The sequences of the human variants of the transcript leader of CDKL5 were acquired from the GRCh38.p13 assembly of Ensembl (Howe et al., 2021) and published literature (Hector et al., 2016). Sequence conservation analysis was performed starting from a nBLAST search (Sayers et al., 2021) in the nucleotide database. The search parameters were optimized for non-coding regions, setting a word-size of seven bases (BlastN algorithm), and an expected threshold (E) of 0.1. Match/mismatch scores of 1/-2 and gap costs of 1 for existence and 1 for extension were selected. nBLAST search in the EST database was conducted with the same parameters. Multiple sequence alignment of the chosen 26 sequences (Supplementary material 2) was performed using Clustal-omega (Sievers and Higgins, 2014) and visualized through Consurf (Ashkenazy et al., 2016). In Consurf, we selected the calculation method maximum likelihood (ML) and the evolutionary substitution model HKY85 (Hasegawa et al., 1985). Consurf scores for each nucleotide indicates the conservation at the given position (1 = not conserved; 9 = very conserved). These scores were analyzed through the Prism software (GraphPad Software), applying a sliding window on the arithmetical means of five nucleotide positions to obtain the conservation profile of the sequence of interest.

The prediction of the translation initiation start sites was conducted using two tools with different algorithms: Netstart (Pedersen and Nielsen, 1997) and TIS Miner (Liu et al., 2005). The calculation of the GC content of the sequences was performed trough the mathematical formula:

and expressed as percentage. Watson and Crick secondary structure prediction was performed using RNAfold (Mathews et al., 1999; Hofacker, 2003). The predicted structures were interpreted following the reported guidelines (Gruber et al., 2008): (i) the stability of the structures predicted trough the positional entropy; (ii) the ensemble diversity (ED), as a measurement of the variability of the possible secondary structures folded from the same sequence; (iii) the distance between the GMFE the GCE indices, as a measurement of the reliability of the predicted MFE structure. RNA structures were visualized through forna (Kerpedjiev et al., 2015). The QGRS Mapper program (Kikin et al., 2006) was used to predict G-quadruplex motifs.

SHSY-5Y cells was cultured in RPMI 1640 media (Gibco) supplemented with 10% Fetal Clone III (Thermofisher Scientific), 1% penicillin/streptomycin (Thermofisher Scientific), and 1% non-essential amino acids (Cyagen). After trypsinization with 0.25% tripsyn/EDTA, Sigma/Aldrich cells were plated in six-well multiwell plates and used for transfection.

CD-1 wild-type mice were treated following the European Community Council Directive 2010/63/UE for the correct care and use of experimental animals. When the experimental animals reached the 30th postnatal day, they were dissected to take cortices. RNA Later Buffer (Invitrogen) was used to preserve the quality of collected samples at −20°C before RNA extraction.

Cells cultures were washed with PBS and lysed with PureZOL (BioRad) 5 min at room temperature directly on plates (300 μl/well). Lysate was collected, 60 μl/well of chloroform (Merck) was added and total RNA was prepared according to the standard PureZOL protocol. The same protocol was adopted for mice sample, except for the addition of a previous step, in which mouse cortices were homogenized in 500 μl of PureZOL, using the Tissues Lyser (Qiagen). Once purified, the RNA samples were quantified with Nanodrop (ThermoFisher) and retrotranscribed with the RT2 Easy First Strand Kit (Qiagen).

We designed the PCR primers by means of NetPrimer (PREMIER Biosoft), and we checked their quality through in-silico PCR (UCS Genome Browser). The list of primers used for the amplification of murine samples is as follows:

The list of primers designed for the amplification of SHSY5Y samples is as follows:

PCR reaction was carried out by Taq Polymerase (ThermoFisher) in 50 μl starting from 100 ng of cDNA with annealing at 63°C and extension at 70°C. The amplification products, collected at the 27th cycle for GAPDH and the 35th cycle for CDKL5, were loaded on a 2% agarose gel and visualized by a Gel Doc XR (BioRad).

