Alternative splicing regulates adaptor protein binding, trafficking, and activity of the Vps10p domain receptor SorCS2 in neuronal development

SORCS2 alternative splicing generates four variants with different ICDs

We searched the National Center for Biotechnology Information Expressed Sequence Tag database and identified several human SORCS2 clones displaying exon–exon junction variation. The putative translational products indicated the existence of four ICDs. To assess whether human SorCS2 mRNA is indeed alternatively spliced, human adult brain RNA was subjected to RT-PCR, amplifying fragments encoding SorCS2 ICD. Two DNA fragments denoted SorCS2 long and SorCS2 short, respectively, were amplified (Fig. 1A). DNA sequencing confirmed the identity of the Expressed Sequence Tag clones. The variants were denoted as SorCS2A, SorCS2B, SorCS2C, and SorCS2D (Fig. 1B). The differential 5′ splice site in intron 26 introduces an acidic cluster of 14 amino acids in variant SorCS2A, giving rise to SorCS2B. The differential 3′splice site in intron 25 introduces a serine residue at position 1104 (S1104) in variant SorCS2A giving rise to SorCS2C, whereas use of both alternative splice sites gives rise to SorCS2D (Fig. 1B). The acidic cluster motif within the long variants B and D is found in 31.2% of SorCS2 transcripts from human adult brain (Fig. 1C). The mixed complementary DNA (cDNA) sequencing profiles of the amplified long and short RT-PCR fragments allowed quantification of the ratio between SorCS2 transcripts with or without the three-nucleotide insertion encoding S1104 (Fig. 1D). This further enabled us to estimate the relative distribution of the four SORCS2 mRNA isoforms in whole brain tissue. Expression of SORCS2A mRNA prevailed, while SORCS2D exhibited the lowest level of expression (Fig. 1E). Interestingly, only two Sorcs2 transcripts were detectable in mouse brain, one of which contains an alternatively spliced insertion of three nucleotides in analogy to the human SORCS2C isoform (Fig. 1F). Hence, SorCS2 variants expressed in mouse brain do not contain the acidic cluster motif. Like human brain, expression of the Sorcs2A variant prevailed in mouse brain tissue being more than 3-fold higher than that of the mouse Sorcs2C variant (Fig. 1G).

SorCS2 splicing is differentially regulated during development and in tissues

We next sought to determine the relative distribution of human SorCS2 splice variants in different tissues. Cerebral cortex, hippocampus, substantia nigra, cerebellum, and spinal cord all showed expression of both SorCS2 long (SorCS2B+D) and short (SorCS2A+C) mRNA (Fig. 2, A, B, and D). The relative expression of SORCS2A was 2- to 3-fold higher than SORCS2B and SORCS2C. SorCS2D expression was further 2- to 3-fold reduced compared to SORCS2B and SORCS2C (Fig. 2, C and E). In fetal brain, an even higher relative expression of SORCS2A than SORCS2B and SORCS2D was observed, suggesting that SorCS2 splicing is developmentally regulated. In contrast to the CNS, dorsal root ganglia, heart, kidney, and adrenal gland expressed SORCS2A and SORCS2B at a more similar level. SORCS2C and SORCS2D are also found at roughly similar levels, however the relative expression is approximately 2- to 3-fold reduced compared to SORCS2A and SORCS2B (Fig. 2, C and E). To study how SorCS2 splicing varies among individuals, we performed RT-PCR on RNA extracted from blood samples of eight men and 16 women. Similar to neuronal tissues, the relative expression of SorCS2 short (SorCS2 A+C) prevails in both men and women (Fig. 2, F and G). To study SorCS2 splicing variation in CNS tissues, we studied RNA sequencing data from The Sequence Read Archive. These data do not allow for statistical analysis; however, it is clear that the relative expression of splice variants in amygdala and prefrontal cortex is similar to that observed in other CNS tissues (Fig. 2H). Surprisingly, there was a complete lack of SORCS2A expression in the dorsolateral prefrontal cortex.

