cytoplasmic, nuclear and MT-associated genes was preserved in KIF2A^H321D; however, the identities were different (Fig. 7C-E). The most striking observation was the depletion of mitochondrial proteins from the interactome of KIF2A^H321D in comparison with KIF2A.1 (16 mitochondrial proteins out of a total of 63 proteins for KIF2A.1 and one out of 42 for KIF2A^H321D).

Our proximity biotinylation results from Neuro2A cells revealing a large number of unexpected potential interactions with mitochondrial proteins for isoforms KIF2A.1 and KIF2A.2, but far fewer for KIF2A.3 isoform and KIF2A^H321D mutant, prompted us to further analyze mitochondrial proximity of KIF2A in primary cortical neurons. We tested whether individual KIF2A isoforms and mutants displayed differential levels of co-localization with mitochondria by co-transfecting primary cortical neurons with EGFP-KIF2A fusion constructs together with mCherry fused to the outer mitochondrial protein TOMM20 (mCherry-TOMM20). Two days after transfection we imaged neurons by confocal microscopy and calculated Manders’ co-localization coefficients for soma, axon and dendrite compartments (Fig. 8A,C). We used cytoplasmic EGFP-expressing neurons as negative controls. In the soma, KIF2A.1 and KIF2A.2 isoforms displayed levels of mitochondrial co-localization which were generally comparable with those of the negative controls, and KIF2A.3 showed a slight but statistically significant reduction (Fig. 8B; Fig. S8). Within the axon and dendrite compartments both KIF2A.1 and KIF2A.2 isoforms displayed significantly increased mitochondrial co-localization; however, this effect was less pronounced for KIF2A.3 in axons and was not significantly different from negative controls in the dendritic compartment (Fig. 8B). Intriguingly, in comparison with wild-type KIF2A.1, the mutants KIF2A^S317N and KIF2A^H321D showed a drastic reduction in mitochondrial co-localization in soma, axons and dendrites (Fig. 8C,D). These results are generally consistent with our BioID findings, in which KIF2A.1 and KIF2A.2 isoforms displayed the largest number of mitochondrial proteins in their interactome datasets.

Mitochondrial localization to neuronal processes is unchanged in cortical neurons of Kif2a knockout mice (Homma et al., 2003). Considering the possibility of compensatory mechanisms that might operate during a relatively longer period of brain development in mice, we asked whether an acute depletion of KIF2A in primary cortical neurons would cause mitochondrial localization defects. After co-transfection with shKif2a and mCherry-TOMM20, we quantified mCherry signal in neuronal soma, axon and dendrites. Our results show that mitochondrial localization in these compartments is comparable between Kif2a silenced and control neurons (Fig. S9). Taken together, our results suggest a potential isoform-specific interaction between KIF2A and mitochondria, specifically in the axonal and dendritic compartments of neurons. The potential functional role of these interactions remain to be solved in the future.

Kif2a is an alternatively spliced gene, and isoform abundance is developmentally regulated (Ince-Dunn et al., 2012; Zhang et al., 2016). Here, we reveal that all three KIF2A isoforms can rescue a dendritic pruning defect caused by knockdown of all KIF2A expression in cortical neurons. However, only KIF2A.1 and KIF2A.2, but not KIF2A.3, are able to support migration of neurons in the developing cortex. KIF2A.3 differs from the other two isoforms by a 20 amino acid-long serine-rich disordered protein region located immediately N-terminal to its dimerization domain. Currently, we do not know how this short stretch of amino acids affects KIF2A function and how these functional changes might impact neural progenitor division or neuronal migration in the developing cortex. However, the results from our isoform-specific proximity-labeling screen demonstrated that the protein interactome of KIF2A.3 is significantly different from the other two isoforms.