Small interfering RNA transfections were performed using Lipofectamine 3000 Reagent (Thermo Fisher) with 100 pmol/ml of siRNA following the instructions. eIF4B siRNA was purchased by Dharmacon (ON-TARGETplus Human eIF4B siRNA, SMARTPool format). SHSY-5Y cells were transfected at 80% of confluency and then incubated for 72 h at 37°C. After this period, we performed cell lysis using RIPA Buffer (50 mM Tris/HCl pH 8, 150 mM NaCl, 1% NP40, 0.1% SDS, 0.5% sodium deoxycholate. In the study, 5 mM EDTA and protease inhibitor mix (CLAP: chymostatin, leupeptin, aprotinin, PMSF). Detection of proteins was performed by SDS-PAGE followed by Western blot analysis. The primary antibodies were αtubulin (Cell Signaling, 3873); CDKL5 (Santa Cruz Biotech., Sc-376314); eIF4B (Cell Signaling, 3592); and XIAP (Cell Signaling, 14334). Detection and quantification of chemiluminescence signals were performed by SuperSignal West Pico substrate (ThermoFisher) using the Chemidoc Imaging System and the ImageLab software (BioRad).

We purchased the pBRm2L plasmids with the inserts of interest (5′UTRs sequences) from Synbio Technology after providing the original plasmid (De Pietri Tonelli et al., 2004).

The CDKL5 inserts of pBRm2L were amplified by PCR. SalI and NcoI restriction sites were added at the end of the primers. After digestion of the amplified products with the two restriction enzymes (NEB) for 2 h at 37°C, the fragments were inserted by ligation with the Quick Ligation Kit (NEB) in the pBATmod2 plasmid (De Pietri Tonelli et al., 2003) digested with the same enzymes and pretreated with rSAP (NEB). The new plasmids were verified by LightRun sequencing service (Eurofins Genomics) and named pBATmod2Ex1205, pBATmod2Ex202up, pBATmod2UTex2, and pBATmod2EVA.

SHSY-5Y cells at 70%–80% of confluency on 24-well plates were first infected with the Modified Vaccinia virus Ankara (MVA), carrying the T7 RNA polymerase gene, and then transfected (according to De Pietri Tonelli et al., 2003). For DNA transfection, we used 500 ng of vector using Lipofectamine 3000 (ThermoFisher) following the manufacturer protocol. Cells were incubated at 37°C for 5 h, washed twice with PBS, lysed with passive lysis buffer and processed to detect the signals from FLuc and RLuc using the Dual-Luciferase Reporter Assay System (Promega) on a GloMax reader (Promega). We calculated the translation efficiency of FLuc as the ratio between FLuc and RLuc signals. All the measurements were repeated twice for each sample (20 μl of lysates) from at least three independent experiments performed in triplicate.

We performed transcription start site (TSS) peak quantification analysis on CAGE libraries of the following human tissues: adult brain, adult kidney, adult testis, adult spleen, adult lung, adult heart, adult liver, and fetal brain, made available by the FANTOM5 consortium. CAGE tags were analyzed using CAGEr package, freely distributed by R/Bioconductor (Haberle et al., 2015). Normalization of the analysis was performed in accordance with Balwierz et al. (2009). The same analysis was performed also on murine CAGE libraries from various tissues (cortex, lung, liver, heart, testis, and kidney) and developmental stages (embryonic, neonatal, and adult).

To establish the conditions to investigate the translational regulation of CDKL5, we listed all the known 5′UTR and built a comprehensive catalog of all possible variants (Supplementary material 1). Consistent with the CDKL5 transcript leaders reported in the NCBI reference sequences database (NM003159.3/NM003159.3 and NM001323289.2), all the analyzed sequences have a common organization, sharing the sequence that is proximal to the main translational start site, consisting of 162 nucleotides in the exon 2 (UTex2), and differing at the distal 5′ region (Figure 1). We considered the 16 alternative first exons reported in Table 1 arising from various transcription start sites (TSSs), as previously claimed in other studies (Hector et al., 2016). A part of these sequences is only predicted (see Ensembl), whereas the experimentally defined first exons (Hector et al., 2016, 2017b) are not included in any reviewed databases. Moreover, most of the exons differ for the length at the 5′ end, a feature often associated with technical limitations (i.e., degradation of the mRNA or interrupted retrotranscription due to tight structures) in defining the first nucleotides of a transcript.

