Alternative translation initiation produces synaptic organizer proteoforms with distinct localization and functions

SUMMARY

While many mRNAs contain more than one translation initiation site (TIS), the functions of most alternative TISs and their corresponding protein isoforms (proteoforms) remain undetermined. Here we showed that alternative usage of CUG and AUG TISs in neuronal pentraxin receptor (NPR) mRNA produced two proteoforms, of which the ratio was regulated by RNA secondary structure and neuronal activity. Downstream AUG initiation truncated the N-terminal transmembrane domain and produced a secreted NPR proteoform sufficient in promoting synaptic clustering of AMPA-type glutamate receptors. Mutations that altered the ratio of NPR proteoforms reduced AMPA receptors in parvalbumin-positive interneurons and affected learning behaviors in mice. In addition to NPR, upstream AUU-initiated N-terminal extension of C1q-like synaptic organizers anchored these otherwise secreted factors to the membrane. Together, these results uncovered the plasticity of N-terminal signal sequences regulated by alternative TIS usage as a potentially widespread mechanism in diversifying protein localization and functions.

Graphical Abstract

eTOC blurb

Unlike most mRNAs that encode one protein, Lee et al. discover that several neuronal synaptic organizer mRNAs encode more than one protein product (proteoform) by using multiple translation initiation sites. Alternative translation initiation remodels the N-terminal signal sequence, diversifies proteoform localization and functions, and impacts animal behaviors.

INTRODUCTION

Eukaryotic mRNAs are considered predominantly monocistronic. After scanning from the 5′ end of mRNAs, ribosomes initiate translation usually at an AUG TIS, translocate while decoding one codon at a time through the open reading frame (ORF), and terminate at the first in-frame stop codon, producing one polypeptide chain¹. Deviations from this conventional model such as internal ribosomal entry sites², ribosomal frameshifting during elongation³, and stop-codon read-through⁴ had typically been considered either rare or restricted to viral mRNAs. Countering this conventional view, high-throughput translatomics studies especially those using ribosome profiling have identified numerous previously unannotated ORFs^5,6, many of which can be validated by proteomics and when disrupted can cause detectable cellular phenotypes⁷. While these studies hint at the potential contribution of alternative ORFs to greatly enhance proteome complexity, the extent to which these alternative translation events and/or their protein products may have physiological functions is unclear. A large category of alternative ORFs initiate at alternative TISs that are in-frame with the canonical TISs and are expected to produce either N-terminal extended or truncated isoforms of the canonical proteins^8–10. To date, only a few N-terminal proteoforms have been functionally characterized^11–15. Whether alternative TISs may regulate protein functions in any generalizable mechanism remains unclear.

Initially identified as binding proteins for the snake venom toxin taipoxin, neuronal pentraxins comprise a family of synaptic organizers, which have been shown to recruit AMPA receptors to excitatory synapses through direct interactions between the pentraxin domain and the N-terminal domain of AMPA receptor subunits^16–18. Neuronal pentraxin 1 (NP1) and 2 (NP2, also known as neuronal activity-regulated pentraxin or NARP), encoded by NPTX1 and NPTX2 genes, respectively, are secreted factors with N-terminal cleavable signal peptides. In contrast, a third member in this family, neuronal pentraxin receptor (NPR), encoded by the NPTXR gene, is a type II membrane protein with a single N-terminal transmembrane domain, which also functions as a signal anchor targeting the ribosome-mRNA-nascent peptide chain complex to the endoplasmic reticulum (ER) surface. Previous work has shown that neuronal pentraxins can assemble into heteromeric complexes¹⁹ and play an important role in the abundance of GluR4-containing AMPA receptors in parvalbumin-positive (PV⁺) interneurons in vivo²⁰. Intriguingly, early work on NPR has found that the N-terminal methionine residue corresponds to a CUG TIS in its mRNA sequence²¹. This CUG codon in NPR mRNA is highly conserved among vertebrates (Figure 1A), hinting at its potentially unique functional role that cannot be readily fulfilled by an AUG TIS.

Figure 1. Alternative translation of NPR mRNA produces two N-terminal proteoforms.

(A) Peptide sequence (top) and genome alignment (bottom) of the NPR N-terminal region across vertebrates. The annotated CUG TIS and a downstream AUG are indicated (red).

(B) Endogenous NPR in forebrain mouse synaptosomes before or after deglycosylation (−glyc).

(C) Endogenous NPR in various brain regions (OB: olfactory bulb, FC: frontal cortex, HP: hippocampus, MB: midbrain, CB: cerebellum).

(D) Neuronal expression of wild-type (CUG) and mutant NPR cDNAs in which CUG TIS was replaced with AUG or the first 11 codons were deleted (ΔN).

(E) Expression in primary neurons of NPR cDNAs with TIS mutations in primary neurons. All samples were deglycosylated in subsequent Western blot analyses.

Here, by investigating the functional role of this noncanonical TIS in NPR mRNA, we uncovered a general mechanism by which alternative TIS usage remodeled the N-terminal signal sequence and altered the localization of the resulting proteoforms. Alternative translation initiation at a downstream AUG TIS truncated the N-terminal signal anchor to a cleavable signal peptide, thereby converting transmembrane NPR to a secreted, short NPR proteoform sufficient in promoting AMPA receptor clustering. A CTG-to-ATG knock-in mouse model with an altered NPR proteoform ratio showed reduced AMPA receptor abundance in PV⁺ interneurons and changes in learning behaviors. We further identified dozens of alternative TIS-associated proteoform pairs from a variety of biological pathways including all four members of the C1q-like (C1QL) synaptic organizers, suggesting widespread impact of alternative TIS usage on protein localization and functions.

RESULTS

Alternative translation of NPR mRNA produces two N-terminal proteoforms

As NPR has been shown to primarily regulate excitatory synapses^17,20,22, we analyzed its endogenous expression in mouse forebrain synaptosomes. As expected for most membrane proteins and consistent with previous studies²¹, NPR is glycosylated at multiple asparagine residues, as indicated by a reduction of apparent molecular weights after treatment by a mix of N- and O-glycosidases (Figure 1B). After deglycosylation, however, two distinct species with slightly different molecular weights were still observed (Figure 1B). The two species were readily detectable in most brain regions except for the cerebellum, in which NPR abundance was low and the short species was less apparent (Figure 1C). To test whether the distinct NPR species may be associated with the CUG TIS, we ectopically expressed either wild-type (WT) NPR mRNA along with its native 5′ untranslated region (UTR), or a mutant in which the CUG TIS was replaced by AUG. We found that WT NPR mRNA also produced two proteoforms in primary cortical neurons at a similar ratio to that in the brain (Figure 1D). In contrast, mutant NPR mRNA with an AUG TIS produced predominantly the long form (Figure 1D; Figure S1A), indicating that the short NPR proteoform is not due to post-translational modification nor partial degradation from the long proteoform, and that its expression is specific to WT NPR mRNA with a CUG TIS.

The dependency on CUG TIS suggested that the heterogeneity of NPR proteoforms might be due to alternative translation initiation. Examination of previously published²³ initiating ribosome profiling data obtained from harringtonine-treated rat hippocampal neurons revealed substantial ribosome-protected fragments (RPFs) at a conserved, in-frame AUG 11-codon downstream from the CUG TIS (Figure S1B), suggesting this downstream AUG might be an alternative TIS. Consistent with this possibility, deletion of the first 11 codons yielded an N-terminal truncated product with a similar size to that of short NPR (Figure 1D), whereas substituting CUG with CUA reduced the long but not short NPR proteoform (Figure 1E). Interestingly, substituting AUG with GCG did not substantially decrease short NPR (Figure S1A), suggesting that other TISs such as a CUG immediately downstream of AUG (Figure 1A) could also initiate translation when AUG was mutated. Indeed, mutating both downstream alternative TISs eliminated short NPR (Figure 1E; Figure S1A). These results indicate that both endogenous and ectopically expressed NPR mRNAs undergo alternative translation initiation at either the previously annotated CUG TIS or the downstream AUG/CUG, producing two N-terminal proteoforms.

A stable RNA secondary structure promotes CUG initiation in mammals

Both the upstream CUG and downstream AUG TISs are conserved in vertebrates (Figure 1A). To test whether the production of dual proteoforms is also conserved, we in vitro transcribed four reporter mRNAs, each of which contained the first 135 nucleotides of NPR coding sequence from either human, mouse, frog, or zebrafish, followed by a GFP sequence, and translated each mRNA in rabbit reticulocyte lysates. Both proteoforms were produced from each of the four NPR orthologs (Figure 2A), suggesting that alternative translation initiation of NPR is indeed evolutionarily conserved. However, the CUG initiation efficiency was substantially higher in human and mouse than that in frog and zebrafish NPR mRNAs (Figure 2A). Because all four reporter mRNAs were translated in the same in vitro system, the difference in CUG initiation efficiency must be due to the presence of one or more cis-regulatory elements.

Figure 2. A stable RNA secondary structure promotes CUG initiation in mammals.

