Srsf1 and Elavl1 act antagonistically on neuronal fate choice in the developing neocortex by controlling TrkC receptor isoform expression

Abstract The seat of higher-order cognitive abilities in mammals, the neocortex, is a complex structure, organized in several layers. The different subtypes of principal neurons are distributed in precise ratios and at specific positions in these layers and are generated by the same neural progenitor cells (NPCs), steered by a spatially and temporally specified combination of molecular cues that are incompletely understood. Recently, we discovered that an alternatively spliced isoform of the TrkC receptor lacking the kinase domain, TrkC-T1, is a determinant of the corticofugal projection neuron (CFuPN) fate. Here, we show that the finely tuned balance between TrkC-T1 and the better known, kinase domain-containing isoform, TrkC-TK+, is cell type-specific in the developing cortex and established through the antagonistic actions of two RNA-binding proteins, Srsf1 and Elavl1. Moreover, our data show that Srsf1 promotes the CFuPN fate and Elavl1 promotes the callosal projection neuron (CPN) fate in vivo via regulating the distinct ratios of TrkC-T1 to TrkC-TK+. Taken together, we connect spatio-temporal expression of Srsf1 and Elavl1 in the developing neocortex with the regulation of TrkC alternative splicing and transcript stability and neuronal fate choice, thus adding to the mechanistic and functional understanding of alternative splicing in vivo.


INTRODUCTION
The plethora of pr ojection neur on subtypes in the cerebral neocortex is generated during embryonic de v elopment by a transient pool of neural progenitor cells (NPCs) (reviewed in ( 1-3 )).While the ultimately six-layered neocortex is one of the defining features of mammals, the numbers and ratios between subtypes are species-specific, and an increased complexity of this organization is generally accepted to hav e enab led heightened cogniti v e function in primates (4)(5)(6).Abnormalities in the positioning, morphology Nucleic Acids Research, 2023, Vol. 51, No. 19 10219 or numbers of cortical neuron subtypes often result in dev elopmental neuropsy chiatric disor ders that impair cogniti v e, sensory and motor functions (7)(8)(9).Ther efor e, unravelling how NPCs generate the correct numbers of the different pr ojection neur on subtypes is pivotal for understanding the de v elopment of both the healthy and the diseased neocortex.
The dif ferentia ti v e behavior, and hence fate choices, of cortical NPCs is governed by their cellular identity, which, in turn, is governed by a precise transcriptomic composition.We currently still only have a partial understanding of the factors determining the specific transcriptome in NPCs and its causal relationships to what neuron subtypes these cells produce.In general, RNA-binding proteins (RBPs) are known to considerably impact the transcriptome ( 10 , 11 ) as they regulate all steps of RNA processing.This e v entually determines the abundance and composition of mature transcripts, and some RBPs are known to exhibit strikingly diverse and dynamic spatio-temporal expression profiles in the de v eloping corte x (12)(13)(14).
Among the many processes that RBPs control, alternati v e splicing and dif ferential stabiliza tion of (pre-m)RNAs are highly prevalent in the brain, and the protein di v ersity that these processes enables is thought to have contributed to the evolutionary expansion of cortical complexity (15)(16)(17)(18)(19)(20)(21)(22)(23).Furthermore, RBPs regulating alternati v e splicing (termed splicing factors in the following, in short, SFs) have been implicated in several steps of cortical projection neurogenesis, such as NPC maintenance, neurogenic division, and neuron migration and morphology acquisition (re vie wed in ( 24 ), and see ( 25 )).Howe v er, to date, we have very limited knowledge on how SFs contribute to neuronal subtype fate acquisition, despite some SFs being expressed in pa tterns tha t are specific to NPCs or during key stages in the production of distinct neuron subtypes (12)(13)(14).
We previously showed that the NPC levels of an alternati v ely spliced isoform of the neur otr ophin-3 receptor TrkC determine the acquisition of corticofugal projection neuron fa te (CFuPN) a t the expense of callosal pr ojection neur on fate (CPN) ( 26 ).This control is crucial due to the striking functional differences of the two neuron subtypes in the cortical circuitry, with CFuPNs projecting outside of the cerebral cortex and CPNs within.Based on this finding, we hypothesized that CFuPN-CPN fate determination through TrkC-T1 is the result of dynamic RBP-controlled mRNA isoform expression in the developing cortex.Here, we show that the CFuPN-CPN fate choice, as dictated by the le v els of TrkC isoforms, is orchestrated by two key SFs, Srsf1 and Elavl1.We first report that TrkC-T1 and TrkC-TK+ levels ar e finely r egulated in NPC-and neuron-specific ratios in the de v eloping corte x.Through in vitro and in viv o dosa ge manipulations, we uncover that the RBPs Srsf1 and Elavl1 act as antagonistic regulators of TrkC mRNAs, ensuring the NPC-and neuron-specific balance between the two receptor transcript and protein variants, and thereby the generation of the correct proportion of CFuPNs to CPNs.Additionally, we show that the combination of Srsf1 and Elavl1 le v els defines distinct, cell type-specific environments for regulating the le v els of Tr kC isoforms.Finally, we identify an exonic splicing enhancer in the T1-specific exon 13A as an Srsf1-dependent cis-regulatory element.This is the first demonstration of Srsf1 and Elavl1 co-regulating binary neuronal fate decisions in the de v eloping mammalian cortex and provides an in vivo example to underline the importance of RBPs in neurogenesis.

Immortalized cell line culture and treatment
N2A and HEK293T cells were, unless otherwise specified, cultured in DMEM (4.5 g glucose / l, supplied with Gluta-MAX L-glutamine, Gibco, 10566) supplemented with 10% fetal bovine serum (Biochrom) and penicillin-streptomycin (1000 U / ml, Gibco, 15140122).For transfection, we used the TurboFect reagent (Thermo Scientific, R0533), according to the instructions of the producer.In the respecti v e e xperiments, actinomy cin D was applied for the indicated length of time (final concentration: 20 g / ml, Sigma-Aldrich, A9415).

Primary neuron pr epar ation and nucleofection
Primary cortical neurons wer e cultur ed in dishes that were coated overnight at room temperature with pol y-L-l ysine (final concentration 10 g / ml, Sigma-Aldrich, A-005-M) and laminin (final concentration 0.2 g / ml, Sigma-Aldrich, L2020).For harvesting the neurons, pregnant dams were sacrificed at the indicated days of embryonic de v elopment, the embryos released from the uterine horns, and their cortical hemispheres excised while removing the hippocampal anlage and detaching from the ganglionic eminences.After rinsing in HBSS -(Gibco, 14175095), the cortex pieces were dissociated using a 0.3125% trypsin solution in HBSS-(Gibco, 15090046) for 15 min at 37 • C and then treated with DNase I (final concentration of 0.05 mg / ml, Roche, 10104159001) for 2 min at room temperature.The cortical cells were resuspended in embryonic neuron culture medium (1 ml 50 × B27 supplement without vitamin A, Gibco, 12587010, 500 l GlutaMAX supplement, 500 l of penicillin-streptomycin stock, added to 48 ml Neurobasal medium, Gibco, 12348017).

Cell and tissue staining procedures
For chromogenic RNA in situ hybridization, embryonic brain slices were incubated with digoxigenin (DIG)-labeled probes.The primers used to generate the RNA probes contained SP6 RN A pol ymerase promoters and had the sequences listed in the table below.For the transcription, 1 g linearized plasmid were mixed with 2 l 10 × transcription buffer, 2 l of 100 mM DTT, 0.5 l RNase inhibitor, 2 l of 10 × DIG labeling mix (Roche, 11277073910), 20 U SP6 RN A pol ymerase (Thermo Scientific, EP0131), to 20 l with RNase-free MilliQ water.All containers and solutions used prior to and during the RNA probe hybridization were treated against RNase contamination by heating them at 200 • C for two hours.For the prehybridization, 1 ml of hybridization solution (50% deionised formamide p.a, 0.1 mg / ml yeast tRNA, 10% dextran sulphate, 1:50 dilution of Denhardt's solution, Thermo Fisher, 750018 and a 1:10 dilution of a salt solution containing 2 M NaCl, 50 mM EDTA, 100 mM Tris-HCl pH 7.5, 50 mM NaH2PO4 •2H2O, 50 mM Na2HPO4) was applied per slide and the slides incubated at 65 • C for 1 h.200 l of the probe mixture (50 l probe solution denatured in 100 l formamide for 5 min at 95 • C) were applied per slide and slides were incubated as above overnight.Unbound probe was r emoved by thr ee washes in a stringent washing solution (50% formamide, 1 × SSC, 0.1%, Tween-20), after which the slides were washed with MABT buffer (100 mM maleic acid, 150 mM NaCl, 0.1% Tween-20, pH 7.5), then blocked for 1 h at room temperature in a 2% blocking reagent solution (Roche, 11 096 176 001) with 10% sheep serum in 1x MABT.A 1:1500 solution of alkaline phosphatase-coupled anti-DIG antibodies (Roche, 11093274910) in MABT was applied overnight at 4 • C. Unbound antibodies were washed of f a t room tempera ture in 1x MABT buf fer, then in prestaining solution (4 ml of 5M NaCl, 10 ml of 1 M MgCl 2 , 20 ml of 1 M Tris pH 9.5, 0.2 ml of Tween 20 in 166 ml of MilliQ water).Slides were then incubated at 37 • C in staining solution with chromogenic AP substrate until the color ed pr ecipitate could be observed.Staining solution: 0.8 ml of 5 M NaCl, 2 ml of 1 M MgCl 2 , 4 ml of 1 M Tris pH 9.5, 13.2 ml H 2 O, 40 l Tween-20, 40 l of NBT (1000 × = 100 mg / ml in 70% DMSO), 40 l of BCIP (1000 × = 50 mg / ml in 100% DMSO), supplemented up to 40 ml with 10% PVA in H 2 O. Coverslips were mounted with Entellan (Sigma-Aldrich, 107960).
For immunofluorescent staining, 50 m brain sections were blocked for 30 min in blocking solution (10% horse serum, 0.1% Triton X-100 in PBS), then incubated overnight with primary antibody in blocking solution with gentle shaking.On the second day, sections were washed 4 × 10 min in an excess of PBS, then fluorophore-coupled secondary antibodies were applied for 4 h at room temperatur e.Sections wer e then mounted onto SuperFrost Plus glass slides with ImmuMount (Thermo Scientific, 9990402).