Cellular stress affects SorCS2 splicing

The differential SorCS2 splicing in tissues and during development prompted us to address whether SorCS2 splicing is also regulated by different cellular conditions. We subjected human embryonic kidney (HEK293) cells to various stress conditions and assessed changes in SorCS2 splicing pattern by RT-PCR. UV radiation of cells did not affect the ratio between SorCS2 long and short. Heat shocking of cells at 42 °C for 2 h increased although not significantly the relative amount of SorCS2 long relative to SorCS2 short (Fig. 3, A and B). However, the splicing pattern was significantly changed when subjecting cells to hypoosmotic medium for 24 h, leading to approximately 2-fold increase in SORCS2B and SORCS2D transcripts (Fig. 3C). Similar findings were found for the liver cell line HepG2 and the neuronal cell line SH-SY5Y when subjected to hypoosmotic shock (Fig. 3, D–I). Strikingly, although untreated HEK293, HepG2, and SH-SY5Y cells possess different SORCS2 mRNA isoform expression profiles under normal culture conditions, hypoosmotic stress resulted in a strikingly similar expression profile of SORCS2 isoforms in all three cell types. Interestingly, hypoosmotic stress had no effect on Sorcs2 splicing in mouse hippocampal neurons. However, there was a clear increase in total Sorcs2 transcripts as well as translated product (Fig. 3, J–L).

To study how pathological conditions might affect SORCS2 expression, we compared CNS levels of total SORCS2 transcripts in patients suffering from autism or schizophrenia with healthy control groups using data from The Sequence Read Archive. In autism patients, mean SORCS2 expression was non-significantly increased by ∼60% in prefrontal cortex. Interestingly, SORCS2 expression levels spanned a much greater range for autism patients than the control group (p = 0.00011) (Fig. 3M). Likewise, the variation was markedly increased in amygdala in patients suffering from schizophrenia compared to healthy individuals (p = 0.047), suggesting that SORCS2 brain expression could be perturbed in a general manner in patients suffering from psychiatric disorders (Fig. 3N).

Alternative splicing and posttranslational processing generate eight SorCS2 isoforms

SORCS2 mRNA is translated into a proprotein containing a short propeptide, which is proteolytically processed to generate a single-chain receptor, which may eventually be processed into two-chain SorCS2 (9). So far, all biochemical and functional studies using transfected cells, including receptor-processing experiments, have been performed using cDNA encoding the SorCS2A isoform (9, 10, 11, 20). To gain insight into the processing of the additional SorCS2 splice variants, we cloned SorCS2B, SorCS2BC, and SorCS2BD and followed receptor maturation by ³⁵S-pulse chase analysis in transfected HEK293 cells. All four receptor variants were processed from their proform into single- and two-chain receptors in a similar manner (Fig. 4A). We also assessed steady state distribution of pro-, one-, and two-chain receptors for each of the splice variants by Western blotting using HEK293 cells. All receptor-processing variants are present at similar levels with one-chain receptor being the most abundant species, followed by two-chain SorCS2 (Fig. 4B). Thus, it can be predicted that posttranslational proteolytic processing gives rise to a total of eight SorCS2 isoforms (Fig. 4C).

The presence of Ser1104 inhibits SorCS2 neurotrophic activity

The SorCS2 ICD is required for BDNF-induced branching of hippocampal neurons (11). We therefore assessed if alternative splicing of SorCS2 affects this process. Thus, primary cultures of SorCS2 KO mouse hippocampal neurons were transfected with cDNA encoding the individual SorCS2 splice variants and cultured in the absence or presence of 1 nM BDNF for 72 h. As previously reported, transfection with a SorCS2 truncation variant lacking the entire ICD (SorCS2 tailless) failed to rescue BDNF-induced branching of SorCS2 KO neurons (11). In contrast, transfection with SorCS2A and SorCS2B significantly rescued BDNF-induced branching (Fig. 5B). However, no significant rescue was observed with SorCS2C and SorCS2D, suggesting that insertion of a serine residue at position 1104 in SorCS2 ICD compromises BDNF-signaling activity.