(A) Expression of GFP carrying the N-terminal sequences from four NPR orthologs in rabbit reticulocyte lysate (top). Quantified proteoform ratios were correlated with their respective predicted MFEs (bottom). Data represent mean ± SD. n=3 independent experiments.

(B) Predicted secondary structure of the first 100 nucleotides of mouse NPR coding sequence. Mutations that destabilized or restored the stem-loop structure are indicated in red and blue, respectively.

(C) Proteoform ratios of wild-type (WT), destabilized, and compensatory mutant NPR cDNAs expressed in N2A cells. Predicted MFEs are shown. Data represent mean ± SD. n=3 independent experiments. ***, p<0.001, Student’s t tests.

Previous studies have shown that downstream RNA secondary structures can enhance initiation efficiency at near-cognate TISs as well as AUG TISs within non-optimal Kozak contexts^24,25. Interestingly, in silico analysis²⁶ predicted the presence of stable RNA stem-loop structures located 10–12 nucleotides downstream of the CUG TISs in human and mouse NPR mRNAs, with predicted minimum free energies (MFE) of −58.3 and −55.5 kcal/mol, respectively (Figure 2A,B). In comparison, the predicted secondary structures in frog and zebrafish NPR mRNAs were less stable (MFE= −38.0 and −39.1 kcal/mol, respectively) (Figure 2A; Figure S2).

To test the causal role of the predicted RNA structure in alternative TIS usage, we first introduced synonymous mutations to the left arm of the stem to disrupt the secondary structure (MFE = −39.3 kcal/mol) (Figure 2B), which substantially reduced CUG initiation efficiency (Figure 2C). Furthermore, we restored the base pairs by introducing compensatory mutations to the right arm (MFE = −58.5 kcal/mol) (Figure 2B), which fully restored the expression of long NPR (Figure 2C). These results indicate that the NPR mRNA secondary structure is both required and sufficient to enhance initiation at the CUG TIS in mammals.

Alternative AUG initiation converts NPR to a secreted proteoform

To begin to characterize the functional impact of alternative TIS usage, we first monitored the localization of FLAG-tagged NPR proteoforms on the cell surface of primary cortical neurons (Figure S3A). While long NPR was widely distributed across the cell surface, short NPR were concentrated in synapse-like punctate structures (Figure S3A), which may partly explain the stronger enrichment of endogenous short NPR in the synaptosome fraction (Figure 1B) than in whole-cell lysates (Figure 1C). These results were further confirmed by the localization of NPR mutants with the CUG-enhancing stem-loop structure either destabilized (expressing mostly short NPR) or restored (expressing both short and long NPRs) (Figure S3B), suggesting that the two NPR proteoforms may be differentially localized in cells.

Like other type II single-pass membrane proteins, the N-terminal transmembrane domain of CUG-initiated NPR (amino acids 3–23) presumably acts as a signal anchor to target the ribosome-mRNA-nascent peptide chain complex to the ER surface (Figure 3A). Therefore, N-terminal truncation caused by downstream AUG initiation may affect the site of translation and/or protein localization. To test whether the truncated N-terminus could still function as a signal anchor, we fused the N-terminal sequences from either the long or short NPR to GFP and monitored both cell-surface and intracellular expression. As anticipated, the intact N-terminus from long NPR was sufficient to anchor GFP localization on the outer surface in both HEK293T cells (Figure S3C) and primary hippocampal neurons (Figure 3B). In contrast, the truncated N-terminal sequence (ΔN) from short NPR resulted in no detectable GFP signal on the cell surface (Figure 3B; Figure S3C), suggesting that the truncated N-terminus can no longer function as a signal anchor.

Figure 3. Alternative AUG initiation converts NPR to a secreted proteoform.

(A) N-terminal sequence of mouse NPR, showing the transmembrane domain/signal anchor (gray) and the predicted signal peptide (blue).

(B) Surface and total GFP signals in hippocampal neurons expressing either SA- or ΔN-GFP. SA, signal anchor. Scale bar, 5 μm.

(C) Endogenous NPR expression in whole-cell lysates, membrane fraction, and culture media of primary cortical neurons. Membrane (pan-cadherin) and cytosolic (GAPDH) protein controls are shown.

(D) Expression of FLAG-tagged long and short NPRs in culture media and whole-cell lysates of primary cortical neurons.

(E) Endogenous NPR in the insoluble and soluble fractions of detergent-free extracts as well as detergent-assisted extracts from mouse hippocampus.

Instead, the truncated N-terminal sequence of short NPR was computationally predicted with high confidence to act as a cleavable signal peptide^27,28 (Figure 3A, Figure S3D), which would cause short NPR to be secreted from cells. Indeed, not only endogenous NPR was readily detectable in the neuronal culture media, but it also appeared smaller than the canonical NPR species in the membrane fraction (Figure 3C). Expression of the long proteoform alone by substituting CUG with AUG did not produce detectable NPR in the media (Figure 3E), arguing against the source of secreted NPR being the previously shown cleavage of long NPR by tumor necrosis factor-α converting enzyme (TACE)²². In contrast, expression of short NPR alone (ΔN-NPR) resulted in robust secretion (Figure 3D), confirming the prediction that N-terminal truncation caused short NPR to become a secreted factor analogous to NP1 and NP2.

To test whether short NPR was also secreted in vivo, we compared NPR expression between mouse hippocampus lysates prepared with and without adding detergents to disrupt the plasma membrane. While both long and short NPRs were detected when detergents were added, only short NPR appeared in the soluble fraction of the detergent-free extracts (Figure 3E), further supporting that downstream TIS usage remodeled the NPR signal anchor to a cleavable signal peptide and thus converted the membrane protein into a secreted factor in vivo. Notably, the transformation of a type II membrane protein signal anchor to a cleavable signal peptide was first described in mutagenesis studies more than 30 years ago²⁹. Our results suggest that NPR represents a natural example of such transformation.

Neuronal activity regulates NPR alternative TIS usage

Both NP1 and NP2 have been shown to be regulated by neuronal activity^30,31. To test whether the NPR proteoform ratio may also be regulated by neuronal activity, we treated primary cortical neurons with bicuculline, a GABA_A receptor antagonist, and monitored both intracellular and extracellular NPR proteoform expression. While long NPR expression was largely unaffected, both intracellular and secreted short NPR levels were increased after stimulation (Figure 4A). Neuronal activity-regulated NPR expression was absent in a mutant mRNA that only expressed the long proteoform (Figure S4A), arguing against the increase in short NPR being the result of post-translational cleavage of the long proteoform. KN93, a Ca2+/calmodulin-dependent protein kinase II (CaMKII) inhibitor, blocked the bicuculline-stimulated increase in short NPR translation (Figure S4B). Furthermore, this increase was abolished in the stem-loop-destabilized mRNA but restored by compensatory mutations (Figure S4C), suggesting this stem-loop structure was not only important for efficient CUG initiation under basal conditions but also required for neuronal activity-dependent regulation.

Figure 4. Secreted NPR enhances synaptic clustering of AMPA receptors in vitro.

(A) Expression of FLAG-NPR in the culture media and whole-cell lysates of primary cortical neurons after DMSO or bicuculline (40 μM, 12 hours) treatment (left). Quantifications of relative secreted NPR levels in the media (middle) and short NPR fractions in cell lysates (right) are shown. **, p<0.01; ***, p<0.001, Student’s t-tests.

(B) Soluble NPR levels in hippocampi harvested from mice with or without having explored an enriched environment. n=3 per group. *, p<0.05, Student’s t-test.

(C) Surface GluR1, PSD-95 and MAP2 immunostaining (left) in cultured hippocampal neurons expressing either FLAG (vector), long, or short NPR cDNA. Scale bar, 5 μm. Quantification of surface GluR1 puncta density is shown (right). Data were pooled from three independent experiments. ***, p<0.001, Kruskal-Wallis tests comparing short NPR to either vector or long NPR groups.

(D) Schematic illustration of the experimental procedure testing the effects of secreted NPRs in the media (left). Surface GluR1, PSD-95 and MAP2 immunostaining in NPR-depleted recipient hippocampal neurons treated with conditioned media with or without NPR depletion (middle). Scale bar, 5 μm. Quantification of surface GluR1 puncta density is shown (right). siNT, non-targeting siRNA. **, p<0.01; ns, p>0.05, one-way ANOVA, multiple comparisons test.

To monitor the activity-dependent changes in TIS usage, we performed ribosome profiling after harringtonine and puromycin treatment to enrich initiating ribosomes (Table S1)³². As expected, the 5′ ends of RPFs exhibited strong 3-nucleotide periodicity near annotated TISs (Figure S4D). Consistent with the observed increase in secreted NPR, RPFs with the inferred P sites mapping to the downstream AUG were increased relative to CUG after bicuculline stimulation (Figure S4E). In addition, RPFs extracted from translating ribosomes shifted towards the coding sequence downstream of AUG after stimulation (Figure S4F). Similar results were observed in two independent datasets from previous studies^33,34 comparing cultured neurons before and after KCl-induced depolarization (Figure S4F), further supporting that neuronal excitation increased alternative AUG TIS usage in NPR mRNA.