Image acquisition, processing and quantification
The slides resulting from in situ hybridization were imaged on a Zeiss BX60 system.Linear modifications of brightness were performed using ImageJ software.For the immunofluorescently stained tissue preparations, we used a Leica Sp8 confocal laser scanning system with a DMI6000CSB microscope (BioSupraMol facility at Freie Uni v ersit ät Berlin).When analyzing fate acquisition, we marked around 100-300 GFP-positi v e cells per analyzed electroporation site for Ctip2 and Satb2 or Cux1 co-expression and counted each dual labeling using the Cell Counter plugin in ImageJ.Counting was performed blinded.For each electroporated litter, brain sections were matched for anteroposterior and lateromedial position of the electroporation site.To quantify the fold change in fate, individual brains were compared to the mean percentage of doub le positi v e cells of tha t fa te in the littermate controls.

Fluor escence-activ ated cell sorting (FACS)
Primary cortical cells, pr epar ed as described above, were resuspended in PBS, stained with the APC-coupled antiprominin-1 or isotype control antibody plus propidium iodide in PBS on ice for 30 min, and then sorted for PI and APC signal using a BD FACSCanto or FACSMelody sorter.PI-negati v e cells were collected in two separate tubes, depending on the presence or absence of APC signal.The collection medium was based on our Neurobasal culture medium and supplemented with recombinant murine EGF (final concentration: 40 ng / ml, ImmunoTools, 12343406) and FGF2 (final concentration: 40 ng / ml, ImmunoTools, 12343623).Cells were then pelleted by centrifugation and the pellets snap-frozen in liquid nitrogen for downstream applications.

Quantitative real-time PCR
For tissues, RNA was extracted using the ReliaPrep RNA extraction kit (Promega, Z6212) and re v erse transcribed into first strand cDNA with an oligo(dT) primer (Promega, C1101) and the Promega GoScript re v erse transcription system (A5000).TaqMan RT-qPCR was performed using the FastAdvanced Master Mix (Thermo Fisher, 4444557) on a StepOne Plus RT-qPCR cycler (Thermo Fisher / Applied Biosystems, 4376600).Reactions were set up according to the master mix protocol using the equivalent of 25 ng re v erse transcribed RNA per 10 l reaction.Reactions were performed in technical quadruplicates and the number of biological replicates indicated in the figures.The following TaqMan probes were used: for TrkC-T1, VIC-tagged Mm01317842 m1 and for TrkC-TK+, a custom-designed exon junction spanning FAMtagged probe (AR47VWU), both from Thermo Fisher.
For cultured cells, the RNA was extracted using a standard phenol-chloroform extraction procedure employing the TRIzol reagent (Ambion / Invitrogen, 15596018) and followed by a DNase treatment (Lucigen, D9905K) and a phenol-chloroform-isoamyl alcohol extraction (Carl Roth, X985.1).The resulting RNA was re v erse-transcribed using MMuLV re v erse transcriptase (Enzymatics / Qiagen, P7040L) using either oligo(dT) primers (RT-qPCR) or gene-specific re v erse primers (splicing-sensiti v e RT-PCR), according to the producer's protocol.For SYBR Green RT-qPCR, we used the Promega GoTaq RT-qPCR system (A6001), according to the producer's protocol.Amplifica tion ef ficiency was calcula ted using the Thermo Fisher qPCR efficiency calculator.

Radioacti ve splicing-sensiti ve PCR
Splicing-sensiti v e PCRs were performed with transcript variant-discriminating primers either with radioacti v e labelling of primers or without.Radioacti v e RT-PCRs were performed as described ( 27 ).Briefly, 200 ng of the primer binding to both transcript variants were labelled with 32P--ATP (Hartmann Analytic, SRP-501) using 10 units T4 PNK (Molox) for 1h at 37 • C, and then purified and precipitated using the PCI protocol as described above.Primer pellets were resuspended in 80 l H 2 O, and 1 l of this labelled primer was used per 20 l PCR reaction.After the PCR, products wer e mix ed 1:1 with formamide loading buf fer, dena tured alongside the marker (NEB, N3032S) for 5 min at 95 • C, and 5 l were applied to a denaturing polyacrylamide-urea gel (7 M urea, 8% polyacrylamide in 0.5 × TBE).Once the desired degree of resolution was r eached, gels wer e fix ed, transferr ed to W ha tman paper, vacuum-dried and finally assembled with a photostimulatable phosphor plate in photo gra phic cassettes.The plates were then imaged on a GE Healthcare Typhoon 7000 FLA Phosphorimager and the result quantified using the Image-Quant TL software, version 8.1.

UV cr osslinking of radioactiv e RNA pr obes to nuclear extract proteins
T25 flaks with N2a cells at 80% from confluency were transfected with the empty vector (pCAGIG) or Srsf1 ov ere xpression vector (pCAG-Srsf1).Transfection efficiency was confirmed by epifluorescence microscopy.Of these cells, nuclear extracts were prepared by nuclear-cytosolic fractionation in RNase-free buf fers.W hile on ice, cells were washed twice with ice-cold PBS, then gently resuspended in a volume equal to fiv e times the packed cell volume of the low salt CTX buffer (10 mM HEPES, 1.5 mM MgCl 2 , 10 mM KCl).
After fiv e min, the same volume of CTX buffer with 0.2% NP-40 was added and the suspension was gently pipetted up and down, then left on ice for another fiv e min.Nuclei were pelleted by centrifugation at 6500 rpm for three min (4

Bioinformatic analysis of RNA sequencing data
For the analysis of single cell datasets, E 14, E 16 and E 18 snRNAseq raw data provided by ( 29 ) (GSE153164) was aligned to GRCm38 using cellranger 7.0 with '-includeintrons'.Count matrices were then imported into Seurat v4 and quality filtered to remove cells containing < 5% mitochondrial transcripts and nFeature RNA < 1000 and nFea ture RNA < 5000.Da tasets were integra ted using SCT, followed by RunPCA(npcs = 30), RunUMAP(), FindNeighbors(dims 1:20, k.param = 10) and FindClusters(algorithm = 1, resolution 0.3).Markers were then found using FindAllMarkers, and manually annotated according to known biology.Cells belonging to the pyramidal lineage were then subset into a new Seurat object for plotting.For plotting expression of Srsf1 and Elavl1 in progenitor cells, cells were subset if they had > 0 expression of either Pax6 or Tbr2 .Raw Srsf1 and Elav1 values were then fit to a negati v e binomial with 'celltype' ( Tbr2-positi v e, Pax6positi v e, Tbr2 / Pax6 -positi v e) and 'stage' (E 14, E 16, E 18) as interaction terms using the MASS package in R.
Coefficients from this model can be seen in the table in Supplementary Figure 6D.For stage-specific differences in Srsf1 and Elavl1 expression distributions, we used a nonparametric Wilco x on signed-rank (paired) test to assess whether their population mean ranks differ by cell type.N and p values from the Wilco x on signed rank test can be found summarized per stage in the table in Supplementary Figure 6E and in detail for each cell type and stage in Supplementary Figure 6F.P value adjustment for multiple testing was performed with the Benjamin-Hochberg method.