Our BioID assay for KIF2A carried out in a neuronal cell line identified a number of centrosomal proteins with roles in bipolar spindle assembly, of which NUMA and CEP170 are known KIF2A interactors (Gaetz and Kapoor, 2004; Welburn and Cheeseman, 2012; Zhang et al., 2019). In mitotic cells KIF2A is essential for bipolar spindle assembly and at the poles it depolymerizes minus-end MTs to contribute to poleward chromosome movement (Ali et al., 2017; Ganem and Compton, 2004; Manning et al., 2007). Here, we provide additional candidate KIF2A-interacting proteins such as KMT2E, CKAP5, RANBP2 and CHD4, which also localize to bipolar spindles. Further investigations on how KIF2A and these candidate proteins functionally interact is likely to provide new mechanistic understanding of bipolar spindle assembly and chromosome segregation. These interactions may be particularly relevant for dividing neural progenitors in the VZ of the developing cortex.

We also identified unexpected KIF2A interactors such as a number of proteins targeted to organelles, mostly to the mitochondria, RNA processing enzymes and translation factors. We recognize that a subset of these hits may represent potential contaminants in our dataset. However, given that many of them display clear isoform-specificity we wish to speculate on the exciting implications of these findings. Several of these proteins have well-described roles in MT-based cargo trafficking. Interestingly, the original biochemical characterization of KIF2A revealed clear interactions with vesicle structures in neurons that were distinct from synaptic vesicles (Noda et al., 1995). It is possible that organelle and vesicle trafficking appear to be unaffected in Kif2a knockout cells (Homma et al., 2003) owing to genetic redundancy with other kinesins. In support of this idea, our BioID results identify KTN1 and TRAK1, which are both kinesin adaptor proteins with essential roles in mitochondrial transport along MTs, as potential KIF2A interactors (Barel et al., 2017; Brickley and Stephenson, 2011; López-Doménech et al., 2018; MacAskill et al., 2009; Tanaka et al., 1998).

Perhaps the most interesting aspect of these unexpected hits is the fact that mitochondrial proteins are enriched in KIF2A.1 and KIF2A.2 interactomes and largely reduced in the dataset from KIF2A.3, the isoform which cannot support neuronal migration. Furthermore, we observed new candidates (such as SRP68, PES1 and EIF4ENIF1) that are only identified in the interactome of KIF2A.3. Consistent with our BioID analysis we further confirm that indeed KIF2A.1, KIF2A.2 and, to a much reduced extent, KIF2A.3 display mitochondrial co-localization in axons and dendrites of primary cortical neurons. Paradoxically, KIF2A does not harbor a mitochondrial-targeting sequence, is not listed as a mitochondrial protein in a mitochondrial proteome database (Calvo et al., 2016) and mitochondrial distribution is not affected in Kif2a knockout cells (Homma et al., 2003). Currently we do not know whether and how KIF2A-mitochondria co-localization affects mitochondrial trafficking, protein targeting or its metabolic functions. Intriguingly, several important studies have uncovered a role for MT- and kinesin-based trafficking of ribosome-associated mRNA complexes targeted to the mitochondrial surface for co-translational mitochondrial import (Fazal et al., 2019; Gagnon and Mowry, 2011; Kanai et al., 2004; Messitt et al., 2008). Whether KIF2A plays a direct role in MT-based RNA-protein complex transport, or might generally be in close proximity to a number of MT-associated-cargo due to its function in MT depolymerization, remains to be discovered.