To test NPR proteoform expression might be regulated by physiological neuronal activity in vivo, we quantified the abundance of secreted NPR within the detergent-free soluble fraction of hippocampal lysates after the animals explored a novel and enriched environment, which has been shown to increase neuronal activity and immediate-early gene expression³⁵. Compared to controls, animals exposed to the enriched environment showed higher NPR secretion in the hippocampus (Figure 4B), suggesting that physiological neuronal activity enhanced alternative AUG TIS usage in endogenous NPR mRNA in vivo.

Secreted NPR enhances synaptic clustering of AMPA receptors in vitro

The distinct localization of long and short NPRs as well as their differential regulation by neuronal activity suggested that these two proteoforms might have distinct functions. To isolate the impact of long versus short NPRs on post-synaptic AMPA receptor clustering, we replaced endogenous NPR with individual proteoforms in primary hippocampal neurons by first knocking down endogenous NPR expression with a 3′ UTR-targeting siRNA (Figure S4B) and re-introducing siRNA-resistant long or short NPR cDNAs by lentivirus. We then monitored the synaptic clustering of AMPA receptors in primary hippocampal neurons by immunolabeling under non-permeabilized conditions and compared neurons expressing each of the two NPR proteoforms. Long NPR expression alone did not increase the density of surface GluR1 puncta (Figure 4C) but reduced their intensity (Figure S4C). In contrast, short NPR expression alone significantly increased both the density (Figure 4C) and intensity (Figure S4C) of surface GluR1 puncta. These results suggest that long and short NPRs are functionally distinct, and that short NPR can function independently of its transmembrane counterpart in promoting AMPA receptor clustering in vitro.

Considering that short NPR was secreted into the media, we asked whether the effect of short NPR on AMPA receptor clustering required direct cell-cell contact (i.e., synaptic connections), or it could be mediated by diffusion. We first collected control and secreted NPR-depleted conditioned media from primary neurons treated with either control or NPR siRNAs, respectively, and then added these conditioned media to new batches of NPR-depleted neurons (Figure 4D). Neurons treated with control media showed higher surface GluR1 puncta density than those treated with either fresh or NPR-depleted conditioned media (Figure 4D), indicating that short NPR could promote AMPA receptor clustering at least partly through secretion and diffusion, independent of direct cell-cell contact.

Altered NPR proteoform ratio reduces GluR4 in PV neurons in vivo and affects behaviors

Previous studies have shown that neuronal pentraxins play crucial roles in the formation and organization of excitatory synapses in parvalbumin (PV)-positive interneurons^20,36. While knocking out both NP2 and NPR in mice causes a nearly complete loss of GluR4-containing AMPA receptors in PV neurons in the hippocampus²⁰, conventional gene knockout studies cannot address proteoform-specific contributions. To specifically alter proteoform ratio without changing the overall NPR abundance, we took a proteoform-specific targeting strategy based on our finding that alternative initiation at downstream AUG required leaky scanning at the CUG TIS (Figure 1D). Using CRISPR/Cas9-based homologous recombination, we generated knock-in (KI) mice carrying one or both Nptxr alleles with the CTG TIS replaced by ATG. Both heterozygous (KI/+) and homozygous (KI/KI) ATG KI mice were born at expected Mendelian ratios and did not display gross physical abnormalities. Consistent with the expectation that substituting the CUG TIS with an AUG should effectively block leaking scanning, KI/+ and KI/KI mice expressed progressively less short NPR (Figure 5A), whereas total NPR abundance remained unchanged. When no detergent was added during lysate preparation, only short NPR was detected in the soluble fractions of both wild-type (+/+) and KI/+ but not KI/KI forebrains (Figure 5A), further supporting alternative TIS usage being the primary source of secreted NPR in vivo. To examine the impact of altered NPR proteoform ratio in vivo, we first quantified the abundance each of the four AMPA receptor subunits (GluR1–4) by proteomics. Consistent with the previously reported role of neuronal pentraxins on GluR4⁺ AMPA receptors in PV neurons^18,20, GluR4 abundance was more affected than other subunits in the KI/KI forebrains (Figure S5A). To validate these measurements, we performed immunohistochemistry for GluR4, which was, as expected, highly enriched in PV neurons (Figure 5B). GluR4 levels were significantly reduced in PV neurons in the hippocampal CA1 region of KI/KI mice compared to +/+ littermates (Figure 5B), suggesting that the NPR proteoform ratio indeed plays a role in GluR4⁺ AMPA receptor expression in these cells. Consistent with the reduced excitation of PV neurons, perineuronal net density measured by wisteria floribunda agglutinin (WFA) labeling was also reduced in KI/KI mice (Figure S5B).

Figure 5. Altered NPR proteoform ratio reduces GluR4 in PV neurons in vivo and affects behaviors.

(A) Endogenous NPR expression in the insoluble and soluble fractions of detergent-free hippocampal extracts from wild-type (+/+), heterozygous (KI/+), and homozygous (KI/KI) ATG knock-in mice. Long NPR fractions, total and secreted NPR levels were quantified.

(B) Immunohistochemistry (left) and quantification (right) of GluR4 in PV neurons in hippocampal CA1 regions of adult (P30) ATG KI mice. Scale bar, 20 μm. n=7–10 animals per genotype. *, p<0.05; ns, p>0.05, one-way ANOVA, multiple comparisons test.

(C) Spontaneous locomotor activity. n=5–6 animals per genotype. ns, p>0.05, two-way ANOVA.

(D) Frequencies of spontaneous alternations in a Y-maze. n=7–8 animals per group *, p<0.05, Mann-Whitney test.

(E) Freezing responses to tone preceding shock during training session (left) and to tone presentation during cued memory test (right). n=3–5 animals per genotype. *, p<0.05 interaction; ns, p>0.05, two-way repeated measures ANOVA.

PV neurons are broadly involved in cognitive functions and behaviors, with a particularly important role in working memory^37,38. We therefore tested these KI mice in several behavioral paradigms that measure baseline function and different forms of short- and long-term memory. While animals of all three genotypes showed similar ambulatory locomotor activity (Figure 5C), KI/KI mice made fewer spontaneous alternations in a Y-maze compared to +/+ littermates (Figure 5D), suggesting that KI/KI mice exhibited impaired working memory. In a separate paradigm, KI/KI mice displayed increased within-session freezing to a shock-paired tone after the first tone-shock pairing (Figure 5E), suggesting that loss of NPR proteoform diversity increased the sensitivity to a stressful stimulus and enhanced short-term associative fear learning. Compared to the immediate enhancement, the difference in freezing in response to the tone alone 48 hours later was diminished, suggesting that long-term fear memory was less affected (Figure 5E). The increase in short-term tone-fear reactivity and the reduction of spontaneous alternations in a Y-maze may be due to a differential role for PV neuron activity in fear learning³⁹ versus working memory³⁷. Taken together, these results indicated that a large change in NPR proteoform ratio caused both cellular and behavioral changes in vivo, supporting the notion that NPR proteoform diversity is necessary for the proper development of PV neuronal connectivity as well as cognitive behaviors that rely on intact PV neuron function.

N-terminal extension by upstream alternative TISs converts secreted C1QLs to transmembrane proteoforms

Having observed the signal anchor-to-signal peptide conversion through the N-terminal truncation of NPR, we reasoned that the reverse conversion might also occur when an upstream, in-frame alternative TIS extended the signal peptides of known secreted factors. Among the candidate proteins previously predicted^9,10 by sequence analysis to have conserved N-terminal extensions are the complement component 1, q subcomponent-like proteins (C1QLs), which have been shown to regulate the formation and maintenance of excitatory synapses^40,41. In contrast to NPR, C1QLs had been known as secreted synaptic organizers indirectly anchored at the presynaptic termini through one or more membrane proteins such as neurexins⁴¹. They bind to several postsynaptic partners including the cell-adhesion G protein-coupled receptor BAI3⁴² and kainate receptors⁴¹. Intriguingly, all four C1QL paralogs have PhyloCSF candidate coding regions⁴³ with high ratios of synonymous to non-synonymous mutations indicative of conserved protein-coding functions immediately upstream of their annotated AUG TISs, each proceeded by a conserved AUU codon (Figure 6A).

Figure 6. N-terminal extension by upstream alternative TISs converts secreted C1QLs to transmembrane proteoforms.

(A) N-terminal region of C1QL2, showing the upstream AUU (red), PhyloCSF candidate coding region, predicted signal anchor and signal peptide.

(B) Ectopic expression of WT C1QL1/2/3 cDNAs in HEK293T cells.

(C) Expression of C1QL2 mutant cDNAs in HEK293T cells. Mutations are indicated in red. ΔN begins at the annotated AUG TIS.

(D) Expression of long and short C1QL2 proteoforms in the membrane fractions and culture media of primary cortical neurons.

(E) Surface (non-permeabilized) and total (permeabilized) GFP immunostaining in HEK293T cells expressing GFP carrying the putative signal anchor (SA) or signal peptide (SP) from long and short C1QL2, respectively. Scale bar, 20 μm.