Statistical analysis
Statistical analysis for laboratory experiments was performed using the Prism software (GraphPad), in accordance with the nature of the experimental setup and employing the tests indicated in the experiments.First, fulfilment of the assumptions r equir ed for statistical testing was verified by interrogating whether samples were taken from normally distributed and equal-variance populations with the Shapiro-Wilk and F test, respecti v ely.Based on the resulting information and the type of experimental setup, the statistical test was chosen with, if needed, the appropriate corr ections.P air ed tests wer e chosen when comparing gene expression between control and experimental animals from the different litters, because cortical differentiation is highly dynamic and small differences in the exact developmental time point can affect overall gene expression.ANOVA post-hoc tests were chosen based on whether samples were compared pairwise or all with a control sample.A detailed description of the statistical test decisional tree employed by GraphPad Prism can be found in the software's documenta tion ( https://www.graphpad.com/guides/prism/latest/ statistics/index.htm).

T wo alternativ e isoforms of TrkC, TrkC-T1 and TrkC-TK+, ar e expr essed in a stage-and cell type-specific manner in the developing cortex
In Parthasarathy et al. ( 26 ), we characterized two TrkC splice variants that result in the isoforms T1 and TK+ (Figure 1 A), and how the finely tuned le v els of Tr kC-T1 steer the CFuPN-CPN fate choice.Based on these findings, we hypothesized that the quantity of each receptor variant and, thereby, their ratio, is precisely regulated during cortex development.To test this, we assessed the TrkC-T1 to TrkC-TK+ transcript ratio by multiplex TaqMan RT-qPCR in cortical tissue across key stages of corticogenesis (Figure 1 B).We found that, as the corte x de v elops, the ratio of T1 and TK+ mRNA in bulk cortical tissue gradually shifts in favor of the latter variant, with T1 decreasing from a proportion of around 35% in the total TrkC transcript quantity at embry onic da y (E) 11.5 to around 15% at E 18.5 (Figure 1 B).This is reminiscent of our previous observations on the protein le v el ( 26 ), and suggests that the regulation of the TrkC protein isoform ratio is controlled through differential mRNA isoform expression, caused, for example, by either different stabilities of the transcript isoforms through their different 3 UTRs or due to alternati v e splicing.
Based on our observation that the T1 to TK+ ratio gradually changes during cortex development, we wondered whether this effect could be due to ratios specific to the various cortical cell types, whose numbers change in de v elopment and thereby may contribute to the shifting ratios observed in the cortical tissue in bulk.At early de v elopmental stages, the cortex consists of a multitude of NPCs and then, as these NPCs divide asymmetrically, becomes gradually enriched in neurons ( 30 ).We ther efor e investigated whether cortical NPCs and neurons exhibit specific ratios of the two TrkC transcript variants.To this end, we sorted primary cortical cells at E 12.5 by FACS using prominin-1 as a marker for apical radial glial cells (aRGCs), a type of cortical NPC.Gi v en tha t, a t E 12.5, the cortex consists primarily of aRGCs and the neurons they produce by direct neurogenic divisions ( 30 , 31 ), this allowed our sorting paradigm to discriminate between NPCs and neurons (Figure 1 C-E, Supplementary Figure 1A).We observed a significant difference in the ratio of T1 to TK+ between prominin-1positi v e (aRGs) and -negati v e cells (neurons) (Figure 1 E).In neurons, the percentage of T1 of total TrkC transcripts is around 15%, less than half of the percentage seen in NPCs.This indicates a cell type-specific balance between the two transcript variants.Since neurons exhibit a T1 to TK+ ratio that is strongly shifted in favor of TK+, their gradually rising numbers may explain the developmentally increasing dominance of this isoform in the cortex at large.Taken together, we find that the ratio of the TrkC transcript variants is maintained at specific le v els in cortical NPCs and neurons.

Srsf1 and Elavl1 regulate TrkC alternative splicing antagonistically
To address the mechanistic basis for differential TrkC isoform expression, we first considered a potential involvement of micro RNAs (miRNAs) in establishing the levels of T rkC-T1 and T rkC-TK+.How ever, when w e interrogated the 3 UTR sequences of T1 and TK+ with the miRNA binding site prediction tool TargetScan ( 32 , 33 ) (Supplementary Figure 1B), we could not identify any miRNAs whose expression patterns in the de v eloping corte x are in line with the observed patterns of T1 and TK+ expression ( 26 ).
Based on these findings, we hypothesized that the de v elopmental regulation of the TrkC tr anscript r atio in the neocortex likely occurs at the le v el of alternati v e splicing.Many RNA-binding proteins (RBPs) have been shown before to be crucial for corticogenesis ( 25 , 34-36 ), fulfilling manifold roles and exhibiting variable expression patterns (12).We ther efor e wonder ed whether RBPs could contribute to the cortical regulation of Tr kC alternati v e splicing.To identify such RBPs, we searched for those de v elopmentally dynamic SFs which are also predicted to bind in the region of the TrkC pre-mRNA that is relevant to the alternative splicing outcome (Figure 2 ).We found 32 splicing factors (SFs) that fulfilled these r equir ements by querying both putati v e RBP binding sites in the TrkC pre-mRNA, as predicted by four online tools (CISBP ( 37 ), RBPDB ( 38 ), ATtRACT ( 39 ) and RBPmap ( 40)), and SFs found to be dynamically transcribed in the de v eloping corte x ( 12 , 13 ) (Figure 2 A).
To test whether any of these SFs regulate TrkC transcript variant balance (Figure 2 B-E), we examined the impact of individual factors by knocking them down with an established siRNA library in N2A cells ( 41 , 42 ), and then assessing the resulting ratio of T1 to TK+ using a radioacti v e, splice variant-discriminating RT-PCR (Figure 2 B and C).Of the 32 tested SFs, Srsf1 and Elavl1, which are strongly expressed in wild type N2A cells (Supplementary Figure 2A), had, by far, the largest impact on Tr kC alternati v e splicing.The Elavl1 knockdown increased the proportion of T1 in the total TrkC transcript pool by over 70% of its levels in the control sample and the knockdown of Srsf1 decreased it by around 50% (Figure 2 D, E and Supplementary Figure 2A and B).
The ELAV-like family of RNA binding proteins is comprised of four r epr esentati v es in mice and humans.Previous r esear ch has shown that Ela vl2, Ela vl3, and Elavl4 share a number of properties, especially in terms of functional redundancy in the nervous system ( 43 ).To test whether this particular subgroup of RBPs has an impact on TrkC alternati v e splicing, we knocked down Elavl2 in N2a cells and assessed the effect on TrkC AS by splicing-sensitive RT-PCR (Supplementary Figure 2C).The knockdown did not significantly change Tr kC-T1 le v els in the total TrkC-T1 + TrkC-TK+ transcript pool, as opposed to the knockdown of Elavl1, which, as seen in previous experiments, increased Tr kC-T1 le v els compared to the control siRNA sample.
To investigate a potential cross-regulation of Srsf1 and Elavl1, we altered the protein le v els of each of the factors by ov ere xpressing or knocking them down in N2a cells.We then monitored the resulting Srsf1 protein le v els by Western blotting (Supplementary Figure 3A) and the Elavl1 protein le v els by immunofluorescence (Supplementary Figure 3B).We observed no significant changes of neither Srsf1 nor Elavl1 under these conditions, except for the ones caused, for each of the proteins, by the ov ere xpression plasmid encoding it and the knockdown construct directed against it (Supplementary Figure 3A-C).This indica tes tha t Srsf1 and Elavl1 operate independently from one another to regulate TrkC AS and, potentially, other splicing e v ents.
In order to test if Elavl1 and Srsf1 control TrkC alternati v e splicing in cortical cells, we isolated primary cortical cells from E 13.5 embryos, ov ere xpressed either Srsf1 or Elavl1 in these cells, cultured them for two days in vitro (DIV), and then quantified the TrkC isoform ratios on the transcript and protein le v el (Figure 2 F-H).As before, we saw an increase in the proportion of TrkC-T1 when overexpressing Srsf1.Conversely, the proportion of T1 decreased when ov ere xpressing Elavl1.We could observe these effects both via RT-qPCR (Figure 2 G) and Western blotting (Figure 2 H).This showed that, indeed, Srsf1 and Elavl1 influence Tr kC alternati v e splicing in primary cortical cells as seen in N2A cells.