SorCS2 alternative splicing affects receptor trafficking and interactions with sorting APs

Intracellular sorting of SorCS2 is governed by its ICD (9), and we therefore assessed the subcellular localization of each of the splice isoforms by immunostaining of transfected HEK293 cells. We found that all four splice isoforms were localized in vesicle-like structures and at the cell surface. However, the staining pattern of SorCS2A and SorCS2C lacking the acidic cluster motif was more paranuclear localized compared to SorCS2B and SorCS2D (Fig. 6A). We therefore monitored trafficking of SorCS2 isoforms by labeling surface-localized SorCS2 on ice with an antibody directed against the extracellular domain, followed by incubation at 37 °C for defined time intervals, whereafter cells were fixed and stained with secondary antibodies. At 15 min, all four receptor variants were mainly found in smaller vesicles. However, at 30 min, SorCS2A and SorCS2C mainly localized to paranuclear compartments in contrast to SorCS2B and SorCS2D, which were still observed more distant from the nucleus (Fig. 6B). Internalized SorCS2A and SorCS2C showed partial overlap with the trans-Golgi marker TGN46, suggesting retrograde transport to the trans-Golgi network (Fig. 6, C and D).

The SorCS2 ICD contains a tyrosine-based internalization and sorting motif YAQM conforming to the consensus sequence YXXØ (where X is any amino acid and Ø is a hydrophobic, bulky residue) (Fig. 1D). Clathrin AP complexes are known to bind such motifs in other receptors, and thereby mediate their sorting between the cell surface, endosomes, and TGN (21, 22). Sorting of the Vps10p domain receptor sortilin relies on AP-1 (23) whereas sorting of SorLA depends on both AP-1 and AP-2 (24), but so far, the interaction between SorCS2 and APs has not been studied. We therefore generated fusion proteins between SorCS2 ICD and glutathione-S-transferase (GST) and assessed pull down of endogenous AP complexes from HEK293 cell lysate. We used the ICD of sortilin fused to GST as positive control and confirmed the interaction with AP-1 (23) but also observed that sortilin binds AP-2 and AP-3, which has not previously been described (Fig. 7A). Interestingly, SorCS2A ICD conferred the most robust pull down of AP-1, AP-2, and AP-3, followed by SorCS2C ICD. SorCS2B and SorCS2D pulled down AP-1 and AP-2 to a lower extent but showed no interaction with AP-3, suggesting that the presence of the acidic cluster in SorCS2B and SorCS2D potentially obstructs the interaction with AP complexes. The adaptin ε subunit from AP-4 was not pulled down by any of the SorCS2 ICD fusion proteins. The interaction between SorCS2 ICD and AP-2 was confirmed using human brain homogenate (Fig. 7B), however we were not able to detect human AP-1 and AP-3 in neither pull downs nor homogenates from human brain tissue by Western blotting.

SorCS2 alternative splicing regulates interactions with motor proteins kinesin and dynein