Previous findings have demonstrated an essential role for KIF2A in axonal pruning (Homma et al., 2003; Maor-Nof et al., 2013), neuronal migration (Gilet et al., 2020; Homma et al., 2003) and in primary cilium disassembly during mitotic division of neural progenitors (Broix et al., 2018; Ding et al., 2019; Miyamoto et al., 2015; Zhang et al., 2019). As Neuro2A cells are actively dividing and do not possess primary cilia, our BioID assay did not identify primary cilium-specific proteins. However, several of the candidate KIF2A interactors are likely to have important functions during development of the cerebral cortex. For example, we have verified interactions between the neurite outgrowth inhibitor RTN4 and endogenous KIF2A in Neuro2A cells. Interestingly, RTN4 acts a negative regulator of radial neuron migration, and cortical lamination is defective in Rtn4 knockout mice (Mathis et al., 2010). Among other interactors is CEP170, which is a KIF2A.2-specific interactor, regulates primary cilia disassembly and neural progenitor division in the VZ by functioning in a previously described WDR62-CEP170-KIF2A pathway (Zhang et al., 2019). CHD4 is a candidate interactor of KIF2A.1 and KIF2A.3 isoforms, and is a chromatin remodeling histone deacetylase, promoting the proliferation of neural progenitors at the cortical VZ during development (Nitarska et al., 2016). Finally, mutations in KMT2E, a lysine methyl transferase localized to spindle poles and a candidate interactor of all KIF2A isoforms, result in a range of neurodevelopmental disorders in humans, often manifesting with microcephaly and epilepsy (O'Donnell-Luria et al., 2019). Further characterization of these potential KIF2A interactors will reveal important details of how KIF2A controls early cortical development.

Mutations in the ATP binding sites of the KIF2A motor domain cause malformations of cortical development (Broix et al., 2018; Cavallin et al., 2017; Poirier et al., 2013; Tian et al., 2016). Patients manifest with a spectrum of symptoms which may include frontal band heterotopia, pachygyria, microcephaly, lissencephaly and neonatal onset seizure (Broix et al., 2018; Cavallin et al., 2017; Poirier et al., 2013; Tian et al., 2016). These mutations have been associated with severe migration defects of neurons in the developing cortex and defects in the MT depolymerization activity of KIF2A (Broix et al., 2018; Gilet et al., 2020), ability to regulate the structure and function of the MT-based structure the primary cilium and cell cycle defects in cortical neural progenitors (Broix et al., 2018). However, how these individual processes culminate in the symptomatic manifestations described in humans is unclear. Here, we investigated how one of these previously described motor domain patient mutations in KIF2A affect its interactome. Interestingly, we identified that the proximity-labeling-based interactome of KIF2A^H321D exhibits clear changes from its wild-type isoform (KIF2A.1). In particular, we observed a strong depletion of mitochondrial proteins in the interactome dataset of KIF2A^H321D and the presence of ectopic new proteins, the functions of which generally center on RNA regulation and transport. To date, few studies have investigated mitochondrial function in cortical neuron migration and lamination and have provided conflicting results. For example, although inhibition of mitochondrial protein adenine nucleotide transferase 1 (ANT1; SLC25A4) had no effects on cortical development (Lin-Hendel et al., 2016), knockout mice for mitochondria localized glutamic acid rich protein (Mgarp) showed defective radial migration and cortical axon and dendrite polarization, associated with abnormalities in motility and distribution of mitochondria (Jia et al., 2014). Our results suggest that a role for kinesins, and KIF2A in particular, in mitochondrial regulation during neuronal migration and cortical development should be investigated. Although further analysis is essential to resolve the biological relevance of our findings, the utility of our approach to identify potential interactors of different KIF2A isoforms and a mutant with relevance to pathogenesis of human cortical malformations is revealed.