(F) Surface and total GFP immunostaining in primary hippocampal neurons expressing GFP carry the putative SA or SP from long and short C1QL2, respectively. Scale bar, 5 μm.

Consistent with a previous study detecting multiple C1QL2/3 protein species by immunoblotting⁴¹, ectopically expressed C1QL1/2/3 along with their native 5′ UTRs each produced two proteoforms (Figure 6B). The apparent molecular weights of C1QL2 proteoforms were similar to those of two mutants in which either the AUU codon was substituted by AUG or the putative N-terminal extension was deleted (ΔN) (Figure 6C). Either substituting AUU with CUA or changing the AUU-flanking sequence to an anti-Kozak context eliminated the expression of long C1QL2 proteoform (Figure 6C), confirming the AUU codon as the upstream alternative TIS. Like the CUG-downstream sequence in NPR, AUU-downstream sequences of all four C1QLs were also predicted to form stable RNA secondary structures (Figure S6A).

Consistent with C1QLs being known as secreted proteins, all four N-terminal sequences of AUG-initiated short C1QL proteoforms were predicted by Phobius as cleavable signal peptides (Figure S6B). In contrast, all four AUU-initiated long C1QL proteoforms were predicted to have N-terminal transmembrane domains (Figure S6B), suggesting potential conversions from signal peptides to signal anchors. To test these predictions, we expressed each of the two C1QL2 proteoforms in primary cortical neurons. While short C1QL2 was mostly secreted into the media as expected, long C1QL2 was predominantly membrane-bound (Figure 6D). Furthermore, the extended N-terminal sequence of long C1QL2, but not the canonical N-terminal sequence of short C1QL2, was sufficient to direct the cell-surface expression of GFP both in HEK293T cells (Figure 6E) and in primary cortical neurons (Figure 6F), consistent with in silico prediction of a signal peptide-to-signal anchor conversion.

Prediction of additional alternative TIS-mediated proteoform conversions

To explore the broader scope of this mechanism, we asked whether alternative TIS-mediated signal anchor/signal peptide conversions could be found in other mRNAs beyond those encoding synaptic organizers. Using previously compiled lists of alternative TISs in human⁷ and mouse cells⁵ as well as those identified using our TIS mapping data (Table S1), we applied Phobius^27,28 to predict signal peptides and transmembrane domains for each annotated N-terminal sequence and its alternative TIS-associated variant. This analysis identified 39 secreted proteins with cleavable signal peptides that were predicted to acquire a transmembrane domain through upstream TIS-mediated N-terminal extension, as well as 23 proteins with single predicted transmembrane domains that were changed to cleavable SPs upon N-terminal truncation (Table S2). These putative N-terminal proteoforms included an additional C1q-related factor (C1QTNF1) as well as those in a wide variety of functional categories including extracellular matrix components (e.g., laminin), signaling (e.g., WNT3/5A, gremlin) and cell adhesion molecules. Ectopic expression of some of these predicted candidates in HEK293T cells indeed yielded multiple proteoforms (Figure S6C). These results, together with those from NPR and C1QLs, indicated an unexpected level of plasticity of N-terminal signal sequences tuned by alternative TIS usage, which enabled dual encoding of secreted and transmembrane proteoforms (Figure S6D).

DISCUSSION

In this study, we show that a conserved CUG TIS in NPR mRNA enables alternative translation initiation at the downstream AUG TIS, yielding an N-terminal truncated proteoform with distinct subcellular localization and synaptic functions. To understand the functional impact of NPR proteoform diversity, we generated an ATG KI mouse model in which NPR proteoform ratio, but not overall abundance, was altered. Despite the redundancy between neuronal pentraxins^20,44 and a variety of other synaptic organizers⁴⁵, the imbalance between long and short NPR proteoforms, caused by a single-nucleotide substitution (C-to-A) reduced GluR4+ AMPA receptors in hippocampal PV interneurons. Concomitantly, the loss of NPR proteoform diversity affected cognitive functions, with mice expressing predominantly long NPR exhibiting sensitized short-term associative fear learning and impaired working memory measured by spontaneous alternations in a Y-maze.

In stark contrast to the widely accepted notion that nearly all multi-exon genes produce alternatively spliced mRNA isoforms, each mRNA has been typically considered to encode a single protein. Reflecting this conventional view, most if not all current genome annotations assign a single TIS and a single stop codon to each transcript isoform. In the past decade, this view has been challenged by high-throughput translatomics studies based on ribosome profiling, especially those that specifically enrich initiating ribosomes to globally map TISs across the transcriptome^5,32. Instead of finding one TIS per mRNA, these studies have repeatedly shown that alternative TIS usage is prevalent across the transcriptome, with more than half of the detected transcripts having more than one TIS. Using these expanded TIS annotations, we searched for additional N-terminal signal sequence switching events analogous to NPR and the C1QL proteins. Indeed, we were able to identify dozens of other mRNAs encoding distinctly localized proteoforms, hinting at a broad impact of alternative TIS usage on protein localization. However, current TIS mapping methods based on ribosome profiling are affected by multiple sources of false positives and false negatives. For example, to reduce RPFs from actively translating ribosomes, these experiments typically involve treating cells with translation inhibitors to allow translating ribosome to run off. Over the duration of this treatment, new initiation events can occur and accumulate at positions that may not represent physiological TIS usage. To partly address this caveat, QTI-seq enriched initiating ribosomes after cell lysis, thereby minimizing new rounds of initiation³². Puromycin was added to induce the release of the nascent polypeptide chain and the translating ribosome, which reduces RPFs within coding sequences. Nonetheless, ribosome profiling-predicted alternative TISs will require validation by orthogonal approaches, such as mass spectrometry-based N-terminal proteomics⁸ or, in the cases of NPR and C1QLs, evolutionary conservation (PhyloCSF), protein expression analysis, and mutagenesis studies.

The biogenesis of transmembrane and secreted proteins heavily relies on their N-terminal signal sequences. Type II single-pass transmembrane proteins use an N-terminal signal anchor that also functions as a transmembrane domain, whereas secreted proteins have cleavable signal peptides. While both types of signal sequences induce ER targeting of the mRNA-ribosome-nascent peptide chain complex through the SRP complex, these proteins enter unique downstream processing pathways, possibly at different subdomains of the ER⁴⁶. The processing of signal anchors versus signal peptides requires distinct ER-resident factors capable of detecting the differences between the two types of signals. For instance, the nascent polypeptide chain of a type II transmembrane protein is inserted into the ER membrane head-first and undergoes a signal anchor inversion within the ER-Sec61α complex, bypassing signal peptidase-dependent cleavage of signal peptides of secreted proteins. In the context of bifunctional mRNAs that produce both transmembrane and secreted proteoforms, how these distinct pathways may be spatially coordinated at the translation site on the ER surface remains an open question.

Another outstanding question regarding mRNAs with multiple TISs is whether all TISs are used in each mRNA copy. Translation of a single copy of NPR mRNA, for example, may be initiated at either CUG, AUG, or both TISs. Single-molecule imaging-based analysis of mRNA translation may elucidate the mechanistic details underlying alternative TIS selection⁴⁷. Considering the distinct morphology of neurons characterized by a high degree of compartmentalization, TIS selection of many neuronal mRNAs may be impacted by their subcellular localization⁴⁷. In addition, our analysis of NPR cDNA expression across different cell types indicated a high long-to-short NPR proteoform ratio in neurons. While mouse NPR mRNA expressed mostly the short proteoform in rabbit reticulocyte lysate, both endogenous and ectopically expressed NPR mRNA produced a larger fraction of long proteoform in primary neurons, suggesting that one or more cellular factors that regulate alternative TIS usage, such as eIF1, eIF1A, and eIF5, may be differentially expressed between cell types. Considering that efficient initiation at the CUG TIS requires a highly stable downstream stem-loop structure, the folding equilibrium of the structure as well as its dynamic unwinding during ribosome scanning may also be regulated by cellular factors differentially expressed between cell types. Furthermore, our findings on the activity-dependent TIS usage in NPR mRNAs provided another example for the selection between alternative TISs being regulated by cell states. Unlike previous studies showing global changes in TIS selection (e.g., the preference of AUG over non-AUG TISs) during mitosis^15,48, starvation³², and integrated stress response^49,50, the activity-dependent increase in AUG initiation appeared to be specific to NPR mRNA, hinting at the involvement of additional cis-regulatory elements in this regulation. Indeed, the same stem-loop structure that promotes CUG initiation was found to be also required for the activity-dependent regulation. This result is reminiscent of a recent study showing that RNA structures downstream of alternative, regulatory AUGs broadly mediate immune challenge-induced translation events in plants⁵¹. It will be interesting to identify the trans-acting factors that mediate the structure-dependent translation control and test whether other mRNAs with TIS usage influenced by stable secondary structures may be similarly regulated by neuronal activity.