To search for additional sorting APs contributing to SorCS2 isoform specific trafficking, we performed genome-wide yeast two-hybrid screening at Hybrigenics using SorCS2D ICD as bait and screened a human adult brain prey library containing 10 million primary clones. Thirty one protein fragments representing a total of seven cytosolic proteins were found to interact with SorCS2D ICD. Each hit was scored on a scale from 1 to 6 based on local and global technical parameters, with six representing very high confidence in the interaction. Twelve fragments, all of which obtained a score of 6, covered kinesin KLC1 a subunit of the motor protein kinesin responsible for cargo transport along microtubules toward the plus end (anterograde transport). Eleven fragments, again all of which obtained the highest score, covered DYNLT3 a subunit of the motor protein dynein, which moves cargo along microtubules toward the minus end (retrograde transport) (Fig. 8A). We subsequently assessed KLC1 and DYNLT3 binding to all four SorCS2 isoforms by yeast two-hybrid assay using plasmids obtained from the global yeast-two hybrid screen. We confirmed an interaction between SorCS2D ICD and KLC1 but also SorCS2B ICD–bound KLC1. We also confirmed the interaction between SorCS2D ICD and DYNLT3 and in addition found that the remaining three SorCS2 ICD isoforms also interact with DYNLT3 (Fig. 8B). We corroborated these findings by pull-down experiments using GST-SorCS2 ICD constructs to pull down endogenous KLC1 from HEK293 cell lysate (Fig. 8C) and GST-DYNLT3 to pull down SorCS2 isoforms from transfected HEK293 cell lysate (Fig. 8D). In order to further explore the specificity of SorCS2 splice variants in terms of motor protein adaptor binding, we sought to define the epitope of SorCS2 ICD responsible for the interaction with KLC1 and DYNLT3. To this end, we performed yeast two-hybrid assay using C terminally truncated variants of SorCS2D ICD as bait. Removal of the most C-terminal 20 amino acids fully abrogated interactions with both KLC1 and DYNLT3, suggesting that amino acids within this sequence and/or within the most C-terminal part of the acidic cluster motif contribute to KLC1 and DYNLT3 binding (Fig. 8E). To fine map the binding epitope of SorCS2D, we performed alanine scanning of SorCS2D C terminus, including part of the acidic cluster. Pairwise alanine substitutions of Ser⁵¹-Gln⁵² and Thr⁵³-Leu⁵⁴ located within the acidic cluster and to some extent Gly⁵⁵-Asn⁵⁶ located outside the acidic cluster (numbered according to the cytosolic sequence of SorCS2D ICD) abrogated KLC1 binding (Fig. 8F). In contrast, only alanine substitutions of amino acids located outside the acidic cluster (Gly⁵⁵-Asn⁵⁶ and His⁵⁷-Ser⁵⁸) inhibited DYNLT3 binding. Hence, KLC1 and DYNLT3 bind to distinct yet partly overlapping epitopes with KLC1 binding to the acidic cluster motif.

Thus, the differential trafficking properties of the SorCS2 splice isoforms may in part be explained by isoform-dependent preference for KLC1 and DYNLT3.

To study the influence of DYNLT3 on SorCS2 trafficking, we mutated the DYNLT3-binding site of SorCS2A by substituting His⁵⁷ and Ser⁵⁸ with alanine (SorCS2A DYNLT3mut). We then monitored trafficking of DYNLT3mut and wt SorCS2A by antibody uptake studies in transfected HEK293 cells. After 30 min, SorCS2A DYNLT3mut was mainly found in vesicle-like structures evenly distributed throughout the cytoplasm compared to wt SorCS2A, which showed a more paranuclear staining pattern (Fig. 9A). Hence, colocalization between SorCS2A DYNLT3mut and TGN46 was significantly reduced compared to that for wt SorCS2A, consistent with decreased retrograde transport of SorCS2A DYNLT3mut. We next assessed the involvement of KLC1 in SorCS2 trafficking by mutating part of the KLC1-binding site of SorCS2B by substituting Ser⁵¹ and Gln⁵² with alanine (SorCS2B KLC1mut). Whereas wt SorCS2B was mainly found in vesicular structures distributed throughout the cytoplasm after 30 min, SorCS2B KLC1mut staining was more paranuclear (Fig. 9B). The results are consistent with reduced anterograde transport of SorCS2B KLC1mut, shifting its subcellular distribution from cytoplasmic to paranuclear. Hence, our findings suggest that alternative splicing regulates SorCS2 anterograde and retrograde transport through interactions with kinesin and dynein motor proteins.

Constructs

SorCS2A (GenBank OQ616758) was inserted into pcDNA3.1/zeo (Invitrogen). SorCS2 splice variants were cloned by RT-PCR amplification of cDNA from human fetal brain RNA (Clontech). C-terminal fragments encoding the C terminal of human SorCS2B (GenBank OQ616759), SorCS2C (GenBank OQ616760), and SorCS2D (GenBank OQ616761) were produced. Fragments were substituted with the corresponding sequence in SorCS2A cloned into pcDNA3.1/zeo using an inherent and XhoI restriction site.