The TrkC-T1-specific e x on 13A harbors a splicing enhancer regulated by Srsf1
We next wished to understand which sequence elements contribute to the regulation of TrkC AS.The cassette exons of interest, 13A and 14A (Figure 1 A), are flanked by large introns (see Figure 3 A, 51.6 kb for intron 13 between exons 12 and 13A, 1.5 kb for intron 14 between exons 13A and 14A, and 40 kb between exons 14A and 13).Therefore, in order to examine alternative splicing regulation, we first predicted the splice site strength of the cassette exons and the flanking constituti v e e xons (termed 12 and 13).Using the HBond and MAXENT algorithms ( 44 , 45 ), we assessed the splice site score of the 3 and 5 splice sites (Figure 3 A).The analysis re v ealed a strong 5 splice site for exon 12, the last constituti v e e xon shared by Tr kC-T1 and Tr kC-TK+, and similar splice site scores for the 3 splice site of exon 13A (MaxEnt score of 9.21) and 13 (MaxEnt score of 7.34).As no transcript variants have been documented in w hich onl y exon 14A is included while exon 13A is skipped (Ensembl genome bro wser, Ho we et al. , 2021), the splicing decision between exon 13 and exon 13A likely determines Tr kC isoform e xpression.Gi v en the rather similar splice site strength of the competing 3 splice sites, this splicing decision is likely controlled through additional trans -acting factors that can contribute to cell type-specific splicing patterns.For a first detailed analysis, we have chosen the TrkC-T1-determining exon 13A and used the HEXplorer tool ( 46 ) to predict regulatory regions.The resulting probability profile re v ealed three main putati v e cis-acting regions (Figure 3 B).The 5 region (fragment 1, 13A-1) most likely acts as a splicing silencer, the middle region (fragment 2, 13A-2) as an enhancer, and the last one (fragment 3, 13A-3) contained both potentially enhancing and potentially silencing regions.To determine whether these predictions translate into functional roles and identify potential splicing factor binding sites, we tested the three fragments using a splicing r eporter vector (Figur e 3 C, ( 47)).Consistent with the pr edictions, fragment 1 from exon 13A favored exon skipping and fragment 2 exon inclusion (Figure 3 D and E).Fragment 3 proved to act as an exonic splicing enhancer as well, albeit less potent.
The analysis of the nucleotides essential for the maintenance of the splicing-regulatory properties of fragment 3 with HEXplorer suggested that a single nucleotide substitution could se v erely disrupt the ability of this fragment to act as an enhancer (Supplementary Figure 3A).To confirm this potential splicing-regulatory element by a second, independent algorithm, we analyzed the sequence of fragment 13A-3 using the ESEfinder tool ( 48 ).We found three GArich (GAR) elements predicted to be bound by Srsf1 with high probability (Figure 3 B and Supplementary Figure 3B).Additionall y, w hen anal yzing the sequence with the mutation predicted to be disrupti v e by HEXplorer (Supplementary Figure 3A), ESEfinder did not detect any putati v e Srsf1 binding at this site.Indeed, introducing this mutation in the fragment 3 splicing reporter strongly reduced exon inclusion (Figure 3 D and E), underscoring the importance of this region in exon 13A inclusion and hence TrkC-T1 formation.To test whether Srsf1 is indeed r equir ed for the inclusion of the e xon 13A-3, we e xpr essed the r eporter vector harboring this sequence and sim ultaneousl y knocked down Srsf1 (Figure 3 F).This resulted in a significant reduction of exon inclusion (Figure 3 G), confirming that this enhancer is responsi v e to Srsf1 le v els, which, in turn, control TrkC-T1 formation.Importantl y, the GAR-m utated reporter vector 13A-3mut did not respond to Srsf1 knock down, suggesting a direct and sequence-specific role of Srsf1 in controlling this splicing e v ent (Figure 3 F, G).Additionally, the knockdown did not impact the splicing reporters containing the other two fragments of exon 13A (Figure 3 F and G), suggesting regulation through the 13A-3 sequence.
We also investigated whether the inclusion of exon 13A is influenced by Elavl1 le v els.To this end, we used knockdown or ov ere xpression of Elavl1 in cells transfected with the splicing reporters (Supplementary Figure 4C and D).Neither decreasing nor increasing Elavl1 le v els changed the splicing behavior of any of the three exon 13A fragments (Supplementary Figure 4D).Howe v er, the mutated reporter E 13A-3 mut responded to ov ere xpressing Elavl1, but not to its knockdown.We did not pursue this avenue further, as the mutated sequence is an artificially generated one, and ther efor e likely not present in a wild type setting.
Gi v en the known role of Elavl1 in regulating mRNA stability, we also tested whether Elavl1 differentially affects the stability of TrkC-T1 and TrkC-TK+.To test this, we knocked down Elavl1 in N2A cells and then inhibited transcription using actinomycin D, monitoring the ratio of TrkC-T1 to TrkC-TK+ after three and six hours of treatment (Supplementary Figure 4F).After the knockdown, we observed an increase of TrkC-T1 in the total transcript quantity similar to the siRNA knockdown experiments (DMSO + siCtrl versus DMSO + siElavl1, compare to Figure 2 D and E), but, at the same time, a decrease in the proportion of TrkC-T1 after six hours of actinomycin D treatment as compared to the vehicle control (DMSO + siElavl1).This suggests two modes of action of Elavl1.On the one hand, it controls splicing of TrkC pre-mRNA and suppresses the generation of the TrkC-T1 isoform independent of the exon 13A sequences that we assayed in our reporter vectors.On the other hand, once the TrkC-T1 transcripts are generated, Elavl1 reduces the stability of this isoform, which is in line with the presence of Cortices from full litters of E 13.5 embryos were microdissected, dissociated into primary cortical cells and nucleofected with Srsf1 or Elavl1 expression plasmids, or empty expression constructs (pCAGIG = EV).Nucleofected cells wer e cultur ed for two days in vitro (DIV), after which total RNA or protein were extracted.( G ) Srsf1 and Elavl1 alter transcript variant ratio of TrkC-T1 and TrkC-TK+ in cortical neurons.RT-qPCR on material from the nucleofected, cultured primary cortical cells, percentage of TrkC-T1 from total TrkC transcripts (T1 plus TK+) is shown.Lines r epr esent pair ed r eplicates from the same experiment (cortical cells from one full litter split into the three nucleofection conditions).N = 4. P values from one-way ANOVA; overall P value: 0.0046.( H ) Western blot of samples from (F), probed with a pan-T rkC antibody , which detects T rkC-TK+ (130 kDa) and TrkC-T1 (100 kDa).GAPDH was detected as a loading control.N = 4; P values from one-way ANOVA with Dunnett's T3 multiple comparison post-hoc test; overall P value: 0.0278.3C).
To further understand the mechanism of TrkC AS regulation by Srsf1, we designed a 2 -MOE (antisense oligonucleotide) complementary to region 3 of exon 13A (Figure 3 H), which w e show ed to contain an Srsf1-dependent element (Figure 3 F-G).Upon transfecting this 2 -MOE into N2a cells (Figure 3 I), the proportion of TrkC-T1 dropped in an ASO-dose dependent manner to around half of its le v el in the control samples at the highest concentra tion, showing tha t this enhancer element is crucial for the control of this alternati v e splicing e v ent also in the endogenous conte xt.Ne xt, we in vitro tr anscribed r adioacti v ely labelled RNA probes with sequences corresponding to E 13A-3 and E 13A-3mut (Figure 3 J).We performed UV crosslinking (Figure 3 K) using these probes and nuclear extracts obtained from N2a cells that had been transfected with either the Srsf1 ov ere xpression plasmid or the empty vector (Supplementary Figure 4G).Ov ere xpressing Srsf1 clearly increased the intensity of a band corresponding to the size of Srsf1, demonstrating that Srsf1 directly binds to the E 13A-3 sequence element.When using the E 13A-3mut probe, the binding slightly decreased but was not fully abolished, which is consistent with the presence of additional Srsf1 binding sites in this exon (Supplementary Figure 4B).

The relative levels of Srsf1 and Elavl1 directly impact the outcome of TrkC alternative splicing
Up to this point, we had shown that Srsf1 and Elavl1 each have a significant impact on TrkC AS, with Srsf1 promoting the formation of TrkC-T1 and Elavl1 that of TrkC-TK+ (Figure 2 ).We sought to explore whether the ratio between the le v els of Srsf1 and those of Elavl1 directly dri v es the outcome of this AS e v ent.To do so, we transfected combinations of constructs aiming to sim ultaneousl y change the levels of Srsf1 and of Elavl1 in N2a cells (Figure 4 A).We either upregulated the le v els of both RBPs (pCAG-Srsf1 + pCAG-Elavl1), knocked both of them down (si Srsf1 + si Elavl1 ), or upregulated one while downregulating the other (pCAG-Srsf1 + si Elavl1 and pCAG-Elavl1 + si Srsf1 ), and then assessed TrkC AS outcomes by splicing-sensitive RT-PCR (see Supplementary Figure 5 for relati v e quantification of Srsf1 and Elavl1 transcript le v els in relation to matching control samples).We first observed that, in all of the con-trol samples (pCAGIG, siCtrl, pCAGIG + siCtrl), Srsf1 levels and Elavl1 transcripts were distributed in an about 60%-65% to 35-40% ratio in the total Srsf1 + Elavl1 transcript pool.Increasingly large alterations of the Srsf1 to Elavl1 ratio led to changes in the ratio of TrkC-T1 to TrkC-TK+ that were in line with our previous observations (Figure 4 B).W hen Srsf1 domina ted the Srsf1 + Elavl1 ra tio (pCAG-Srsf1, pCAG-Srsf1 + siCtrl and pCAG-Srsf1 + siElavl1), the TrkC AS outcome also increasingly shifted towards an increase in TrkC-T1, reaching proportions as high as around 70% from the total TrkC transcript pool.In contrast, when Elavl1 dominated the Srsf1 + Elavl1 transcript pool (pCA G-Elavl1, pCA G-Elavl1 + siCtrl, pCA G-Elavl1 + siSrsf1), the production of TrkC-T1 was suppressed down to le v els as low as 10% from the total TrkC tr anscript pool.Over all, the pr oportion of TrkC-T1 fr om the total TrkC transcript pool significantly correlated with the Srsf1-to-Elavl1 ratio (Figure 4 C, R 2 for Pearson goodness of fit of linear r egr ession: 0.8203, P < 0.0001).