A recent study has shown that N-terminal truncations widely affect protein localization in yeast⁵², suggesting deep evolutionary conservation of the plasticity of N-terminal signal sequence regulated by alternative TISs. Beyond N-terminal signal sequences, alternative translation can conceivably impact protein functions in diverse manners. For instance, alternative TISs of upstream ORF (uORF) and out-of-frame TISs can regulate the amount of functional protein output. In addition, alternative TISs may add or subtract critical residues for post-translational modification or protein-protein interactions potentially impactful on protein folding and catalytic functions. Therefore, our findings on alternative translation initiation, together with other translational recoding mechanisms such as ribosomal frameshifting and stop-codon read-through, hint at a vastly underexplored contribution of alternative translation events in diversifying the functional output of the transcriptome.

Limitations of the study

In order to specifically alter the relative proteoform ratio while keeping the overall NPR abundance constant, we generated a CTG-to-ATG KI mouse model to shift the relative ratio towards long NPR. As such, while the cognitive and synaptic phenotypes observed in these mice support the significance of the proper proteoform ratio, the underlying mechanism remains to be further investigated. In one scenario, these phenotypes may result from either the near-complete loss of short NPR and/or the overexpression of the long NPR. In a separate scenario, the proteoform ratio may represent the correct subunit stoichiometry critical for the formation of functional heteromeric complexes either between long and short NPRs or between NPR and NP1/NP2, as previously reported¹⁹. Lastly, a previous study has shown that (long) NPR can undergo TACE-mediated cleavage²², resulting in secreted fragments of NPR, some of which resemble short NPR produced by alternative TIS usage in both form and function. While we did not quantify the level of long NPR cleavage in primary neurons, changes in subunit stoichiometry could conceivably affect the efficiency of TACE cleavage. Future in vitro and in vivo studies with additional proteoform-specific perturbations, including both gain- and loss-of-function mutations of each TIS, should help disentangle these possibilities.

STAR METHODS

Resource availability

Lead contact

Requests for further information or reagents should be directed to the lead contact, Junjie Guo (junjie.guo@yale.edu).

Materials availability

All reagents generated in this study are available upon request from the lead contact with a completed Materials Transfer Agreement.

Data and code availability

• High-throughput sequencing data reported in this paper have been deposited in Gene Expression Omnibus (GEO) under the accession number GSE255191 and are publicly available as of the date of publication. High-throughput sequencing data from previously published studies, including those by Glock et al., Duffy et al., and Hacisuleyman et al., can be accessed via NCBI BioProject (PRJNA634994) and GEO (GSE180240 and GSE213083), respectively. Original Western blot and microscopic images are available at Mendeley Data and are publicly available as of the date of publication. The DOI is listed in the key resources table. All remaining data reported in this manuscript will be shared by the lead contact upon request.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse NPR	Santa Cuz	Cat# sc-39008
Chicken GFP	Aveslab	Cat# GFP-1020
Rabbit GFP	Thermo Fisher Scientific	Cat# A-6455
Mouse FLAG	Sigma-Aldrich	Cat# F1804
Rabbit FLAG	Rockland	Cat# 600-401-383
Rabbit GAPDH	Sigma-Aldrich	Cat# G9545
Mouse Pan-cadherin	Sigma-Aldrich	Cat# C1821
Mouse Vinculin	Sigma-Aldrich	Cat# V9264
Rabbit GluR1	Sigma-Aldrich	Cat# ABN241
Rabbit GluR4	Sigma-Aldrich	Cat# AB1508
Chicken MAP2	Synaptic Systems	Cat# 188 006
Mouse PSD-95	Sigma-Aldrich	Cat# MAB1596
Guinea pig Parvalbumin	Synaptic Systems	Cat# 195 004
Bacterial and virus strains
5-alpha Competent E.coli	NEB	C2987H
Stable Competent E.coli	NEB	C3040H
Chemicals, peptides, and recombinant proteins
Bicuculline	Tocris	Cat# 0130
KN-93	Sigma-Aldrich	Cat# 422708
Natural Mouse Laminin	Life Technologies	Cat# 23017015
Poly-DL-ornithine hydrobromide	Sigma-Aldrich	Cat# P0421
Wisteria Floribunda (Japanese Wisteria) Lectin (WFL), fluorescein (FITC)	Thermo Fisher Scientific	Cat# L32481
DMEM, high glucose	Gibco	Cat# 11965092
Fetal Bovine Serum (FBS)	Gibco	Cat# 16140071
Lipofectamine 2000	Invitrogen	Cat# 11668019
OptiMEM	Gibco	Cat# 31985062
NuPAGE® MOPS SDS Running Buffer	Life Technologies	Cat# NP0001
NuPAGE® LDS Sample Buffer	Life Technologies	Cat# NP0007
NuPAGE™ Transfer Buffer	Thermo Fisher Scientific	Cat# NP00061
HBSS	Life Technologies	Cat# 14170112
Dulbecco’s PhosphateBuffered Saline (DPBS)	Life Technologies	Cat# 14190144
Tris Buffered Saline	Bio-Rad Laboratories	Cat# 1706435
Neurobasal	Life Technologies	Cat# 21103049
L-Glutamax	Thermo Fisher Scientific	Cat# 35050061
B27 Plus	Thermo Fisher Scientific	Cat# A3582801
Cycloheximide	Sigma-Aldrich	Cat# C7698
Puromycin dihydrochloride	Sigma-Aldrich	Cat# P8833
Homoharringtonine	Sigma-Aldrich	Cat# SML1091
RNase I	Thermo Fisher Scientific	Cat# EN0602
SUPERase·In™ RNase Inhibitor	Thermo Fisher Scientific	Cat# AM2696
RNasin® Plus Ribonuclease Inhibitor	Promega	Cat# N2615
Novex™ TBE-Urea Sample Buffer (2X)	Thermo Fisher Scientific	Cat# LC6876
Novex™ TBE-Urea Gels, 15%	Thermo Fisher Scientific	Cat# EC68855BOX
Novex™ TBE-Urea Gels, 10%	Thermo Fisher Scientific	Cat# EC68755BOX
TBE Buffer (Tris-borate-EDTA) (10X)	Thermo Fisher Scientific	Cat# B52
Sodium creatine phosphate dibasic tetrahydrate	Sigma-Aldrich	Cat# 27920
Spermidine	Sigma-Aldrich	Cat# S0266
Creatine Phosphokinase, Rabbit Skeletal Muscle	Sigma-Aldrich	Cat# 2384
ATP	NEB	Cat# P0756S
TRIzol™ Reagent	Thermo Fisher Scientific	Cat# 15596018
Critical commercial assays
GFP-Trap magnetic agarose beads	ChromoTek	Cat# gtma
Anti-FLAG® M2 magnetic beads	Sigma-Aldrich	Cat# M8823
Sheep anti-mouse antibody (M280)-conjugated dynabeads	Thermo Fisher Scientific	Cat# 11201D
KAPA HiFi HotStart plus dNTPs	Roche	Cat# 7958897001
Q5 High-Fidelity DNA Polymerase	NEB	Cat# M0491S
NEBuilder HiFi DNA Assembly Master Mix	NEB	Cat# E2621S
Rabbit reticulocyte lysate system	Promega	Cat# L4960
HiScribe T7 ARCA mRNA Kit	NEB	Cat# E2060
Protein Deglycosylation Mix II	NEB	Cat# P6044S
AAV extraction kit	Takara	Cat# 6666
Lysing Matrix-D	MP Biomedicals	Cat# 116913100
Qubit RNA High Sensitivity Assay Kit	Thermo Fisher Scientific	Cat# Q32855
Human-Mouse-Rat Ribo-Seq riboPOOL	TOOLs BIOTECH	Cat# dp-P024-50
T4 Polynucleotide Kinase	NEB	Cat# M0201L
T4 RNA Ligase 2, truncated KQ	NEB	Cat# M0373L
T4 RNA Ligase 1	NEB	Cat# M0437M
SuperScript IV Reverse Transcriptase	Thermo Fisher Scientific	Cat# 18090050
ExoSAP-IT™ PCR Product Cleanup Reagent	Thermo Fisher Scientific	Cat# 78200.200.UL
Dynabeads™ MyOne™ Silane	Thermo Fisher Scientific	Cat# 37002D
Deposited data
Ribosome profiling of mouse cortical neurons treated with DMSO or bicuculline	This study	GEO: GSE255191
Western blots and microscopy images	This study	10.17632/8m3psfj729.1
Ribosome profiling of rat primary neurons treated with harringtonine for 150 seconds	Glock et al.23	BioProject: PRJNA634994
Ribosome profiling of human NGN2 neurons treated with DMSO or KCl	Duffy et al.33	GEO: GSE180240
Ribosome profiling of mouse cortical neurons treated with DMSO or KCl	Hacisuleyman et al.34	GEO: GSE213083
Experimental models: Cell lines
HEK293T	This paper	N/A
N2A	This paper	N/A
Primary cortical and hippocampal neurons	This paper	N/A
Experimental models: Organisms/strains
CTG-to-ATG knock-in mice	This paper	N/A
Wildtype C57/Bl6 mice	Charles River Laboratories	N/A
Oligonucleotides
Accell Mouse Nptxr siRNA targeting the 3ʹ UTR of NPR mRNA (5ʹ-CTTGCAAACTGAATTCCTA-3ʹ)	Dharmacon	A-046750-16-0050
Accell Non-targeting Control siRNA	Dharmacon	D-001910-01-50
Recombinant DNA
AAV2-hSyn-3xFLAG	This paper	N/A
AAV2-hSyn-CTG…ATGCTG Wildtype Nptxr-3xFLAG	This paper	N/A
AAV2-hSyn-ATG…ATGCTG Long Nptxr-3xFLAG	This paper	N/A
AAV2-hSyn-N terminal deletion (ΔN) ATGCTG Nptxr-3xFLAG	This paper	N/A
AAV2-hSyn-CTA…ATGCTG Nptxr-3xFLAG	This paper	N/A
AAV2-hSyn-CTG…GCGCTG Nptxr-3xFLAG	This paper	N/A
AAV2-hSyn-CTG…GCGCTC Nptxr-3xFLAG	This paper	N/A
AAV2-hSyn-Destabilized Nptxr-3xFLAG	This paper	N/A
AAV2-hSyn-Compensated Nptxr-3xFLAG	This paper	N/A
AAV2-hSyn-Signal anchor Nptxr-3xFLAG	This paper	N/A
AAV2-hSyn-Signal peptide (ΔN) Nptxr-3xFLAG	This paper	N/A
AAV2-GFP	This paper	N/A
AAV2-hSyn-Signal anchor C1ql2-GFP	This paper	N/A
AAV2-hSyn-Signal peptide (ΔN) C1ql2-GFP	This paper	N/A
AAV2-DJ	This paper	N/A
AAV2-Helper	This paper	N/A
pCMV-Human NPTXR1-135-GFP	This paper	N/A
pCMV-Mouse NPTXR1-135-GFP	This paper	N/A
pCMV-X.tropicalis NPTXR1-135-GFP	This paper	N/A
pCMV-D.rerio NPTXR1-135-GFP	This paper	N/A
pCMV-Wildtype mouse Nptxr-GFP	This paper	N/A
pCMV-Destabilized Nptxr-GFP	This paper	N/A
pCMV-Compensated Nptxr-GFP	This paper	N/A
pCMV-Signal anchor Nptxr-GFP	This paper	N/A
pCMV-Signal peptide (ΔN) Nptxr-GFP	This paper	N/A
pCMV-Wildtype C1ql1-GFP	This paper	N/A
pCMV-ACCATTACC…ATG Wildtype C1ql2-GFP	This paper	N/A
pCMV-ACCCTAACC…ATG C1ql2-GFP	This paper	N/A
pCMV-TTTATTCCC…ATG C1ql2-GFP	This paper	N/A
pCMV-ACCATGACC…ATG Long C1ql2-GFP	This paper	N/A
pCMV-N terminal deletion (ΔN) ATG Short C1ql2-GFP	This paper	N/A
pCMV-Wild type C1ql3-GFP	This paper	N/A
pCMV-Signal anchor C1ql2-GFP	This paper	N/A
pCMV-Signal peptide (ΔN) C1ql2-GFP	This paper	N/A
pCMV-WNT5A-GFP	This paper	N/A
pCMV-WNT3-HA	This paper	N/A
pCMV-LAMB2-HA	This paper	N/A
pCMV-ST3GAL4-HA	This paper	N/A
pCMV-CLN5-HA	This paper	N/A
pCMV-CYB5R3-HA	This paper	N/A
Software and algorithms
Phobius	Käll et al., 200427; Käll et al., 200728	https://phobius.sbc.su.se/
RNAfold	Gruber et al.26	http://rna.tbi.univie.ac.at/cgi-bin/RNAWebSuite/RNAfold.cgi
ImageJ	Schneider et al.53	https://imagej.net/ij/
STAR	Dobin et al.54	https://github.com/alexdobin/STAR
Cutadapt	Marcel Martin, 201155	https://cutadapt.readthedocs.io/en/stable/
RiboMiner	Li et al.56	https://github.com/xryanglab/RiboMiner
Samtools	Danecek et al.57	https://www.htslib.org/
fastx_toolkit	FASTA/FASTQ files preprocessor	https://github.com/agordon/fastx_toolkit
MATLAB	MathWorks	https://www.mathworks.com/products/matlab.html
RStudio	R	https://posit.co/products/open-source/rstudio/
GraphPad Prism	GraphPad Software	www.graphpad.com
Other
Predicted proteoform conversions by alternative translation initiation in Table S1.	This paper	N/A
Predicted alternative TISs in primary cortical neurons in Table S2.	This paper	N/A