For these reasons, we performed sequence conservation analysis to screen sequences in the predicted CDKL5 transcripts and to play a role in protein synthesis modulation, following the assumption that conserved untranslated sequences have a higher chance to retain a regulatory function (Hardison, 2000). Starting from UTex2, the sequence conservation analysis predicted that it is anciently conserved, with homology in more than 206 genomes, including mammals, reptiles, and birds (not shown). This result is strengthened by the returned hits in the expressed sequence tags (EST) database, confirming the UTex2 detection in at least one bird transcriptome (Lonchura striata domestica: DC278838.1). Concerning the first exons, almost all the sequences reported in Table 1 are not conserved. The first exon sequences that show a reliable conservation are the Ex1 205 (88 nucleotides) and Ex1 202 (113 nucleotides), with homologous hits in 64 mammalian species since marsupials, hinting the possible origin of these sequences in this infraclass (not shown; see also Supplementary material 2). By comparing Ex1 202 (i.e., 25 nucleotides + Ex1 205) and UTex2, it is interesting to note that the first exon is more conserved that UTex2 despite being evolutionary more recent, pointing toward a possible important function within the first exon sequence in mammals (Figures 2A, B). Interestingly, Ex1 202 and Ex 205 were the only ones to be confirmed by a search in the EST database, where both returned homology also in non-primate transcriptomes. The sequence Ex1 202 was found in four EST libraries from various primates and rodents. Notably, one of this returned a record, belonging to the marmoset monkey (Callithrix jacchus: HX595850.1), that exceeds the 5′ limit of the human query, suggesting the existence of a longer version of the Ex1 202 also in humans. To verify this possibility, we performed a PCR-based approach on cDNA obtained from SHSY5Y cells and p30 murine cortices (Figure 3A). The analysis detected the complete sequence of Ex1 202 but also confirmed the presence of additional nucleotides upstream the reported TSS of the 202 variant (+10 nucleotides, Figure 3B). This longer version of the Ex1 202 was named Ex1 202up.

To evaluate the presence of cis-acting regulatory elements in the 5′UTR of CDKL5, we performed a predictive analysis on all the CDKL5 transcript leaders. We did not obtain any significant prediction of upstream start sites using two different algorithms (Netstart, Pedersen and Nielsen, 1997; TIS Miner; Liu et al., 2005, Supplementary Table S1). We then focused on possible regulatory secondary structures that might play a role in the modulation of translation initiation rate (Hinnebusch et al., 2016; Leppek et al., 2018). First, we calculated the GC percentage of the exonic sequences. Our analysis highlighted that the majority of the first exons and the common UTex2 did not have a particularly high GC percentage, while the conserved 205 and 202 sequences, as well as HecA1, exceed the conventional threshold of 60% of GC nucleotides (Figure 4A). This threshold is commonly associated to the presence of regulatory secondary structures (Davuluri et al., 2000). In line with this assumption, the global estimated Watson and Crick folding of all the possible CDKL5 5′UTRs returned disorganized and disordered prediction, except for the 205 and 202 variants, as shown by positional entropy (Table 2). These two sequences are characterized by a negative ΔGMFE of −91.6 and −101.4 kcal/mol, respectively. These values, and the corresponding predicted structures, can be considered highly reliable according to the guidelines of RNAfold since GCE is comparable to ΔGMFE and the Ensemble Diversity is relatively small (Gruber et al., 2008) (Table 2). Moreover, also the MFEden normalization (Trotta, 2014) that considers the length of the sequence confirms the reliability of the prediction (Supplementary Table S2). This analysis strengthened the selection of 202 and 205 sequences as interesting from a functional viewpoint. In the 205 5′UTR sequence, the 88 nucleotides composing the first exon give the greater contribution to the stability of the structure, involving various stem-loops, some of which formed together with the UTex2 nucleotides, with very low positional entropy (Figures 4B, C). The longer variants 202 and 202up returned, as expected, a very similar structure. However, the presence of the 25 (or 35 in the case of 202up) upstream nucleotides before the putative 205 TSS are characterized by a localized higher entropy level (Figures 4D, E). Since this entropy level profile in a G-rich region suggests the presence of G-quadruplexes, other secondary structures involved in translational regulation (Zhang et al., 2011; Bugaut et al., 2012; Takahashi and Sugimoto, 2021), we performed a G-quadruplex prediction on these sequences (as well as on all the reported exonic sequences of the CDKL5 transcript leader). The prediction returned the presence of three distinct motifs (namely, pG4₁, pG4₂, and pG4₃) on the sequence Ex1 202up. While pG4₂ and pG4₃ are shared with the 205 sequence, pG4₁, the most conserved of the three, lies in the exclusive 35 nucleotides of the variant 202up (nucleotides 13–24), hinting a possible functional difference between the two alternative first exons (Figure 5). From these bioinformatic analyses, the conserved and structured 5′UTR variants 205 and 202 (or 202up) appear to be the transcript leaders with the highest probability to play a role in the translational control of CDKL5.