Mouse Kif2a.1, Kif2a.2 and Kif2a.3 were cloned by reverse transcription from total RNA prepared from postnatal day (P) 0 mouse cortex, restriction digestion and ligation into the EcoRI/HindIII sites in the pEGFPC1 backbone vector. Subsequently myc-tagged versions of all isoforms and the mutants were subcloned into pcDNA4A-myc backbone vector using the EcoRI/BamHI sites. Mouse Kif2a.1, Kif2a.2, Kif2a.3 and Kif2a^H321D were subcloned from the pEGFPC1 vector into XhoI/EcoRI sites of the pcDNA3.1(-)-mycBirA backbone vector (BirA*, N terminal; kindly provided by Elif N. Fırat-Karalar, Koç University, Istanbul). The primer sequences were as follows: Kif2a_BirAN_F (AAACTCGAGATGGTAACATCTTTAAATGAAGAT); Kif2a_BirAN_R (AAAGAATTCTTAGAGGGCTGGGGCCTCTTGGG). H321D and S317N point mutations were created by site-directed mutagenesis of mouse pEGFPC1-Kif2a.1. The primer sequences were as follows: Kif2a_c959G>A_F (GCAGACTGGAAATGGGAAAACTCATACTATGG); Kif2a_c961C>G_F (GCAGACTGGAAGTGGGAAAACTGATACTATGG); Kif2a_sdm_R (CCATAAGCAAAGCACGTAGCCATGCCCCTTT). Non-silencing (NS) and shKif2a shRNA were cloned into pSUPER-neo-EGFP (www.oligoengine.com) as described in the manufacturer's manual. For in utero electroporation experiments, the EGFP cassette was exchanged with an mCherry cDNA by subcloning into the AgeI/NotI sites of pSUPER-neo. Three isoforms of shKif2a-resistant cDNA (resKif2a.1, resKif2a.2 and resKif2a.3) and mutants of shKif2a-resistant cDNA (resKif2a^S317N and resKif2a^H321D) were created by site-directed mutagenesis in a pcDNA4A backbone vector by introducing three silent mutations (A1521T, C1524G and T1527C) within the shKif2a target sequence (GGAATGGCATCCTGTGAAA) (Jang et al., 2008). The primer sequences were as follows: NS_F (GATCCCCGCGCGATAGCGCTAATAATTTTTCAAGAGAAAATTATTAGCGCTATCGCGCTTTTTA); NS_R (AGTCTAAAAAGCGCGATAGCGCTAATAATTTTCTCTTGAAAAATTATTAGCGCTATCGCGCGGG); shKif2a_F (GATCCCCGGAATGGCATCCTGTGAAATTCAAGAGATTTCACAGGATGCCATTCCTTTTTA); shKif2a_R (AGCTAAAAAGGAATGGCATCCTGTGAAATCTCTTGAATTTCACAGGATGCCATTCCGGG); resKIF2A_F (CCAGGAATGGCTTCGTGCGAAAATACTCT); resKIF2A_R (AGAGATTGTGGCAATCATGCAGGTACGAG). All constructs were confirmed by Sanger sequencing. mCherry-TOMM20-N-10 plasmid was acquired from Addgene (plasmid #55146).

Primary cortical cultures were prepared as previously described (Guner et al., 2017). Briefly, cortices from E14.5-E18.5 embryos were dissected in ice-cold 1× HBSS containing 10 mM HEPES and digested in 20 U/ml papain enzyme at 37°C for 10 min. We used 10 mg/ml trypsin inhibitor to terminate digestion. Tissue was washed with Basal Media Eagle supplemented with 0.5% L-glutamine, 1% penicillin/streptomycin and 5% fetal bovine serum (FBS) (BF) twice at room temperature and triturated three to five times. Then 300,000 cells were plated onto each well of a 24-well plate (pre-coated with laminin and poly-L-lysine) in BF. After 2 h incubation at 37°C, medium was changed into BF containing N-2 and B-27 supplements (Gibco Life Technologies).

Neuro2A and HEK293T cells were maintained in EMEM and DMEM, respectively, and each was supplemented with 10% FBS, 1% penicillin/streptomycin and 0.5% L-glutamine (complete media). Plasmids were transfected into primary cortical neurons at 2 days in vitro (DIV) using Lipofectamine 2000 (Invitrogen), or HEK293T cells using Lipofectamine 3000 (Invitrogen), according to the manufacturer's protocol. Neuro2A cells were transfected using 1 mg/ml Polyethylenimine (PEI) dissolved in sterile water (Polysciences). For immunoblotting, protein lysates were collected 2 days after transfection either in ice-cold RIPA buffer [0.05 M Tris-HCl (pH 7.5), 0.15 M NaCl, 1% Triton X-100, 1% Na-DOC, 0.1% SDS] supplemented with protease inhibitor or Lysis Buffer [10 mM Tris-HCl (pH 7.6) containing 0.5% SDS, 2% NP-40, 150 mM NaCl, 1 mM EDTA] supplemented with 10 mM iodoacetamide and protease inhibitor. Antibodies and dilutions used were as follows: myc (Santa Cruz Biotechnology, SC-40, 1:1000); hnRNP C1+C2 (Abcam, ab10294, 1:2000); biotin (gift from Dr Timothy J. Mitchison, Harvard Medical School, Boston, MA, USA, 1:20,000); Histone H3 (Cell Signaling Technology, 9715S, 1:2500); β-actin (Abcam, ab6276, 1:5000); EGFP (Santa Cruz Biotechnology, sc-8334, 1:1000).