Reverse transcription-polymerase chain reaction

RNA from cell lines were purified using the NucleoSpin gel and PCR clean-up kit from MACHEREY-NAGEL. RNA from human and mouse tissues were purchased from Clontech. Blood samples from human subjects were all part of the Danish PRISME study (47, 48) approved by the Danish Data Protection Agency (2006-41-7032) and the Danish Regional Ethics Committee (RRS 2006-1028 [2747-06]). All subjects were Caucasian of origin and gave written informed consent. The studies abide by the Declaration of Helsinki principles. PAXgene blood RNA tubes were used for collection and immediate stabilization of RNA (PreAnalytiX, Qiagen). RNA was extracted using standard procedures.

mRNA was converted to DNA and a fragment encoding the 3′ end of human SorCS2 including the cytosolic domain amplified with Titanium one-step RT-PCR kit (Clontech) using the primer pairs 5′-CAAGGAAGAGGAGCCTCTCGAGTGATA-3′ and 5′–GTGGTTCTGTGCCCTCCGTGGGTGAAA-3′ and mouse SorCS2 with 5′ – CAGAGGAGCTGCTTGTGACTGTGGTAA – 3′ and 5′ – CACGGGTCTCAGAGGGTGTGTCATT – 3′. Human GAPDH was amplified using the primer pairs 5′-ACCATGGAGAAGGCTGGGGCTCATTT–3′ and 5′–ATGGCATGGACTGTGGTCATGAGTCCTT-3′. Amplified transcripts were separated on a 1.5% agarose gel and quantification of bands performed with ImageJ software (https://imagej.net/ij/). Relevant bands were cut out and purified with RNA NucleoSpin kit (MACHEREY-NAGEL). Sequencing of human SorCS2 was performed with the following primers 5′–ATCAGCTCACCAGGTAGCTGTG–3′ or 5′-CAAGGAAGAGGAGCCTCTCGAGTGATA-3′ and mouse SorCS2 with 5′–CCATAGCCGACTTGTATGTGCTTC–3′ at Eurofins Genomics sequencing facility. Quantification of transcripts containing insertion of nucleotides encoding a serine residue was performed by relating the amplitude of three selected nucleotides from the sequencing profile of transcripts with and without the insertion.

Cell culturing and stress treatments

The HEK293 and hepatocellular carcinoma (HepG2) cell lines were cultured in Dulbecco's modified Eagle's medium (DMEM) (Lonza) and the neuroblastoma (SH-SY5Y) cell line in DMEM F12 (Lonza), all supplied with 10% fetal calf serum (FCS) (Gibco). To promote UV stress, cells were subjected to radiation from an Osram Puritec hns 15 W, G13 UV lamp at a distance of 50 cm for 15 min and harvested 24 h later. Mild heat stress was induced at 42 °C at 2 h prior to harvest. Hypoosmotic stress was induced by addition of 70% distilled water to the culture medium of the indicated cell lines. Cultures of mouse hippocampal neurons were treated in a similar manner at 6 days in vitro for 24 h with 5 to 30% distilled water added to the culture medium. Transient transfection was done with FuGENE 6 (Promega) using 3.2 μl/ml FuGENE and 0.8 μg/ml cDNA.

Pull-down experiments

For preparation of HEK293 cell lysate, confluent cells were lysed in TNE lysis buffer (10 mM Tris, pH 8, 150 nM NaCl, 1 mM EDTA) supplemented with 1% NP40 and complete protease inhibitor cocktail (Roche) and cleared at 2000g for 5 min. For preparation of human brain homogenate, tissue was homogenized with an ULTRA-TURRAX homogenizer in lysis buffer (20 mM Hepes pH 7.6, 125 mM KAc, 2.5 mM MgAc, 320 mM sucrose, 0.1 mM EDTA) supplemented with complete protease inhibitor cocktail and cleared at 1000g for 10 min. The homogenate was then incubated with 1.25% Triton X-100 for 1 h at 4 °C and subsequently cleared by two centrifugation steps, first at 20,000g for 15 min and then at 100,000g for 1 h.