Because of the highly structured 5′UTRs of CDKL5, we hypothesized a translational regulation by eIF4B modulation of the DEAD-box eIF4A helicase (Dmitriev et al., 2003; Shahbazian et al., 2010; Parsyan et al., 2011; Sen et al., 2016). Since eIF4A is an obligatory component of the translational machinery (see, for instance, Sénéchal et al., 2021), we could not consider a gene-silencing approach related to this protein. Therefore, we focused on eIF4B, and we assessed if the downregulation of this translation initiation factor could inhibit CDKL5 expression. To test this hypothesis, we examined CDKL5 protein levels in lysates obtained from control and eIF4B-silenced SHSY5Y cells. We included in the analysis another X-linked apoptosis regulator, XIAP (NM_001167.4: ΔGMFE 5′UTR = −133.60 kcal/mol, GC% = 65.2), as a positive control (Holcik et al., 2000). In fact, XIAP was reported to be translationally underexpressed in Hela cells when eIF4B was silenced (Shahbazian et al., 2010). The result shows that transfection of eIF4B siRNA (Figures 6A, B) halved the expression of CDKL5 as well as of XIAP (Figures 6A, C, D). On the contrary, the levels of αtubulin, a protein with unstructured 5′UTR (NM_001101.5: ΔGMFE 5′UTR = 13 kcal/mol, GC% = 76.2), used as negative control (Shahbazian et al., 2010), remained unchanged over treatment (Figures 6A, E).

With the aim of assessing the capability of the selected 5′UTR variants (205 and 202up) to modulate the translation of a downstream protein, we performed a dual-luciferase reporter assay. In the first experiment, we employed the “bi-monocistronic” two promoter vector pBRm2L (De Pietri Tonelli et al., 2004), that contains two independent transcriptional units under the control of distinct T7 RNA polymerase promoters (Figure 7A). Infecting the cellular system with the MVA virus carrying the T7 RNA polymerase gene, allowed to obtain the rapid, cytosolic transcription of the two cistrons, eliminating in this manner some of the technical limitations, such as the presence of cryptic promoters or splicing events, traditionally linked to standard DNA transfection when translation has to be assessed (Yang and Wang, 2019). The first cistron of the vector encodes the Renilla Luciferase (RLuc) protein downstream a short optimized 5′UTR, while the second cassette encodes the Firefly Luciferase (FLuc) placed downstream the 5′UTRs of interest. Based on the premise of the bioinformatic analysis, we analyzed the 205 (250 nts) and 202up (285 nts) CDKL5 transcript leaders (Figure 7A). Translation efficiency was quantified as the ratio of FLuc and RLuc signals. The sequence under analysis were compared with three controls: (i) the empty pBRm2L vector in which the expression of FLuc is under the control of short, optimized 5′UTR; (ii) the pBRm2L-B1x vector containing the transcript leader of BACE1 (GenBank accession number BG833894), a sequence known to reduce the translation efficiency of the downstream ORF (De Pietri Tonelli et al., 2004); and (iii) a truncated version of the CDKL5 5′UTR composed uniquely by the untranslated region of exon two (162 nts, UTex2), in order to evaluate, by difference, the effect of the first exons on overall translation of CDKL5 (Figure 7A). This reporter gene approach revealed that all the CDKL5 5′UTRs analyzed (205, 202up, UTex2, respectively, cloned in pBRm2L-205, pBRm2L-202up, and pBRm2L-UTex2) did not inhibit the Fluc expression, as the BACE1 5′UTR did, but rather caused an increase in translation efficiency (UTex2: +195%; 205: +119%; 202up: +57% compared to the baseline set by pBRm2L; Figure 7B, statistical analysis in Supplementary Table S3). This result suggests the presence of a cis-acting regulatory motif working as an enhancer of the translation efficiency in the CDKL5 5′UTR, spanning from at least the sequence Ex1 205 to the UTex2. At the same time, we tested the effect of the SNP rs786204994 described by Evans et al. (2005), with the aim to assess a possible inhibitory effect (by the use of the construct pBRm2L-EVA). However, choosing the condition 202up as control, we did not observe any significant difference (Figure 7B).