For immunofluorescence staining, both Neuro2A cells and primary cortical neurons were transfected with Lipofectamine 2000 (Invitrogen) 2 days after plating and fixed with 4% paraformaldehyde (PFA) 2 days after transfection. Neuro2A cells were imaged using a Leica DMI8 SP8 CS/DLS confocal microscope and LasX Life Science software. Primary cortical neurons were imaged using a Nikon Eclipse 90i confocal microscope and NIS-Elements AR software, and a Leica DMI8 SP8 CS/DLS confocal microscope and LasX Life Science software. For co-localization analysis, quantification was performed using JACoP plugin which calculates the Manders’ Coefficients M1 and M2 and these values provide a direct measure of the quantity of interest independent of signal proportionality (Dunn et al., 2011; Manders et al., 1993; Zinchuk and Zinchuk, 2008). Sholl analysis was performed with the Sholl analysis plug-in in ImageJ software (Ferreira et al., 2014; Schindelin et al., 2012). Sholl analysis parameters are: 1 µm starting radius, 300 µm ending radius, 2 µm radius step size, 1 µm radius span, and span type, median. Antibodies and dilutions used were as follows: myc (Santa Cruz Biotechnology, SC-40, 1:1000); EGFP (Santa Cruz Biotechnology, sc-8334, 1:1000); GFP (Aves Labs, GFP-1020, 1:10,000); α-tubulin (Cell Signaling Technology, 3873S, 1:1000); β-III-tubulin (Tuj1) (Abcam, ab7751, 1:500); KIF2A (Abcam, ab71160, 1:1000); Streptavidin Alexa Fluor^®488 conjugate (Life Technologies, S32354, 1:5000); RTN4 (Nogo C-4, Santa Cruz Biotechnology, sc-271878, 1:500); NEUROD2 (Abcam, ab168932, 1:1000); GFAP (Sigma-Aldrich, G3893, 1:200).

Mice cortices at different developmental time points (E14, E16, E18, P0, P7, P14, P21, P28 and adult) were collected in 1× cold HBSS and total RNA was isolated in Trizol (Thermo Fisher Scientific). cDNA was synthesized using Transcriptor High-Fidelity cDNA Synthesis Kit (Roche) and qRT-PCR was performed with Luminaris HiGreen qPCR Master Mix (Thermo Fisher Scientific) using a CFX Connect Real-Time PCR detection System (Bio-Rad). Kif2a mRNA levels were normalized to Gapdh mRNA using the 2^−ΔΔCT method. The primer sequences were as follows: Gapdh_F (CGACTTCAACAGCAACTCCCATCTTCC); Gapdh_R (TGGGTGGTCCAGGGTTTCTTACTCCTT); Kif2a.1_F (CCAGGGTGAAAGAATTGACTG); Kif2a.1_R (GCTTCATGGAAAGTGAACAGC); Kif2a.2_F (GCAAATAGGGTGAAAGAATTGACTG); Kif2a.2_R (TCTGGGAGACAGCTTCATGG); Kif2a.1_ex5long_F (CTCCTTCACGTAGAAAATCCAATT); Kif2a.1_ex5long_R (GTCTCTGATCATACACATAATTTCGT); Kif2a.3_ex5short_F (CAGAATGCACGTAGAAAATCCAA); Kif2a.3_ex5short_R (TCTGAAGTCTCTGATCATACACATA).