Srsf1 and Elavl1 have different expression patterns during cortical neuronal differentiation
The ratio of TrkC-T1 to TrkC-TK+ significantly differs between cortical NPCs and neurons (Figure 1 ).Splicing factors frequently operate in a combinatorial fashion, with levels that differ between cell types and de v elopmental stages ( 12 , 13 ).Thus, we tested whether the difference between T1 to TK+ le v els may result from a cell type-specific di v ergence of Srsf1 and Elavl1 le v els.We performed comparati v e RNA in situ hybridization for Srsf1 and Elavl1 on cortical sections from different de v elopmental stages (Figure 4 A and Supplementary Figure 4).We detected a strong signal for Srsf1 in the stem cell compartments of the dev eloping corte x (v entricular zone and subv entricular zone, short: VZ and SVZ), while le v els in the dif ferentia ting neurons of the intermediate zone (IZ) and the cortical plate were significantly lower.This held true across the stages of CFuPN production (E 12.5-E 14.5 ( 30 )), in which TrkC-T1 le v els are also strongly elevated in NPCs ( 26 ).In contrast, Elavl1 transcripts were distributed more uniformly across the neocortex.Both mRNA expression patterns were maintained up to E 16.5 (Supplementary Figure 6A).Furthermor e, we could r eplicate these r esults for both Srsf1 and Elavl1 by RT-qPCR in prominin-1-sorted E 12.5 NPCs and r egulatory r egions.The arrow points to a nucleotide predicted by HEXplorer to be of particular importance for conferring the splicing enhancer properties to fragment 13A-3.The dotted box marks the region most strongly impacted by this nucleotide and is shown magnified in Supplementary Figure 3B, along with the predicted effects of mutating this nucleotide.C -E .Minigene analysis of exon 13A splicing r egulatory r egions.( C ) Splicing reporter used to assess enhancing or silencing properties of exon 13A fragments.The skipping and inclusion control vectors are described in ( 47 ) ( J, K ).Radioacti v ely labelled, in vitro transcribed RNA probes of the exon 13 A-3 fragment (E 13A-3) or of the same fragment with the mutation described in Supplementary Figure 4B (E 13A-3 mut.) were crosslinked by UV irradiation to nuclear extract proteins from N2a cells transfected either with an Srsf1 ov ere xpression construct (pCAG-Srsf1) or with the empty vector (pCAGIG).Arrow: Srsf1 band (see also Supplementary Figures 3A and 4G).neurons.Here, too, the le v els of Sr sf1 dif fered strongly and significantly between Prom-1-positi v e and Prom-1-negati v e cells, whereas those of Elavl1 did not (Supplementary Figure 6B).Additionally, an analysis of Srsf1 and Elavl1 levels during neurodif ferentia tion of mouse embryonic stem cells (Supplementary Figure 6C) shows Srsf1 le v els decreasing strongly after the start of neuro genesis, w hereas Elavl1 le v els do so at a much milder ra te, ef fecti v ely resulting in the change from a strongly Sr sf1 -domina ted transcript pool (day 0, high Pax6 expression) to one where its levels are reduced f ourf old and are thereby closer to those of Elavl1 (after day 16, low to no Pax6 expression, high Rbfox3 expression).
We also used publicly available single cell sequencing data from E 14, E 16 and E 18 mouse cortex (provided by ( 58 )) in order to estimate cell-specific expression levels of both the Srsf1 and Elavl1 mRNAs.We could confirm that Srsf1 and Elavl1 are more highly co-expressed in progenitors (Figure 5 B).Using the same data, we could observe that, on average, Elavl1 expr ession incr eases in late (E 16) Pax6 + progenitors, wher eas Srsf1 expr ession stays r elati v ely stab le until dropping at E 18 (Figure 5 C).Similarly, in Tbr2 + cells, Srsf1 expression decreases more precipitously from E 14 to E 18 than the expression of Elavl1 does.In order to test if Srsf1 and Elavl1 expression differs significantly across stages in cortical progenitors, we asked first if the relationship between Srsf1 and Elavl1 co-expression is different between stage and cell type, and second, for each combination of stage and cell type, if the distribution of expression values of Srsf1 differs significantly from that of Elavl1 .For the first question, we fit a negati v e binomial model of Srsf1 and Elavl1 expression values with 'stage' and 'celltype' as interaction terms to understand whether Srsf1 le v els depend on Elavl1 le v els.At E 16, Srsf1 and Elavl1 showed a significant interaction (Supplementary Figure 6D, Stage-Mouse E16 adjusted p = 2.97E-07) compared to the reference E 14 stage, while at E 18 they did not.In other words, Elavl1 and Srsf1 expr ession ar e concordant at E 16 while at E 14 and E 18 they are not.Similarl y, w hen we looked at the paired mean rank expression of Srsf1 and Elavl1 at each stage (Figure 5 C), we found that their distributions wer e significantly differ ent in all progenitors at E 14, (adjusted P < 0.001) and E 18 (adjusted P < 0.01), and also specifically in Pax6 -positi v e progenitors at E 14.
Surprisingly, howe v er, when inv estigating Elavl1 protein le v els in the de v eloping corte x, we observ ed a discrepancy between the distributions of its transcript and its protein product (Figure 5 D).The Elavl1 protein is indeed expressed at constant le v els across the germinal zones at E 11.5, and similarly in this cortical zone a t la ter de v elopmental stages.Howe v er, from E 12.5 on, the signal observed for Elavl1 in the cortical plate is vastly stronger than that observed in the VZ / SVZ.This pattern is upheld at least up to E 16.5 and indica tes tha t Elavl1 le v els in the de v eloping neocorte x are primarily regulated on the posttranscriptional le v el.
Altogether, these data are consistent with the idea that the cell type-specific Srsf1 to Elavl1 ratio in progenitors and neurons changes as dif ferentia tion proceeds and, in turn, controls the differing Tr kC alternati v e isoform distribution in these cells.

Both Srsf1 and Elavl1 control the cell fate in the developing cortex
Gi v en the key role of TrkC-T1 levels in the CFuPN-CPN fate choice in the de v eloping corte x ( 26 ) and since Srsf1 and Elavl1 altered the T1 to TK+ ratio (Figure 2 ), we tested if changing the expression levels of these splicing factors in vivo might change the proportion of these cell types in the neocortex.We assessed this by in uter o electropora ting (described in ( 28 )) either ov ere xpression or knockdown constructs for Srsf1 and Elavl1 in the cortical NPCs at E 12.5.Four days post-electropora tion, a t E 16.5, the cortices were analyzed for the proportion of CFuPNs and CPNs (Figure 5 ).All constructs co-express GFP, which enables the identification of the progeny of the electroporated NPCs using immunofluorescence.Ctip2 and Satb2 are key determinants of the CFuPN and the CPN fate, and are routinely used to quantify the proportion of these cell types among electroporated (GFP-positi v e) cells (49)(50)(51)(52)(53).We compared the percentages of each category of double positive cells (Ctip2 + GFP + and Satb2 + GFP + ) to the corresponding category in littermate control embryos, which were electroporated with the empty vector (pCAGIG) or a scrambled shRNA.We found that increasing the le v els of Srsf1 increased the proportion of GFP-Ctip2 doub le positi v e cells at the expense of GFP-Satb2 positi v e ones, whereas the knockdown of Srsf1 had the opposite effect (Figure 5 A).Modulating the le v els of Elavl1 in NPCs had the opposite effects: an increase in Elavl1 decreased the proportion of Ctip2-positi v e cellular progeny and increased that of Satb2positi v e daughter cells, and vice versa for the knockdown of Elavl1 (Figure 5 B).Hence, both Srsf1 and Elavl1 influence the CFuPN-CPN neuron subtype fate choice in opposing ways.