Experimental Model and Study Participant Details

Cell lines

HEK293T and N2A cells were cultured in DMEM with 10% fetal bovine serum.

Primary neuronal culture

Pregnant female C57BL/6J mice (Charles River) at 18 days gestation were anesthetized with isoflurane for 5 min in a chamber and subsequently sacrificed by cervical dislocation. Male and female E18 embryos were then extracted and transferred to a sterile dish containing HBSS. Brain regions of interest, such as cortices or hippocampi, were dissected, removing any surrounding tissue and meninges. The tissues were digested in 5 mL of trypsin-EDTA solution and incubated at 37°C for 10–15 min, followed by gentle titration in DMEM supplemented with 10% FBS (Thermo Fisher). Neurons were pelleted and resuspended in Neurobasal medium supplemented with B27 Plus (Thermo Fisher, A3582801, 50X), L-Glutamax (Thermo Fisher, 35050061, 1%), 33 mM glucose, and 37.5 mM NaCl. Dissociated neurons were plated on coverslips or dishes coated with laminin and poly-ornithine at the desired density. The medium was changed every 3–5 days, and 5 μM AraC was added on day 1 in vitro (DIV1) for 48 hours to inhibit glial proliferation.

Animals

All experimental protocols involving animals were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) at Yale University (Protocol #2024–20207). Animal care and housing were provided by the Yale Animal Resource Center (YARC). The animals were maintained in a 12-hour light/dark cycle with ad libitum access to food and water. ATG KI mice carrying one or both Nptxr alleles with the CTG-to-ATG codon replacement were generated by Yale Genome Editing Center using CRISPR/Cas9-based homologous recombination. Sequencing-validated founder animals were backcrossed to WT C57BL6/J animals (Charles River) to establish a homogenous genetic background. The mutant line was subsequently maintained through het-to-het breeding. Adult (>P30) male littermates were used for behavioral and immunohistochemical experiments.

Method details

Plasmid and viral constructs

For GFP reporter experiments, the following N-terminal sequences were inserted to N-terminus of GFP: 1) NPR-SA: MKFLAVLLAAGMLAFLGAVICIIASVPLAASPARALPGGTDNASA; 2) NPR-SP: MLAFLGAVICIIASVPLAASPARALPGGTDNASA; 3) C1QL2-SA: MTSPGGKEELVAVASRLWQRRRRACLAAVGVLLAMALGLLIAVPLLLQAAPPGAAHY E; 4) C1QL2-SP: MALGLLIAVPLLLQAAPPGAAHYE.

For AAV preparation, HEK293T cells were cultured in DMEM supplemented with 10% fetal bovine serum. Cells were transfected with pAAV-hSyn plasmid containing the cDNA of interest along with the AAV-DJ and pHelper constructs in a T175 flask for 48 h. AAV-producing cells were dissociated by incubation with 0.5 M EDTA (pH 8.0) at room temperature for 10 min. Harvested cells were then centrifuged at 2,000 g for 10 min at 4°C to remove the supernatant. Viruses were extracted and purified by using an AAV extraction kit (Takara).

AAV transduction of primary neurons

For NPR replacement experiments, primary neurons were treated with 1 μM siRNA (Dharmacon: A-046750–16-0050) targeting the 3′ UTR of NPR mRNA (5′-CTTGCAAACTGAATTCCTA-3′) at DIV2. NPR siRNA was supplemented at every medium change. Neurons were transduced with AAV for overexpression of WT or mutant NPR cDNAs simultaneously with siRNA treatment. Neurons were allowed to mature until DIV14–16 for further analyses.

In vitro translation

In vitro translation was performed using a rabbit reticulocyte lysate system (Promega, L4960). Capped and polyadenylated reporter mRNAs containing the first 135 nucleotides of NPR coding sequences (human, mouse, frog, or zebrafish), followed by the GFP sequence were generated by using the HiScribe T7 ARCA mRNA Kit (New England Biolabs, E2060). 1 μg mRNA was first denatured at 65°C for 3 minutes, and then added to the rabbit reticulocyte lysate reaction mix. The translation reactions were incubated for 90 minutes at 37 °C for human and mouse, 25°C for frog, and 28.5°C for zebrafish reporter mRNA. The translation products were analyzed by immunoblotting.