The in utero electroporation experiment was performed as previously described (Guzelsoy et al., 2019; Saito, 2006) using E14.5 pregnant CD1 mice. Briefly, timed pregnant mice (E14.5) were anesthetized using isoflurane. The uterine horns were exposed by laparotomy and lateral ventricle of each embryo was injected using pulled glass capillaries (Drummond^® PCR micropipettes, pulled as 60 µm diameter) with 333 ng/µl NS (control) or shKif2a plasmids combined with 333 ng/µl resKif2a.1, resKif2a.2 or resKif2a.3 and 333 ng/µl pCAGGS_IRES_tdTomato to visualize transfected neurons and 0.1% Fast Green to monitor the injection (all plasmids were prepared using Qiagen Endofree Plasmid Maxi Kit). Plasmid DNA was electroporated into the neuronal progenitors located in the VZ by 0.5 pulses at 30 V for 400 ms intervals with an electroporator (BTX Harvard Apparatus, ECM830) using 5 mm tweezertrodes (BTX Harvard Apparatus). Uterine horns were placed back in the abdominal cavity, and development of embryos was allowed to continue for 3 days (E17.5). Embryo brains were harvested and fixed with 4% PFA in PBS overnight at 4°C. Brains were sectioned coronally at 50 µm using a cryostat (CM1950, Leica Biosystems). Optical z-series through 50 µm images were captured using a Leica DMI8 SP8 CS/DLS microscope and maximum intensity projections were generated in LasX Life Science software. Images were opened in Photoshop CC (Adobe Systems) for positioning the ten bins grid for the cell counting. Number of transfected cells were counted for each bin using the cell counter option of ImageJ (National Institutes of Health; Schindelin et al., 2012) and then data were transferred to Excel (Microsoft) for further analysis. Bins 1-5 represent CP, bins 6-9 represent SVZ-IZ and bin 10 represents VZ. Finally, graphs were plotted using GraphPad Prism 5 software.

Neuro2A cells were transfected with 60 µg BioID constructs using 1 mg/ml PEI. The following day, medium was changed with complete EMEM containing 0.05 mM D-biotin (Life Technologies) and 1 mM ATP. Then, 2 days after transfection, protein lysates were collected in Lysis Buffer supplemented with 10 mM iodoacetamide and protease inhibitor by lysing the cells using an insulin syringe seven times. Protein concentration was measured by BCA Assay (Thermo Fisher Scientific) and equal amounts of protein from each condition was mixed with an equal volume of Streptavidin Plus UltraLink^® Resin (Pierce^®), which was pre-washed with 1× PBS and Lysis Buffer. Proteins were incubated with streptavidin beads overnight at 4°C on a tube rotator. In order to eliminate sample loss, on-bead digestion for streptavidin-captured biotinylated proteins was performed the following day. Beads were washed with Wash buffers 1, 2, 3 and 4 twice for 10 min each [Wash Buffer 1: 2% SDS; Wash Buffer 2: 2% Na-DOC, 1% Triton X-100, 50 mM NaCl, 50 mM HEPES (pH 7.5), 1 mM EDTA; Wash Buffer 3: 0.5% NP-40, 0.5% Na-DOC, 1% Triton X-100, 500 mM NaCl, 1 mM EDTA, 10 mM Tris-HCl (pH 8.1); Wash Buffer 4: 50 mM Tris-HCl (pH 7.4), 50 mM NaCl]. For the affinity capture experiment followed by immunoblotting, proteins were eluted with elution buffer (500 nM D-biotin in 2× Laemmli Buffer with 0.1 M DTT) at 98°C for 10 min with 1000 rpm shaking.