The cell fate effects of Srsf1 and Elavl1 are mediated by TrkC-T1
Next, we inquired whether the effects on neuron subtype fate observed for the Srsf1 and Elavl1 level alterations are mediated by Tr kC alternati v e splicing in vivo .To address this question, we co-electroporated Srsf1 or Elavl1 expression constructs with constructs aimed at compensating for the incr ease, r especti v ely decrease, of Tr kC-T1 le v els caused by these factors and assessed the resulting proportions of Ctip2-and Satb2-positi v e progeny.Combining the splicing factor expression with a modulation of TrkC-T1 levels led to a mitigation of the fate choice phenotypes observed w hen solel y altering Srsf1 or Elavl1 le v els (Figure 5 C).The proportions of GFP-Ctip2 and GFP-Satb2 double positive progeny were not significantly different from those in the littermate controls.Altogether, we present evidence that two splicing factors, Srsf1 and Elavl1, regulate TrkC alternati v e splicing in a cell type-specific manner, which then contributes to control dif ferentia tion of NPCs into functionally different neurons.

DISCUSSION
De v elopmental fate choices in the production of projection neuron subtypes are crucial for generating both the connections within the neocortex and the ones between the neocortex and other brain structures ( 30 , 54 ).The past years have brought about extensive r esear ch on how this species-and region-specific organization of neuron subtypes is achie v ed during corticogenesis, b ut the contrib uting progenitor-intrinsic molecular mechanisms are still incompletely understood.Numerous studies have either identified individual epigenetic or transcription factors that shape the cortical NPC fate (re vie wed in ( 30 )) or stri v ed to capture the di v ersity of embryonic NPC subtypes at the whole-transcriptome le v el (55)(56)(57)(58)(59).Still, there has been little r esear ch on how the regulation of RNA processing ensures the le v els of fate-determining factors in NPCs that ultima tely dicta te the fa te of their neur onal pr ogeny.In this stud y, we elucida ted an alternati v e splicing-based mechanism that upholds appropriate le v els of Tr kC-T1, an isoform of the neur otr ophin-3 receptor TrkC, which we have previously shown to be a deep layer (CFuPN) neuron subtype determinant ( 26 ).We found that the balance between T rkC-T1 and T rkC-TK+ is stage-and cell type-specific in the de v eloping corte x, and that this balance is controlled antagonistically by the RBPs Srsf1 and Elavl1.Furthermore, w e show ed tha t Srsf1 and Elavl1 exhibit dif fer ential expr ession in different cell types in the de v eloping neocorte x.Finally, we pr esent dir ect in vivo evidence that these two splicing factors steer the CFuPN-CPN fate choice during corticogenesis.To our knowledge, this is the first example of alternati v e splicing regulation that controls the ratio between CFuPN and CPN numbers.

The precise cellular ratio between TrkC-T1 and TrkC-TK+ depends on the cortical cell type
The ratio between TrkC-T1 and TrkC-TK+ has been shown to be crucial for dorsal root ganglion-deri v ed neuron axonogenesis, with the two receptors having dose-dependent antagonistic effects on the number of processes formed ( 60 ).In a previous publication ( 26 ), we showed that TrkC-T1 impacts signal transduction by sequestering the scaffolding adapter molecule ShcA, and this is not observed with the TrkC-TK+, as had been previously suggested ( 61 , 62 ).In the current study, we also detected TrkC-TK+ in sorted cortical NPCs at E13.5 by RT-qPCR (Figure 1 C-E), which is in agreement with findings from a previous study ( 63 ).Additionally, this study found that NPCs are responsi v e to NT-3 signals mediated by TrkC-TK+.Furthermore, in an earlier publication ( 64 ), we found that NT-3 production by postmitotic cortical neurons is a key feedback mechanism for the NPCs switching from deep to superficial layer neur on pr oduction.Taken together, these results argue for the need for a balance between TrkC-TK+ and TrkC-T1 signaling in cell fate decisions in the NPCs.We indeed found that the ratio of T1 to TK+ shifts in favor of TK+ during corte x de v elopment from a whole-tissue perspecti v e (Figure 1 B and ( 26)), and that this is likely due to the change in cellular composition.We found that NPCs and differentiating neurons exhibit different ratios of TrkC-T1 to TK+ (Figure 1 E).While we cannot exclude the effect of other cell types, the change in cell type prevalence in the de v eloping cortex from an NPC-dominated tissue (E 12.5) is overw helmingl y dri v en by the production of postmitotic neurons.This may explain the balance shift observed in bulk tissue.
Pr evious r esear ch indica tes tha t the NPC popula tion at any one cortical de v elopment stage may not be homogeneous in their potential to produce differently fated progeny ( 65 ), and recent single-cell RNA sequencing stud-ies support this hypothesis ( 56 , 58 ).Gi v en the considerable increase in, for instance, intermediate progenitor numbers over the stages in which TrkC-T1 levels drop in the v entricular / subv entricular zone, it may be tha t, a t early corte x de v elopment stages, some NPCs exhibit a CFuPNfavoring T1 / TK+ balance and others a CPN-favoring one.Our RNA sequencing data analysis (Figure 5 B and C) and our previous findings ( 26 ) indicate that a changing ratio of Srsf1 and Elavl1 during dif ferentia tion may cause a shift fr om CFuPN pr oduction to the pr oduction of latergenera ted fa tes, such as the CPN fate.To acquire insight into whether this is truly the case in vivo and how this impacts cell fate, highly sensiti v e mass spectrometry techniques, single-cell proteomics and single-cell RNA sequencing at isoform resolution ( 66 , 67 ) on sorted NPCs may provide answers.

Srsf1 and Elavl1 co-regulate the balance of TrkC-T1 to TrkC-TK+ and steer the CFuPN-CPN fate choice
The work presented here and that of others ( 60 , 68 ) demonstrates the importance of upholding a finely tuned cellular balance between TrkC-T1 and TrkC-TK+ .However, until now, we have had no knowledge on the mechanisms regulating this balance.Here, we show by radioacti v e splicingsensiti v e RT-PCR, RT-qPCR, and Western blotting that the splicing factors Srsf1 and Elavl1 have antagonistic effects on the alternati v e splicing of the TrkC pre-mRNA, with Srsf1 favoring the formation of TrkC-T1 and Elavl1 that of TrkC-TK+ (Figure 2 C-H).Additionally, we show that the precise cellular ratio of Srsf1 to Elavl1 transcripts is sufficient for driving the changes observed for TrkC AS (Figure 4 ).
Previous large-scale RNA sequencing projects and bioinformatic analyses showed that alternati v ely spliced last exons are an especiall y finel y regulated class of alternati v e splicing e v ents in de v eloping neural cells.Their alternati v e inclusion in mRNA often leads to the expression of two main protein isoforms with distinct C-terminal protein domains that frequently undergo signaling-relevant phosphoryla tion ( 19 , 69 ).Similarly, alterna ti v e splicing of penultimate exons whose exclusion induces a frameshift, leading to proteins with altered C-termini, and is highly regulated during neuronal dif ferentia tion ( 70 ).Our findings regarding Tr kC alternati v e splicing and stability regulation fit these patterns.Unexpectedly, though, in this instance, the counterplayer of the regulatory SR protein (Srsf1) is not an hn-RNP protein, as is frequently the case ( 71-73 ), but Elavl1.Elavl1, also known as HuR, has been more commonly associated with mRNA stability and transla tional regula tion rather than alternati v e splicing.It binds to AU-rich elements in the 3 UTR of mRNAs and thereby stabilizes them.Nonetheless, Elavl1 has also been associated with the regulation of alternati v e splicing in some cases ( 43 , 74-78 ).Intriguingly, we see a potential dual role of Elavl1, acting in the splicing choice between the TrkC-T1 and TrkC-TK+ transcript, but also in the differential stabilization of the two (Supplementary Figure 4F).This may be due to the presence of se v eral strong predicted Elavl1 binding sites in the 3 UTR of TrkC-T1 (Supplementary Figure 4E), whereas no such binding sites could be detected in the 3 UTR of TrkC-TK+.In contrast, the role of Srsf1 in TrkC AS clearly depends on a splicing enhancer in the last third of exon 13A (E 13A-3), which is crucial for TrkC-T1 formation (Figure 3 H-K) and loses some of its enhancing ability when Srsf1 is knocked down (Figure 3 F-G).
Based on these results and together with the results of the correlation analysis of the Srsf1 / Elavl1 ratio with T rkC-T1 / T rkC-TK+ (Figure 4 ), we hypothesize that, for TrkC-T1 to be formed, Srsf1 le v els hav e to be considerab ly higher than Elavl1 le v els.If the role of Elavl1 in alternati v e splicing is overridden by Srsf1, then, TrkC-T1 can result from this processing step.As a fine-tuning regulatory step, Elavl1 can then bind to the 3 UTR of the processed TrkC-T1 mRNA, having a modest stabilizing effect.Due to the small effect size of the Elavl1 knockdown on TrkC-T1 le v els in the actinomy cin D-treated cells (around 10% change in PSI compared to the DMSO + siElavl1 sample) versus the much larger effect on alternati v e splicing (around 50% change in PSI compared to the siControl, Figure 2 D-E and Supplementary Figure 2A), we suggest that Elavl1 primarily acts as a splicing regulator to balance the TrkC-T1 and TrkC-TK+ isoforms.Taken together, these findings suggest a regulatory network that fine-tunes TrkC transcript variant levels, orchestrated by Srsf1 and Elavl1.
Ther e ar e only few well-documented cases in which transcript variant ratios are involved in cell fate decisions, and e v en fe wer regar ding neuron subtype decisions.One known case of a splicing factor being involved in a neuron subtype decision in corticogenesis is that of SRRM4 ( 79).This study showed that SRRM4 impacts the numbers of Tbr1-and Satb2-positi v e neurons, but it does not show a regulation of the overarching CFuPN fate.Tbr1-positi v e (corticothalamic pr ojection neur ons, a subset of CFuPNs) and Satb2-positi v e neurons (CPNs) only show minor ov erlap in their generation time frames and minimal shared layer occupancy ( 30 ), which is why the ultimate impact magnitude of this fate control mechanism is unknown.Furthermore, the alternati v ely spliced transcripts mediating this function of SRRM4 were not described ( 79 ).Here, we show, for the first time, that the splicing factors Srsf1 and Elavl1 dri v e significant changes in the fate acquisition process for CFuPN and CPN in the de v eloping corte x (Figure 6 ), an effect mediated by their antagonistic effects on TrkC alternati v e splicing and stability and the resulting balance between the receptor isoforms TrkC-T1 and TrkC-TK+ (Figure 2 F-G, Supplementary Figures 4F and 6A).