Based on the results obtained with the “bi-monocistronic” approach, we considered the presence of a cis-acting enhancing motif in the CDKL5 transcript leader that could have IRES features, allowing CDKL5 transcripts to undergo not only cap-dependent but also cap-independent translation. This dual mode of manage translation is not unusual for neuronal transcripts with important function in development (Audigier et al., 2008; Choi et al., 2019). Therefore, to verify the assumption that also CDKL5 could act in a similar manner, we performed the dual-luciferase reporter assay using the dicistronic pBatmod2RL vector. This construct allows to investigate the presence of a cap-independent translation that involves the presence of an IRES (De Pietri Tonelli et al., 2003). In this vector, the two luciferases coding regions are under the control of a unique T7 promoter that allows the cap-dependent initiation of the first ORF (RLuc) but not of the second one (FLuc), leaving the translation of the latter under the control of non-canonical, cap-independent translational initiation. In this experiment, we analyzed the 205, 202up, UTex2 transcript leaders, as well as the one carrying the rs786204994 SNP (Figure 8A). The FLuc/RLuc ratio values obtained confirmed the presence of an enhancing cis-acting regulatory element involving in cap-independent initiation (Figure 8B). In this case, the values obtained for the 205 and 202up transcript leaders were higher than the one returned from the UTex2 (205: +52.5% and 202up: +28.8% compared to UTex2), suggesting a contribution of the first exon sequence in the IRES function, regardless the presence of the 5′ extension of the 202up variant (Figure 8B, statistical analysis in Supplementary Table S4). Interestingly, rs786204994 SNP described a patient in the 5′UTR of CDKL5 (Evans et al., 2005) produced a decrease of about 25% on the FLuc/RLuc ratio when compared to its control pBatmod2_202up (Figure 8B). This finding suggests that C189 position is involved in the enhancing regulatory motif and that its transition to T might impair the non-canonical translational initiation efficiency, suggesting a possible mechanism in CDD pathogenesis. In any case, the disturbing effect of the C>T transition at position −189 agrees with its predicted effect on ΔG, which becomes less negative compared to the ΔG variations of all the possible C>T transitions in the CDKL5 5′UTR sequence (Supplementary Figure S1).