Hundred micrograms of the cytosolic domain of SorCS2A, SorCS2B, SorCS2C, SorCS2D, sortilin, or DYNLT3 fused to GST was incubated with HEK293 cell lysate or brain homogenate overnight at 4 °C. GST-fusion proteins were then precipitated with glutathione-sepharose beads for 3 h at 4 °C and unspecific binding removed by washing five times for 5 min with lysis buffer. Proteins were eluted by boiling samples in reducing sample buffer (20 mM dithioerythritol, 2.5% SDS) and proteins separated on NuPAGE Bis-Tris 4 to 12% gradient gels from Invitrogen. Antibodies against adaptin γ (610385), μ2 (611350), adaptin δ (611328), and adaptin ε (612018) were all from BD transduction Laboratories. Anti-GST (27457701V) was from GE Healthcare, anti-mouse SorCS2 (AF4237) from R&D Systems, and anti-KLC1 (ab174273) from Abcam. Anti-extracellular SorCS2 F7100 antibody is custom made from DAKO. Images were acquired with the Fujifilm las4000 imaging apparatus using appropriate horseradish peroxidase-conjugated secondary antibodies from DAKO. Further details on the use and performance of antibodies were shared on http://www.pabmabs.com posted by S. Skeldal and S. Glerup.

Immunofluorescence

HEK293 cells–expressing SorCS2 variants were seeded onto poly-L lysine–coated coverslips 24 h prior to analysis. Surface-exposed SorCS2 was labeled with 1 μg/ml purified antiextracellular SorCS2 F7100 antibody (custom made from DAKO) for 2 h on ice, washed in ice-cold complete medium, and chased at 37 °C in complete medium. Cells were then fixed in 4% paraformaldehyde at indicated time points, permeabilized in 0.1% Triton X-100, and blocked in 10% FCS. Secondary antibody was Alexa-conjugated donkey anti-rabbit 488 (A21206) from Molecular probes, and imaging was performed on a Zeiss LSM780 confocal microscope. For costaining of TGN, cells were incubated with anti-TGN46 (1:100) after permeabilization and blocking and detected with Alexa-conjugated donkey anti-sheep 568 (A21099) from Molecular probes. Zeiss Zen software (https://www.zeiss.com/microscopy/en/products/software/zeiss-zen-lite.html) was used to analyze confocal images and quantify pixel-intensity weighted colocalization coefficients (49).

Y2H screen

A commercial Y2H screen was performed by Hybrigenics (http://www.hybrigenics.com) to identify APs targeting the ICD of SORCS2D. The ICD of SorCS2D^{K1101–S1174} was cloned into the Hybrigenics bait vector pB27 in fusion with LexA. An adult human brain random-primed cDNA library, transformed into the Y187 yeast strain was used for mating. To certify data from the primary screening, library and bait plasmids were purchased from Hybrigenics for additional verification using control plasmids without inserts. Successfully transformed cells were suspended in sterile ddH2O and spot-plated (5 μl/spot) on plates prepared with 46 g/l yeast minimal agar SD base (Clontech) and 100 μg/ml Pen/Strep (Gibco), and supplemented with either 640 mg/l Leu/Trp dropout supplement (Clontech) for control plates or 640 mg/l Leu/Trp/His dropout supplement (Clontech) for test plates. 3-aminotriazole (Sigma-Aldrich) was added to the plates to suppress bait autoactivation. SORCS2D dialanine mutations were created by overlapping PCR reactions. SorCS2A, SorCS2B, and SorCS2C cytoplasmic PCR fragments were inserted into pB27.