Srsf1 and Elavl1 levels define cell-type specific splicingregulatory environments in the developing cortex
Across the adult mammalian tissues investigated previously, it has been found that AS frequency is highest in the brain, a phenomenon likely caused by the high number of RBPs expressed in this tissue and their dynamic and variable interaction networks ( 22 , 80 ).A previous study has emphasized the impact of alternati v e splicing on cortex development, as a major tissue-wide splicing switch occurs prena tally, a t E 14.5 ( 81 ).Here, we show how two novel players in alternati v e splicing regulation during neurode v elopment, Srsf1 and Elavl1, steer fate choices early in cortex development.Using RNA in situ hybridization, immunofluorescence, and RT-qPCR on sorted cortical aRGCs and postmitotic neurons, we could show that Srsf1 and Elavl1 expression levels define Tr kC alternati v e splicing-regulatory environments in the de v eloping corte x that are distinct between progenitors and neurons (Figure 5 and Supplementary Figure 6).Although our analysis was focused on Tr kC alternati v e splicing, the balance of Srsf1 and Elavl1 likely affects splicing e v ents and mRNA stability for additional targets in a cell type-specific manner.We found Srsf1 expression to contrast starkly between the two cell types, with far stronger expression in aRGs than neurons, while Elavl1 mRNA le v els were more similar across the cortical tissue (Figure 5 A) and the two cell types (Supplementary Figure 6B and C).
Unexpectedly, the Elavl1 protein distribution does not entirely follow the distribution of its transcript (Figure 5 B), accumulating more strongly in the cortical plate of the dev eloping neocorte x (Figure 5 D).This finding points to a ratio of Srsf1 to Elavl1 in postmitotic neurons that could be much more strongly dominated by Elavl1 than anticipated from the mRNA data.We have recently shown that corte x de v elopment is rife with other instances of stark discrepancies between transcript and protein le v els in different cell types ( 82 ).A good example for this phenomenon is the chroma tin-associa ted CPN fa te marker Sa tb2 ( 51 ), whose protein is solely detected in this postmitotic neuron subtype, whereas its transcripts are present without being translated in a much broader spectrum of cortical cells, including in neural progenitor cells ( 82 ).Our da ta indica te tha t Elavl1 protein and transcripts also exhibit such a discrepancy.
The role of Srsf1 in central nervous system (CNS) development has not been extensively addressed before, likely because the embryonic lethality of Srsf1 deletion mouse lines has been ascribed to cardiovascular and skeletal defects ( 83 , 84 ).As for Elavl1, the only instance of de v elopmental alternati v e splicing regulation through it and other Elav protein family members has only recently been reported in the fruit fly CNS ( 85 ).In mammalian corticogenesis, Elavl1 has solely been shown to act as a stage-specific regula tor of mRNA transla tion in the mouse, exhibiting the same expression pattern we could observe in our work on both RNA and protein le v el ( 86 ).In this earlier publication, Elavl1 was shown to alter the phosphorylation states of core ribosomal components through collaborati v e action with the eIF2-alpha kinase 4, which impacts the association of transcripts with ribosomal components and the forma tion of polysomes.W hile the authors showed changes in Ctip2 mRNA distribution in the unbound versus 40S-60S and polysomal fractions of Elavl1 conditional knockout animals, this was not directly causally linked to a change in the CFuPN / CPN fate.Our results suggest that Elavl1 participates in the CFuPN / CPN fate decision through alternati v e splicing and mRNA stability regulation (Figures 3 and  5 ).Combined, these findings pose the question of whether Elavl1 may have a dual role in establishing the CFuPN fate, both via regulating the alternati v e splicing of TrkC in NPCs and by Ctip2 translation control after cell cycle exit in deep layer neurons.Further studies employing a Dcx-promoterdri v en knockdown or ov ere xpression of Elavl1 may help to disentangle the pre-and postmitotic involvement of Elavl1 in the CFuPN fate.
In conclusion, we show for the first time direct in vivo evidence that Elavl1 and Srsf1 contribute to the fate switch between CFuPN and CPN in the de v eloping corte x, acting at the le v el of Tr kC splicing and stability.Since the cell type-specific distribution of Srsf1 and Elavl1 is maintained up to E 16.5 (Supplementary Figure 5), outside of the time window in which CFuPN and CPN fate acquisition overlap, their balance may participate in other NPC-or neuronspecific splicing e v ents that are independent of TrkC alternati v e splicing but of importance to later de v elopmental processes, such as the neuro genesis-glio genesis switch at E 17.5.This avenue remains to be explored using manipulations of Srsf1 and Elavl1 le v els at other de v elopmental time points.

DA T A A V AILABILITY
The data underlying this article are available in the article and in its online supplementary material.

SUPPLEMENT ARY DA T A
Supplementary Data are available at NAR Online.

Figure 1 .
Figure 1.The balance between the two Tr kC alternati v e splicing isoforms TK+ and T1 is regulated in a de v elopmentally dynamic and cell type-specific manner.( A ) Alternati v e splicing of the Tr kC (Ntr k3) pre-mRNA produces the T1 and TK+ receptor variants.Two groups of m utuall y e xclusi v e e xons (13A-14A and 13-17) gi v e rise to the distinct 3 termini of the T rkC-TK+ and T rkC-T1 transcript variants.Correspondingly , these translate to distinct intracellular domains at the C-termini of the protein isoforms, giving rise to either a kinase domain (TK+) or a catal yticall y inacti v e domain (T1).Stop codons are indicated and demarcate the start of variant-specific 3 UTRs.Binding sites for the probes used in RT-qPCR are indicated at the respecti v e exon junctions.( B ) The balance between TrkC-TK+ and TrkC-T1 changes during cortex development.In RNA pr epar ed from cortices of increasing embryonic age, TaqMan quantitati v e real-time PCR for the two Tr kC isoforms shows that the balance between TK+ and T1 shifts in favor of TK+ as corte x de v elopment pr ogresses fr om embry onic da y E 11.5 to E 18.5.N = 4. Bars: mean percentage of isoform from total Tr kC e xpression (T1 plus TK+) ± SD.P value deri v ed from unpaired, two-tailed Student's t test with Welch's correction.( C-E ).The balance between TrkC-TK+ and TrkC-T1 in the de v eloping corte x is cell-type specific.( C ) Primary cortical cells from whole E 12.5 embryo litters were sorted into neuronal and stem cell populations by FA CS after la belling with an anti-pr ominin-1 antibody (Pr om-1) and collected for further analysis.( D ) Example APC versus count plots used to distinguish viab le Prom-1 positi v e and negati v e cells.Ga ting stra tegy according to signal from isotype control-stained cells.Complete gating strategy is presented in Supplementary Figure 1B.( E ) RT-qPCR was performed on mRNA from the sorted neocortical cells, as described in (B).N = 3. P value deri v ed from paired, two-tailed Student's t test.Pairing efficiency between Prom-1 + and Prom-1-results: r = 0.9962.