To quantify the abundance of the variants carrying the 205 and 202up CDKL5 5′UTRs, we analyzed the percentage of the CDKL5Transcription Start Sites (TSSs) in various adult human tissues by consulting the FANTOM 5 libraries (Noguchi et al., 2017). The evaluation of the cap analysis of gene expression (CAGE) libraries can determine the precise positions of the CDKL5 TSSs, solving the problem of the ambiguous 5′limits described previously. For our analysis, we selected libraries from the following adult organs: brain, heart, kidney, lung, spleen, liver, testis, and fetal brain. This last library was added to assess the presence of an alternative first exon usage in different developmental stages according to the difference in the CDKL5 expression during brain development (Rusconi et al., 2008). The quantification was conducted using the CAGEr package from bioconductor (Haberle et al., 2015). The results show that most of the alternative CDKL5 first exons reported in Ensembl and in the study of Hector et al. (2016) are almost absent in the analyzed transcriptomes since their TSS were not detected at all. On the contrary, the TSS of Ex1 205 (NM001323289/NM003159.3) resulted to be the preferential CDKL5 TSS in each tissue analyzed, ranging from a minimum of percentage of usage of 55.41% in the fetal brain and a maximum of 87.73% in the lungs (Figure 9A). The other reference TSS, belonging to Ex1 204a (NM 001037343) resulted to be expressed uniquely in the testis (the human tissues with more used TSSs), with a percentage of usage of 7.38%. The second most abundant TSS was represented unexpectedly by an unreported TSS lying 26 nucleotides downstream the TSS of 205, that we called 205down (Figure 9B). Its existence is confirmed by the CAGEr peak quantification analysis conducted in various murine tissues, where it has a more weight in determining CDKL5 transcription, with a percentage of usage greater of TSS of 205 in some tissues (Figure 10A). In human tissues, the frequency of transcription of the Ex1 205down averages 20% (max. adult kidney: 23.12%; min. adult kidney: 9.19%). The resulting first exon sequence differs from the longer Ex1 205 from the exclusion of the pG4₂ motif, suggesting a possible functional difference in the structure of the transcript leader or in the definition of the promoter. The other TSS that we identified, the TSS of 202up, previously assumed after the RT-PCR (Figure 9), was less expressed (max. adult testis 4.91%), contrary to the 10 nucleotides downstream the TSS of 202 that is the third most transcription starting point of CDKL5, with a maximum of 19.74% in fetal brain, while it ranges between 0 and 8.33% in adult tissues (Figure 9A). The comparison of the percentage of use level of the TSS of 202 in the adult brain and fetal brain (adult: 5.02; fetal: 19.74%) at the expense of the TSS of 205 (adult: 69.85%; fetal: 55.41%) point toward a plausible differential transcriptional preference never observed in any other adult tissues and hinting a possible role in brain development (Figures 9A, B). This information might turn out to be interesting since we did not observe a similar trend in the murine cortical libraries (taken into consideration in the absence of total brain libraries), where the percentage of use of the TSS of 202 remains stable in a tight range in different developmental stages (adult: 0.99%; neonate: 1.21%; embryo: 0.53%; Figures 10A, B). Our analysis defined the presence of a unique TSS cluster for the gene CDKL5, composing by four start sites: 202up, 202, 205, and 205down (Figure 9B). The result obtained highlights that while the TSS cluster is conserved between the human and the mouse, the percentage of usage of the single TSSs highly differs. Despite the high conservation of the transcript leaders of CDKL5, our finding points toward the possibility to have an alternative use of the various TSS within an evolutionary context, favoring, in humans, longer 5′UTRs, enriched in cis-acting elements.

The growing relevance of CDKL5 as a player in neuronal functions highlights the need of gaining knowledge about the regulation of its expression. In this context, translational regulation seems to have a crucial role based on hints available in the literature (La Montanara et al., 2015) as well as on the complexity of the transcript leader of CDKL5. This necessity arises also from the need to improve the diagnostic accuracy of CDD since some of the untranslated exonic regions of CDKL5 were routinely excluded from the diagnostic screenings because they are considered unfunctional (Hector et al., 2017a,b) although SNPs with possible pathological consequences have been described (Nemos et al., 2009; Mei et al., 2010). A Part of the difficulties in investigating CDKL5 5′UTR is due to the inaccuracy in the definition of the variant sequences and, consequently, to the lack of a common nomenclature. For this reason, we first propose an unambiguous nomenclature based on the Ensembl transcript IDs and the experimental study of Hector et al. (2016). From the possible 5′UTRs emerged that the numerous variants of the 5′UTR maintain an established sequence organization, sharing a sequence proximal to the start codon (UTex2) in the second exon but varying in the first exon(s). This further highlights the presence of various TSSs, as previously hypothesized (Hector et al., 2016). However, the difference of few nucleotides at the 5′ ends of some alternative first exon sequences suggests the presence of errors in the reported variants possibly due to the technical limitation in the determination of the TSSs of the gene (Hector et al., 2016; Adiconis et al., 2018). Our analysis confirmed this possibility, showing that the majority of the CDKL5 first exon variants are not conserved or not reported in the transcriptomes of non-primate species, and their TSSs are absent in FANTOM5 libraries. By CAGEr (Haberle et al., 2015), we validated a TSS cluster around the reference TSS 205 (ChrX: 18425608), composed by other three TSSs closely spaced on the same region, named 202up (ChrX: 18425573), 202 (ChrX: 18425583) and 205down (ChrX: 18425633). Although CDKL5 TSSs identification has been previously attempted (Vitezic et al., 2014; Hector et al., 2016), our CAGE analysis provided new elements to the topic, such as the differential percentage of usage of the variants in the context of human development (adult vs. fetal). Moreover, this approach hints to a multifaceted and dynamic interplay between CDKL5 transcriptional and translational mechanisms with a possible inclusion of part of the transcript leader in the composition of the promoter and vice versa. This agrees with the complex temporal expression pattern of the 5′UTR of CDKL5 in various tissues.