Subcellular fractionation

Cells were harvested in ice-cold subcellular fractionation buffer (20 mM HEPES, pH 7.4, 10 mM KCl, 2 mM MgCl₂, 1 mM EDTA, 1 mM EGTA, 1 μM DTT, supplemented with a cocktail of protease inhibitors) and incubated on ice for 15 min before being passed through a 26-gauge needle 10 times and kept on ice for additional 20 min. The cell lysates were first centrifuged at 720 g for 5 min at 4°C to pellet the nuclei, then further centrifuged at 10,000 g for 5 min at 4°C to pellet the mitochondria. The remaining supernatant was ultracentrifuged at 100,000 g for 1 h at 4°C. The pellet containing the membrane fraction was washed in 2M KCl for 1 h at 4°C to remove peripheral membrane proteins and subsequently ultracentrifuged for another 1 h at 4°C. The membrane pellet was finally resuspended in TBS containing 0.1% SDS.

Brain tissue fractionation

Freshly harvested brains from adult mice were suspended in ice cold Buffer A (320 mM sucrose, 10 mM HEPES pH 7.4, 1 μM DTT, supplemented with a cocktail of protease inhibitors). Tissues were homogenized using a glass Dounce tissue grinder. Part of the homogenates were saved as whole-brain lysates. The rest of the homogenates were centrifuged at 13,500 rpm for 5 min at 4°C. Subsequently, the supernatant was further ultracentrifuged at 40,000 rpm for 1 hour at 4°C. The resulting supernatant was collected as the soluble fraction, and the pellet was resuspended in RIPA buffer and analyzed as the insoluble fraction.

For synaptosomes, tissue homogenates were centrifuged at 800 g for 10 min at 4°C. The supernatant was further centrifuged at 9,000 g for 15 min at 4°C. The pellet was resuspended and washed in ice-cold Buffer A and centrifuged again at 9,000 g for 15 min at 4°C to obtain crude synaptosomes.

To remove glycans from proteins, brain lysates were first incubated with Deglycosylation Mix Buffer 2 (NEB) at 75°C for 10 min. After cooling down to room temperature, Protein Deglycosylation Mix II (NEB) was added and further incubated for 30 min. The mixture was then transferred to a shaker and incubated at 37°C for 1 h.

Western blotting

Cultured cells were lysed in RIPA buffer on ice for 10 min. After 10 min centrifugation at 4°C, 20,000 g, whole-cell lysates were mixed with 4X LDS sample buffer (Thermo Fisher) and denatured at 95°C for 5 min. Samples were loaded on a 4–12% Bis-Tris-SDS-PAGE gel, run at 120 V for 2 h 30 min in MOPS buffer, and transferred onto a nitrocellulose membrane (Bio-Rad) at 15V for 55 min. After 1 h blocking with 5% nonfat dry milk in TBS-T, the membrane was incubated with primary antibodies against protein of interest (NPR, Santa Cruz, sc-39008, 1:100; GFP, Aveslab, GFP-1020, 1:5,000; FLAG, Sigma, F1804, 1:1,000; FLAG, Rockland, 600-401-383, 1:1,000; GAPDH, Sigma, G9545, 1:2,000; Pan-Cadherin, Sigma, C1821, 1:1,000; Vinculin, Sigma, V9264, 1:5,000) diluted in 5% milk/TBS-T, placed on a shaker overnight at 4°C. After incubation, membranes were washed three times with TBS-T, followed by incubation with IR680- or IR800-conjugated secondary antibodies (Li-Cor, 1:5,000) diluted in 5% milk/TBS-T at room temperature for 1 h. Subsequently, membranes were washed with TBS-T and imaged using Odyssey XF system (Li-Cor, 2800).

Immunoprecipitation

For immunoprecipitation of endogenous NPR, culture medium of primary cortical neurons was first centrifuged at 1,000 g for 5 min at 4°C to remove cell debris, and then incubated with an NPR antibody (Santa Cruz, sc-39008, 1:50) overnight at 4°C in a mini-rotator. Samples were then incubated with pre-washed sheep anti-mouse antibody (M280)-conjugated Dynabeads (Thermo Fisher, 11201D) for 1 h at room temperature. Subsequently, the beads were washed three times with PBS supplemented with 0.1% bovine serum albumin and 2 mM EDTA, pH 7.4. Bound proteins were eluted by boiling at 95°C in SDS sample buffer for 10–15 min and were analyzed by western blotting.

For FLAG- and GFP-tagged proteins, culture medium was incubated with either anti-FLAG antibody (M2)-conjugated magnetic beads (Sigma, M8823) or GFP-Trap magnetic agarose affinity beads (Chromotek, gtma-20) overnight at 4°C in a mini-rotator. The beads were washed three times with the appropriate washing buffer (TBS for Anti-FLAG M2 Magnetic beads; 10 mM Tris/Cl pH 7.5, 150 mM NaCl, 0.5 mM EDTA for GFP-Trap magnetic agarose), followed by boiling in SDS sample buffer at 95°C for 10–15 min to elute the bound proteins. Samples were then analyzed by western blotting.

Perfusions and immunohistochemistry

Animals were deeply anesthetized with isoflurane in a chamber and then transcardially perfused with PBS followed by 4% paraformaldehyde (PFA) in PBS. Brains were dissected, postfixed for overnight at 4°C, and cryoprotected in a series of sucrose-PBS solutions. Tissue was sectioned at 40 μm on a microtome (Leica). Free-floating brain sections were permeabilized with 0.3% Triton X-100 in PBS for 1 hour and then blocked for 2 hours at room temperature with 0.3% Triton X-100 and 10% normal goat serum in PBS, followed by incubation with primary antibodies in 0.3% Triton X-100, 5% normal goat serum overnight at 4°C. On the next day, brain sections were washed in PBS and then incubated with secondary antibodies conjugated to Alexa 488, Alexa 555, and Alexa 647 (Thermo Fisher, 1:500) for 2 hours at room temperature. After rinsing with PBS, sections were incubated with DAPI. The following primary antibodies were utilized: GluR4 (Sigma, AB1508, 1:100) and PV (Synaptic Systems, 1:500, 195 004).

Immunocytochemistry

For live-cell labeling of GluR1, DIV14–16 primary hippocampal neurons cultured on coverslips were first incubated with 1.0% bovine serum albumin in Tyrode buffer (15 mM D-(+)-Glucose, 108 mM NaCl, 5 mM KCl, 2 mM MgCl₂, 2 mM CaCl₂, 1 M HEPES, pH 7.4) for 5 min at room temperature. Cells were then incubated with GluR1 antibody (Sigma, ABN241, 1:100) in Tyrode buffer for 12 min at room temperature. Subsequently, cells were washed with PBS and fixed with 4% paraformaldehyde in PBS for 15 min at room temperature. Cells were permeabilized with 0.25% Triton-X in PBS for 5 min and then blocked in 10% normal goat serum in PBS for 1 h at room temperature. For staining of intracellular proteins of interest, such as PSD-95 (Sigma, MAB1596, 1:500 diluted) and MAP2 (Synaptic Systems, 188 006, 1:1000), coverslips were incubated with the primary antibody diluted in 10% normal goat serum overnight at 4°C. After washing with PBS, cells were incubated with secondary antibodies conjugated to Alexa 488, Alexa 555, and Alexa 647 (Thermo Fisher, 1:500 in 10% normal goat serum). After washing, cells were mounted on glass slides in ProLong Diamond Antifade Mountant (Thermo Fisher).

For surface labeling of GFP, cells were fixed with 4% paraformaldehyde in PBS for 10 min at room temperature. After washing with PBS, fixed cells were blocked in 10% normal goat serum in PBS for 1h at room temperature, and then incubated with GFP antibody (Thermo Fisher, A6455, 1:500) in 10% normal goat serum for 2 h at room temperature. After washing with PBS, cells were incubated with secondary antibodies diluted in 10% normal goal serum for 1 h at room temperature. Subsequent staining of intracellular proteins was performed as described above.

Image acquisition and analysis

Z-stack images of primary neurons and brain sections were taken with a Zeiss LSM 800 laser scanning confocal microscope with Airyscan. All quantitative measurements were performed with ImageJ imaging software (NIH). Before quantification, maximum intensity projections for each image were generated. For the quantification of either surface GluR1 or PSD-95 puncta, we restricted our analysis to those residing within or immediately adjacent to MAP2-positive secondary dendrites by masking. After setting minimum intensity threshold level for each fluorescence channel, both the number and mean intensity of puncta were measured by using the ‘analyze particles’ function. The puncta counts were normalized by dendritic length to calculate puncta density.