Pulse-chase experiments

HEK293 cells were seeded into poly-L lysine-coated wells and transiently transfected 24 h later. 24 h post transfection, medium was exchanged with cysteine and methionine free DMEM (Sigma) containing 2% dialyzed FCS and 1% Glutamax for 15 min. Cells were biolabeled for 4 h with ∼300 μCi/ml L-[³⁵S]-cysteine and L-[³⁵S]-methionine (Pro-mix, Amersham) in the same medium and also supplemented with 10 mg/ml brefeldin A. After a quick wash, cells were chased in complete medium and subsequently lysed at indicated time points in TNE lysis buffer (10 mM Tris, pH 8, 150 nM NaCl, 1 mM EDTA) supplemented with 1% NP40 and complete protease inhibitor cocktail. Samples were immunoprecipitated overnight at 4 °C by use of Gammabind G-Sepharose beads (Amersham) coupled with anti-SorCS2 F7100 antibody. Immunoprecipitates were separated on NuPage Bis-Tris 4 to 12% gradient gels from Invitrogen and processed for fluorography using 2,5-diphenyloxazole.

Neuronal branching assay

SorCS2^−/− postnatal day 0 mouse pups of either sex (9) were euthanized by decapitation, brains removed, and hippocampi dissected into ice-cold Leibowitz’s L-15 medium (Life technologies). Hippocampi were dissociated for 30 min using 20 U/ml pre-activated papain (Bionorica) and washed in DMEM containing 0.01 mg/ml DNase 1 (Sigma) and 10% FCS before being triturated in the same buffer. Following trituration, DMEM was substituted with Neurobasal-A medium supplemented with B-27 supplement, 2 mM GlutaMAX (all from Gibco), 100 μg/ml primocin (Invivogen), 20 μM floxuridine, and 20 μM uridine (both from Sigma). Cells were seeded onto precoated coverslips (poly-D-lysine and laminin [InvivoGen]) at a density of 100,000/coverslip and incubated at 37 °C for 24 h. Medium was then substituted with Neurobasal-A medium without antibiotics and transfected using lipofectamine (InvivoGen). For each cotransfection, 125 ng of plasmid encoding SorCS2 and 125 ng plasmid encoding GFP were added to 100 μl Opti-MEM reduced serum medium (Gibco) and 0.25 μl plus reagent added to the diluted DNA and incubated at room temperature for 5 min. Hereafter, 1.25 μl lipofectamine LTX was added, and the solution was incubated at room temperature for 30 min. The DNA–lipid complex solution was then added to neurons and medium substituted with Neurobasal-Amedium with or without 1 nM BDNF (Millipore) after 6 h at 37 °C.

After 72 h, neurons were fixed with 4% paraformaldehyde, and images were obtained using confocal microscopy. Branching pattern of GFP-positive neurons was analyzed using Zen 2011 image processing (https://www.zeiss.com/microscopy/en/products/software/zeiss-zen-lite.html). At least ten neurons were analyzed on three or four coverslips for each condition. The use of animals in this study has been approved by the Danish Animal Experiments Expectorate (License file number 2016-15-0202-00050).

RNA sequencing analysis

Data files were downloaded from Sequence Read Archive using SRAToolkit. For transcript quantification, Kallisto software (https://pachterlab.github.io/kallisto/about.html) was used to normalize to the human genome hg38. For differential expression analysis of SORCS2 in autism and schizophrenia patients versus controls the R package DESeq2 (http://www.bioconductor.org/packages/release/bioc/html/DESeq2.html) was used to normalize Kallisto output.

Data analysis and statistics

Statistical analyses were performed using unpaired two-tailed Student’s t test unless otherwise stated. Variances in SorCS2 mRNA levels in patients suffering from autism or schizophrenia and control groups were compared using an F-test. BDNF treatment responses in different groups of hippocampal neurons were compared by two-way ANOVA. If the analysis of variance revealed significant differences among groups, a post hoc test (unpaired two-tailed Student’s t test) was carried out to elucidate the pattern of differences.

Values are presented as mean ± SD and the significance level was set at p < 0.05.