From a functional perspective, our bioinformatical evaluation predicted a richness of structural cis-acting elements in this exonic region of interest, with a stable global structural folding of the 5′UTR. This feature points to the involvement of the DEAD-box helicase eIF4A and its modulation by eIF4B (Parsyan et al., 2011). Since a strong and prolonged gene silencing of eIF4A, an essential protein of the translational machinery (Sénéchal et al., 2021), would have affected the cell physiology in a complex and unpredictable manner, we decided to focus on the downregulation of eIF4B. Indeed, our hypothesis of the engagement of these motifs in the 5′UTR-mediated translational control of CDKL5 was confirmed by the involvement of this translational initiation factor in setting the CDKL5 protein level. This correlation is particularly relevant in the neuronal context, given the established role of eIF4B in the dynamic modulation of the translation initiation rates (Harms et al., 2014; Sen et al., 2016), and its proposed role in local translation (Bettegazzi et al., 2017) at the level of synapses.

eIF4B is involved in both cap-dependent and cap-independent translational initiation (López de Quinto et al., 2001). While we suggest the presence of a translation control mediated by the 205 and 202up 5′UTRs that is consistent with the eIF4B function in cap-dependent translation, we also found that the CDKL5 5′UTR could support cap-independent translational initiation that enhances the expression of the downstream ORF. More specifically, we propose the presence of an IRES in the CDKL5 transcript leader that, according to the prediction of structural elements, the analysis of the evolutionary conservation, and the results obtained with the UTex2 5′UTR (i.e., a truncated version of 205 and 202up) could span from UTex2 to the proximal part of the first exon. The possible presence of this non-canonical mechanism in CDKL5 translation makes this kinase to fall in roughly 17% of human transcripts able to switch from cap-dependent to the IRES-mediated initiation according to the cellular needs and occurring frequently in brain and neuronal development (Audigier et al., 2008; Choi et al., 2018; Fan et al., 2022). Interestingly, and in agreement with this hypothesis, we used our experimental approach to collect the first evidence of the effect of the transition C>T-189 identified in a CDD patient (Evans et al., 2005). To be associated with the CDD condition, this SNP would be expected to decrease the expression of the downstream ORF. Indeed, we were able to observe this effect when cap-independent translation initiation was assessed, hinting to the inclusion of C-189 in the proposed CDKL5 IRES. IRESes are usually structural elements that can be formed by either Watson and Crick base pairing or Hoogsteens G quadruplex assembly (Morris et al., 2010; Hoque et al., 2022). Our approach predicted both structures in the analyzed 5′UTR of CDKL5. However, the evidence collected by the effect of the C>T-189 SNP on translation and the stem-loop structural context in which this SNP lies points toward a Watson and Crick base pairing in the composition of the IRES. In fact, a Hoogsteens-based in the structure pG4₂ that would support the IRES structure would be unaffected or even reinforced by the C>T-189 SNP, while our experimental evidence with luciferase reporter genes is against this hypothesis.

In our study, we characterized the features of the 5′UTR of CDKL5, demonstrating its potential role in the translational regulation of the downstream protein. Moreover, we provide evidence that a SNP in this region might play a pathological role. SNPs affecting translation efficiency are becoming a growing finding in genetic analysis, especially in neurodevelopmental disorders (see, for instance, Pan et al., 2021; Coursimault et al., 2022). Interestingly, the presence of active cis-acting regulatory elements in the transcript leader could be exploited to boost the expression of the target protein by using approaches such as antisense oligonucleotides (ASOs) (Liang et al., 2017). ASOs, highly selective, minimal invasive, and customizable tools, are already emerged as treatment for loss-of-function monogenetic disorder (Bennett, 2019). Recently FDA approved one ASO (nusinersen) for Spinal Muscular Atrophy (Li, 2020) and ASOs targeting the 5′UTR has shown to be a promising treatment for cystic fibrosis (Sasaki et al., 2019). Therefore, a better understanding CDKL5 5′UTR will improve not only the accuracy of the diagnostic health care but also pave the way to a new therapeutical strategy to treat CDD by altering the efficiency of translation.