Ribosome profiling

E16 primary cortical neurons were cultured in 10 cm dishes for 12 days before being treated with either DMSO or bicuculline (40 μM) for 10 hours. TIS mapping by ribosome profiling was performed by following the previously described QTI-seq method³² with modifications. Neurons were washed twice briefly with ice-cold PBS supplemented with 5 μM harringtonine. 400 μl cold cell lysis buffer (20 mM HEPES, pH 7.4, 100 mM KCl, 5 mM, MgCl2, 500U/ml RNasin-Plus (Promega), 1x protease inhibitor cocktail (200X, Millipore), 5 μM harringtonine) was added directly to the cells, which were then scraped off the dishes and transferred to 2 ml microcentrifuge tube containing Lysing Matrix-D (MP Biomedicals). Cells were lysed by vortexing 6 times (20 seconds each) with 40 s intervals on ice. After centrifugation for 10 min at 13,000 g at 4 °C. The supernatant was transferred into a new microcentrifuge tube and supplemented with 10 mM creatine phosphate, 0.1 mM spermidine, 40 μg/mL creatine phosphokinase, 0.8 mM ATP, and 25 μM puromycin, before incubation at 35 °C for 15 min. After incubation, RNA concentrations of the cell lysates were measured using Qubit RNA High Sensitivity Assay Kit (Thermo Fisher, Q32855). 10U RNase I (Thermo Fisher, EN0601) was added into cell lysate for every 100 μg RNA, and digestion was carried out at 4°C overnight with gentle rotation. The next day, RNase I digestion was stopped by adding Superase-In RNase inhibitor (Thermo Fisher, AM2696) at 1 U/μl. Digested cell lysate was loaded on top of 0.8 ml sucrose cushion (1 M sucrose, 20 mM HEPES, pH 7.4, 100 mM KCl, 5 mM, MgCl2, 5 μM harringtonine, 25 μM of puromycin) in a 1.5 ml ultracentrifuge tube (Beckman Coulter, 343778) and centrifuged for 1.5 h at 75,000 rpm at 4°C using a TLA-120.2 rotor (Beckman Coulter). After carefully removing the supernatant, the ribosome pellet was solubilized in TRIzol (ThermoFisher) for RNA extraction. The extracted RNA was separated in a 15% Urea-TBE gel, and RNA fragments of 25 – 35 nt in length were excised for subsequent library construction.

Sequencing libraries were constructed following a previously established protocol⁵⁸ with modifications. Extracted RNA fragments were first treated with T4 Polynucleotide Kinase (T4 PNK, NEB, M0201L) without adding ATP to remove the 3′ end phosphates. A 5′ pre-adenylated and 3′ blocked adaptor (/5rApp/NNNNNNNNNNCACGGCGATCTTGCCGCC/3ddC/) was ligated to the 3′ ends of RNA fragments by using T4 RNA Ligase 2, truncated KQ (NEB, M0373L). The 5′ ends of RNA fragments were phosphorylated by T4 PNK and ligated to 5′ adaptor (CGATCTCCAATTCCCACTCCTTTCAAGACCTrC) using T4 RNA Ligase 1 (NEB, M0437M). Ribosomal RNA (rRNA) was depleted using RiboPOOL probes against human/mouse/rat rRNAs (siTOOLs BIOTECH). Reverse transcription (RT) was conducted using SuperScript IV Reverse Transcriptase (Thermo Fisher, 18090050) with RT primer GGGCGGCAAGATCGCCGTG. Libraries were amplified by PCR using Q5 Hot Start High-Fidelity DNA Polymerase (NEB, M0493L) with P5 primer (AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCTNNAGGGCGGCAAGATCGCCGTG) and P7 primers (DMSO sample: CAAGCAGAAGACGGCATACGAGATCGGTTCAAGTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTCCAATTCCCACTCCTTTCAAGACCT; Bicuculline sample: CAAGCAGAAGACGGCATACGAGATGCTGGATTGTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTCCAATTCCCACTCCTTTCAAGACCT). After quality control, libraries were pooled and sequenced on a NovaSeq 6000 (Illumina).

Ribosome profiling data analysis

After clipping both 5′ and 3′ adaptors, reads were deduplicated using FASTX-Toolkit. The 10 nt UMIs were then removed using Cutadapt⁵⁵. Reads were first mapped to a mammalian rRNA and tRNA reference using Bowtie. The remaining unmapped reads were then mapped to mouse reference genome GRCm39 and transcriptome by STAR⁵⁴. 28–31 nt footprints were extracted for subsequent analysis using Samtools⁵⁷, with the P site offset for each footprint length being determined by using RiboMiner⁵⁶. For each read, P site was inferred by adding the P site offset to the 5′ coordinate.

To predict alternative TISs, we first analyzed the RPF distributions near annotated TISs. For each gene, we inferred the most abundant transcript isoform. RPF counts within a 11-codon TIS-flanking region (−5 to +5) were extracted and normalized to RPF count at the annotated TIS. This aggregate RPF distribution near all annotated TISs was used as a standard distribution. To globally identify alternative TISs, we scanned each transcript using a 11-codon window centering at each position in-frame with the annotated TIS, from 300-nt upstream to 300-nt downstream from the annotated TIS. RPF distribution within each 11-codon window was calculated and compared to the standard RPF distribution from all annotated TISs. A position was considered as an alternative TIS if the relative RPF counts within that window fell within the 10–90th percentiles of the standard RPF distribution at all 11 codon positions. A minimum of 10 RPF reads at the TIS was required. Alternative TIS in uORFs were excluded by the presence of in-frame stop codons between upstream alternative TIS and the annotated TIS.

Locomotor activity

All rodent behavioral analyses used approximately equal numbers of 10-month-old male littermates at the time of testing. Mice were placed in the testing room at least 30 min before each assay. All experiments and analyses were performed completely in a genotype-blind manner. Locomotor data was collected using an Accuscan Instruments behavioral monitoring system and Fusion software (Omnitech Electronics). Mice were temporarily single-housed on a 12-hour light-dark cycle with food and water ad libitum. Nesting material was removed to prevent obstruction of infrared beams in the locomotor monitoring system. Locomotion was monitored for 72 hours using 12 photocells placed 4 cm apart. Locomotor counts were monitored in 60 s blocks to obtain an “ambulatory activity count” consisting of the number of beam breaks recorded during a period of ambulatory activity. Mice were not disturbed during the testing period. Raw data from the Fusion software was then post-processed using a custom MATLAB script to extract the first 3 hours of data and re-organize into 15-min time bins.

Fear conditioning

Mice were tested in four 6-min sessions (habituation, training, contextual memory test, and cued memory test) across 4 days modified from previously described protocols⁵⁹. All protocols were administered using VideoFreeze software (Med Associates). For habituation on Day 1, mice were placed in a lit plastic chamber (Med Associates) with a cardboard floor and some bedding from the mouse’s home cage (Context A). No shocks or cues were delivered. For training on Day 2, mice were placed in the same plastic chamber with white plastic covering the walls, light, stainless steel grids for shock delivery (Context B), and almond extract underneath the grids to provide a specific odor. After a 160-s baseline, a tone (85 dB, 2.8 kHz) was presented for 20 s, with the last 2 s coinciding with a foot shock (0.5 mA). This same tone-shock pairing was provided a total of 3 times with 40-s inter-trial intervals and a 60-s final resting period. For the contextual memory test on Day 3, mice were again placed in the chamber set up for Context B, but with orange extract underneath the grids to provide the odor. No shocks or cues were delivered. For the cued memory test on Day 4, mice were placed in the chamber set up for Context A, and the sound cues were administered as described for training, but without foot shocks. Mice were returned to their home cage after each session. Freezing behavior during each session was analyzed by the VideoFreeze software, and then post-processed using a custom MATLAB script.

Y-maze spontaneous alternation test

Mice were acclimated to the testing room and handling for at least 3 days prior to the experiment. For the test, the animal was first placed at the center of the Y maze with all three arms initially blocked from entry. After removing the barricades, the animal was allowed to freely explore the three arms for 5 min. An arm entry was recorded when all four limbs of the test animal are within the arm. For each trial, the total numbers of arm entries and alternations were counted. The fraction of spontaneous alternation, starting from the third arm entry, was calculated.

Transmembrane domain and signal peptide prediction

Previously annotated human⁷ and mouse⁵ alternative N-terminal proteoforms based on ribosome profiling results, as well as their corresponding canonical proteoforms, were used to predict bifunctional signal sequences. Both transmembrane domains and cleavable signal peptides were simultaneous predicted by using Phobius (standalone version 1.01)^27,28. Signal anchor-to-signal peptide conversions were defined as N-terminal truncations that converted the canonical proteoform with a single transmembrane domain to a proteoform with a signal peptide and no predicted transmembrane domain. The opposite was defined as signal peptide-to-signal anchor conversions.

Quantification and statistical analysis

Statistical analysis

All statistical details of experiments can be found in the figure legends. All experiments were repeated at least two biological replicates. All statistics were performed by using either GraphPad Prism or RStudio. To determine statistical significance, Student’s t and Mann-Whitney test were performed for pairwise comparisons, and one-way ANOVA and Kruskal-Wallis tests were utilized for comparisons between multiple groups. Two-way repeated measures ANOVA was utilized to test the significance of stimulus to genotpye interactions in fear conditioning experiment. Values are reported as mean ± SD.

HIGHLIGHTS.

• NPR mRNA contains a CUG and an AUG TIS, each producing a distinct proteoform.

• Alternative TIS usage in NPR is regulated by RNA structures and neuronal activity.

• AUG initiation shortens the signal sequence and converts NPR to a secreted factor.

• Altering NPR proteoform ratio impacts AMPA receptors in PV⁺ neurons and behaviors.