For the identification of biotinylated proteins by mass spectrometry, beads were washed with urea solution [8 M urea in 0.1 M Tris-HCl (pH 8.5)] and then reduced with DTT (100 mM in urea solution) at 56°C for 30 min. Beads were washed with urea solution and then incubated with iodoacetamide (0.05 M in urea solution) for 20 min in the dark for alkylation of cysteine residues. Beads were washed with urea solution and 0.05 M ammonium bicarbonate solution (NH₄HCO₃) (0.05 M in HPLC grade water) twice for each buffer. For 600 µg of protein digestion, 4 µg of trypsin (MS Grade Trypsin Protease, Pierce) was added to beads for 16 h at 37°C with 1000 rpm shaking. Peptides were treated with 10% acetic acid in order to terminate trypsin digestion and peptides were eluted using C18 stage tips (Rappsilber et al., 2007). Eluates were dried in a speed vacuum system (Thermo Fisher Scientific Radiant Cover, rc110b) and were analyzed using a Q-Exactive LC-MS/MS mass spectrometer (Thermo Fisher Scientific). The datasets were analyzed against the Mus musculus Swissprot/Uniprot database (UniProt, 2019). For analysis, the spectral counts of proteins were used to calculate fold change ratios and FDR values for each KIF2A isoform and KIF2A^H321D mutant using the qspec-param program of qprot_v1.3.5. Proteins found in less than two biological replicates were filtered away. Proteins with the same gene name were merged via sum of PSM values. Normalization methods, z-statistics and FDR values were calculated as previously described (Choi et al., 2015). Proteins were filtered out if FDR>0.05 or fold change<1.5. Heat map was created with Prism 8. For cluster analysis, significant protein hits were analyzed using the String database v11.0 (Szklarczyk et al., 2019) by Cytoscape StringApp (Doncheva et al., 2019) with 0.4 confidence. Affinity propagation clustering of the network was performed by Cytoscape (version 3.7.2) and its plugin clustermaker (Cline et al., 2007). GO and KEGG enrichment analysis of the network was performed via g:Profiler (Raudvere et al., 2019).

Neuro2A cells were plated on 24-well plates with coverslips. The PLA was performed using Duolink^® In Situ Red Starter Kit Mouse/Rabbit (DUO92101, Sigma-Aldrich) according to the manufacturer's protocol. Briefly, cells were fixed with 3.2% PFA for 15 min and blocked with Duolink^® Blocking Solution for 30 min at 37°C. Cells were incubated with primary antibody pairs or only one primary antibody in Duolink^® antibody diluent overnight at 4°C. Cells were washed with Wash Buffer A and incubated with Duolink^® Plus and Minus PLA probes in antibody diluent for 1 h at 37°C. Cells were washed with Wash Buffer A and ligation was performed using ligase diluted in Duolink^® Ligation buffer for 30 min at 37°C followed by washes with Wash buffer A. Amplification was performed with polymerase diluted in Duolink^® Amplification buffer for 100 min at 37°C. Cells were washed with Wash buffer B followed by 0.01× Wash buffer B and finally mounted with mounting medium containing DAPI. Images were captured with a Leica DMI8 SP8 CS/DLS microscope using the DAPI channel to observe nuclei and the red channel (λex 594 nm; λem 624 nm) to observe PLA dots. Optical z-series images were projected into a flattened plane with the maximum intensity option of the LasX Life Science software. Image analysis was performed with ImageJ software (Debaize et al., 2017). Antibodies and dilutions used were as follows: KIF2A (Abcam, ab71160 1:5000); RTN4 (Nogo C-4, Santa Cruz Biotechnology, sc-271878, 1:500).

All animal experiments were performed in accordance with the guidelines for the care and use of laboratory animals of Koç University and the Ministries of Food, Agriculture and Livestock, Forestry and Water Management of Turkey. All mice were bred, maintained and housed at the Koç University Animal Facility. Ethics approval was obtained from Koç University Institutional Animal Care and Use Committee (license no. 2014–9).

We thank Ahmet Kocabay, Nilhan Coskun, Mustafa Demir and Muhammed Resat Oguzkan Yaralı (Koç University Animal Research Facility, Turkey) for their help in providing timed pregnant mice; the Molecular Imaging Core Facility of Koç University Research Center for Translational Medicine (KUTTAM), Berna Morova, Alper Kiraz from Koç University College of Science and Dr Sercin Karahuseyinoglu from KUTTAM for assistance with confocal microscopy; and Cory Dunn for feedback on the manuscript.

The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.192674.reviewer-comments.pdf