Figure 2 .
Figure 2. Elavl1 and Srsf1 regulate Tr kC alternati v e splicing in N2A cells and in primary cortical neurons.( A ) Selection of splicing factors (SFs) and other RNA-binding proteins (RBPs) with potential involvement in TrkC alternative splicing.B -E .Elavl1 and Srsf1 control Tr kC alternati v e isoform expression.( B ) Str ategy for r adioactive splicing-sensitive RT-PCR for evaluating the TrkC-T1 and TrkC-TK+ splicing event.TrkC AS was assessed in N2A cell samples where RBPs defined in (A) were knocked down using siRNAs.( C ) Exemplary result of a radioacti v e splicing-sensiti v e RT-PCR for TrkC-T1 and TrkC-TK+ on RNA from N2A cells treated with the indicated siRNAs.Percentage of TrkC-T1, as represented in (D) and (E), was quantified using a Phosphorimager and the ImageQuant TL software.Ctrl -siCtrl.( D ) Summary plot for all tested RBPs and their effect on the proportion of the TrkC-T1 transcript variant normalized to TrkC-T1 percentage in the control siRNA samples.Gray dotted circles graduate the plot, indicating increases (positi v e values, outside the zero circle) or decreases (negati v e values, inside the zero circle) in TrkC-T1 percentage as compared to the siCtrl samples.Error bars were omitted for clarity.Statistically significant changes (si Srsf1 and si Elavl1 samples) are represented separately with the corresponding descripti v e and analytical statistical information in (E).( E ) siRNA-mediated knockdown of Elavl1 or Srsf1 in N2A cells changes the ratio of TrkC-T1 to TrkC-TK+ significantly.N = 3. Bars: mean percentage of isoform from total Tr kC e xpression (T1 plus TK+) ± SD.P values deri v ed from Brown-Forsythe and Welch ANOVA with Dunnett's T3 multiple comparison post-hoc test.Overall P value: 0.0002.( F ) Stra tegy for modula ting SF le v els in cortical neurons.

Figure 3 .
Figure 3. TrkC transcript levels are regulated by an Srsf1-dependent exonic splicing enhancer element in the first TrkC-T1-specific ex on, ex on 13A.( A ) Splice site strength prediction of the Ntrk3 primary transcript.( B ) Bioinformatic analysis of exon 13A suggests its subdivision in three major splicing- ( D ) The TrkC-T1 exon 13A r eporter vectors wer e transfected into N2A cells and the splicing outcome assessed by RT-PCR.The mutation in 13A-3 predicted to disrupt Srsf1 binding impedes the splicing enhancing ability of this element, leading to a significant reduction of exon inclusion, as quantified in E. N = 3; P values deri v ed from Brown-Forsythe and Welch ANOVA test with Šidak's post-hoc multiple comparisons test.Overall P value: < 0.0001.The inclusion product of the 13A-3 reporter is larger due to the larger insert size (see B). ( E ) Quantification of (D).( F ) To assess the involvement of Srsf1 in Tr kC alternati v e splicing, TrkC-T1 exon 13A reporter vectors were transfected into N2A cells together with siRNAs as indicated, and the splicing outcome assessed by RT-PCR.N = 3. P values deri v ed from or dinary ANOVA test with Šidak's post-hoc multiple comparisons test.Overall P value: < 0.0001.( G ) Quantification of (F).( H, I ) Blocking the putati v e Srsf1 binding site in exon 13A of the TrkC pre-mRNA leads to a decrease in TrkC-T1 formation.A 2 -MOE antisense oligonucleotide complementary to the putati v e Srsf1 binding site in exon 13A (H) was transfected into N2a cells and its effect on Tr kC alternati v e splicing assessed by RT-qPCR (I).N = 3; P values deri v ed from or dinary one-way ANOVA with Šidak's post-hoc multiple comparisons test.Overall P value: < 0.0001.

Figure 4 .
Figure 4.The ratio of Srsf1 to Elavl1 steers the alternati v e splicing choice between TrkCT1 and TrkC-TK+ .( A ) Srsf1 to Elavl1 ratios were modulated in N2a cells by transfection with either expression constructs, siRNAs against the two transcripts, or combina tions thereof.Combina tions are indica ted above.The resulting ratio of Srsf1 to Elavl1 transcripts was determined by RT-qPCR and plotted below.The percentage of each transcript was calculated based on the C T values by assuming one cycle difference in C T to indicate a twofold difference.The bars are divided at the mean percentage from N = 3 r eplicates.Error bars r epr esent the standar d de viation of the thr ee r esults.The C T values for Srsf1 and Elavl1 transcript le v els relati v e to Hprt transcript le v els and matching control sample are summarized in Supplementary Figure 5. ( B ) The effect of the Srsf1 and Elavl1 modulations in (A) was assessed by radioacti v e, splicing-sensiti v e PCR specific to the Tr kC-T1 / TK+ alternati v e splicing e v ent, as described in Figure 2 B. Each lane corresponds to the Srsf1 / Elavl1 ratio and transfection conditions indicated above it in panel (A).The quantification was performed by normalizing the intensity of the TrkC-T1 band to the total signal from the TrkC-T1 and TrkC-TK+ bands.The bars are divided at the mean percentage from N = 3 replicates.Error bars r epr esent the standar d de viation from the mean of the thr ee r esults.P values deri v ed from or dinary one-way ANOVA with Šidak's post-hoc multiple comparisons test.( C ) Correlation analysis of the percentage of Srsf1 in Srsf1 + Elavl1 transcripts with the percentage of TrkC-T1 in the TrkC-T1 + TrkC-TK+ transcripts for the experiment presented in (A) and (B).Dots r epr esent individual biological replicates.pCAGIG -pCAG-IRES-GFP, empty vector; pCAG-Srsf1 -ov ere xpression v ector containing the Srsf1 CDS; pCAG-Elavl1 -ov ere xpression v ector containing the Elavl1 CDS; siCtrl -control siRNA (siAllstar); siSrsf1 -pool of siRNAs against mouse Srsf1; siElavl1 -pool of siRNAs against mouse Elavl1.

Figure 5 .
Figure 5. Srsf1 and Elavl1 are differentially expressed in the de v eloping neocorte x. ( A ) Srsf1 mRNA le v els are high in the v entricular zone and low in the intermediate zone and cortical plate, while Elavl1 is uniformly expressed.RNA in situ hybridization with probes against Srsf1 and Elavl1 showed different

←
−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−− − expression patterns in the developing cortex.Exemplary coronal sections of brains at the indicated embryonic stages show strong expression of Srsf1 in the ventricular zone, with much weaker expression outside this compartment.VZ -ventricular zone, CP -cortical plate, Pia m. -Pia mater , LV -lateral ventricle.Dashed box es r epr esent ar eas shown ma gnified in the last or last two ma gnification insets, respecti v ely.( B ) Srsf1 and Elavl1 are strongly coexpressed in neural progenitor cells in the developing neocortex.Expression plots generated using Seurat v4 and FindAllMarkers package for cells known to be part of the pyramidal neuron lineage.Expression le v els generated by the SCTransform package.L -layer.NPCs -neural progenitor cells.Immat.neurons -immature neurons.CA1, CA3 -Cornu ammonis areas of the hippocampus; DG -dentate gyrus of the hippocampus.( C ) The ratio of Srsf1 and Elavl1 mRNA le v els in neural progenitor cells flips during corticogenesis.Data obtained from the same analysis as in (B), depicting the stage-specific expression of Srsf1 and Elavl1 mRNAs in different neural progenitor subsets.Asterisks denote significant differences between Elavl1 and Srsf1 expression values when testing the fit with a negati v e binomial model with stage and cell type as interaction terms.Underlying coefficients and further analyses are detailed in Supplementary Figure 6D-F.Pax6 + cells: apical NPCs.Tbr2 + cells: basal NPCs.( D ) Elavl1 protein distribution differs from the distribution of its mRNA in the de v eloping corte x.Imm unofluorescent micro gra phs of E 11.5 to E 16.5 cortex sections stained with antibodies against Elavl1 and the neuronal marker MAP2 show an increased signal intensity for Elavl1 in the nascent and formed cortical plate as compared to the VZ / nascent SVZ as in the CP.VZ -ventricular zone, SVZ -subventricular zone, IZ -intermediate zone, CP -cortical plate, MZ -marginal zone.Asterisk denotes putati v e interneurons, which migrate into the neocortex beginning with E 15.5.Scale bars: 50 m.

Figure 6 .
Figure 6.Srsf1 and Elavl1 act antagonistically on Tr kC-T1 le v els to control the numbers of corticofugal neurons (CFuPNs) and callosal projection neurons (CPNs) in the de v eloping corte x. ( A ) Srsf1 ov ere xpression increases the number of CFuPNs in vivo, while its downregulation decreases it, and elicits the