Attribute Network-based Ranking (ANeR) identifies conserved translational fingerprints of M-phase Authors: Sarah B Eivers, Axelle Baumgartner, Marcio Dorn, Catherine Jessus, Juliana Silva-Bernardes, Enrico Maria Daldello Published: 2026 ABSTRACT AbstractEntry into M-phase requires mRNA translation, but whether mitotic and meiotic divisions share a conserved translational programme and which newly synthesized proteins promote Cdk1 activation remain unclear. We developed Attribute Network-based Ranking (ANeR), a multi-objective network-based method that ranks mRNAs according to their proximity to a training group across heterogeneous two-condition omics datasets. Applied to 30 datasets covering mitotic and meiotic divisions, ANeR identified a conserved M-phase translational fingerprint. Benchmarking ANeR predictions against independent mitotic datasets, together with experimental validation in Xenopus oocytes, confirmed that ANeR prioritizes proteins whose accumulation increases during both mitotic and meiotic M-phases. Application of ANeR to the meiotic translational programme identified Fam76b and Cdca3 as previously unrecognized translationally activated regulators required for timely Cdk1 activation, demonstrating that ANeR can uncover new regulators of meiotic M-phase. Analysis of the oocyte translational programme established how broadly CPE/PAS regulatory rules apply, identified transcripts that do not conform to these rules, and experimentally validated their translational activation. Together, these findings establish that vertebrate mitotic and meiotic transitions share a conserved programme of translationally regulated transcripts and demonstrate that ANeR can identify new regulators of cell division. Broadly, ANeR provides a flexible framework for investigating diverse biological questions using a collection of two-condition datasets. Introduction Cell division is an essential process of life, which accurately and faithfully ensures the survival and propagation of organisms. Underpinning cell division are two separate processes: mitosis (or M-phase of the cell cycle) and meiosis. The final outcomes of meiotic and mitotic divisions differ. Meiosis, which consists of two successive rounds of division without intervening DNA replication, produces four haploid reproductive cells, whereas mitosis gives rise to two daughter cells that are genetically identical to the parent cell. However, since meiosis is widely considered to be evolutionarily derived from mitosis, the core players regulating these cell divisions are similar. Indeed, the activation of the kinase Cdk1 (Cyclin dependent kinase 1) is the universal trigger for cell division in all eukaryotes [1]. In association with its regulatory subunit, Cyclin B, active Cdk1 regulates the downstream activities of necessary kinases and phosphatases involved in protein phosphorylations that orchestrate the full range of cellular rearrangements required for cell division [2]. During cell division, transcriptional activity is strongly diminished in mitotic cells and completely repressed during meiotic divisions of reproductive cells, spermatocytes, and oocytes [3, 4]. Thus, translation of mRNAs becomes a central step of regulation in ensuring mitotic and meiotic progression. Indeed, it is well known that de novo protein translation is required for Cdk1 activation during mitosis and meiosis in most vertebrates [5]. However, the identities of the proteins required for Cdk1 activation still remain largely unknown. Only a few have been so far identified, mostly by biochemical methods, such as Cyclins B1, B2, and A, which bind and activate Cdk1 at the G2/M transition in mitotic cells, and at entry into the first meiotic division in oocytes [6, 7]. In oocytes, the translation of the kinase Mos, an indirect activator of MAPK kinase (MAPK or ERK) is required for the proper progression of meiotic divisions [8, 9]. However, translation of these known regulators does not fully account for M-phase entry, indicating that additional proteins remain to be identified [5]. This calls for unbiased approaches capable of identifying translationally regulated candidates and prioritizing them for functional validation. Given the common evolutionary origin of mitosis and meiosis, it is also plausible to hypothesise that a part of the translational program executed during cell division would be conserved between mitotic and meiotic M-phases. Studies investigating the mitotic M-phase are done in artificially synchronized cells, which could alter patterns of gene expression [10–12]. In addition, as transcription is not fully silenced at the entry into M-phase, programs of transcription and translation overlap during mitosis, adding further complexity to disentangling the two processes. Oocyte meiotic divisions offer a unique system for studying translation during M-phase. In metazoans, oocytes within the ovary are arrested at prophase of the first meiotic division (PRO), a stage equivalent to the late G2 phase of the cell cycle. In response to a hormonal stimulus during ovulation, they resume meiosis, undergo the first meiotic division, and subsequently proceed to the second. In vertebrates, oocytes then arrest at metaphase II (MII), where they remain until fertilization. Thus, the PRO–MII transition represents the meiotic counterpart of the G2–M transition of somatic cells. Unlike mitotic cell cultures, however, oocytes are naturally synchronized, meiotic resumption can be experimentally induced in vitro, and transcription is fully silenced throughout this transition. These features provide a major advantage over conventional mitotic cell culture systems. With the advent of the “omics” era, it is now possible to analyse gene expression in a high-throughput manner. Nonetheless, the power of these studies is often underutilised, with large amounts of data deposited onto databases without further investigation beyond the scope of the original studies. With this rise in big data and omics, computational approaches for analysing these datasets have also advanced rapidly. One issue that arises when extracting useful knowledge from these datasets is that most studies vary in experimental methods, technical approaches, and types of biological samples, making it difficult to compare and classify the data meaningfully. Thus, the challenge is to develop computational methods that can efficiently integrate multiple heterogeneous datasets. To treat such heterogeneous data, machine learning (ML) methods, which are at the core of artificial intelligence and data science, can be applied to large datasets to extract useful properties, which can then be transformed into biological knowledge. Here, we developed an Attribute Network-based Ranking (ANeR) method that integrates heterogeneous two-condition omics datasets and ranks transcripts according to their proximity to a curated training group. We applied ANeR to datasets describing mitotic and meiotic M-phase to identify a conserved translational fingerprint. We then used Xenopus laevis oocytes to validate selected predictions and test their functional relevance, taking advantage of the biochemical accessibility of this model and the strong conservation of the mechanisms controlling meiotic divisions between Xenopus and mammals [13, 14]. We identified Fam76b and Cdca3 as previously unrecognized regulators required for timely Cdk1 activation during meiotic divisions. We then used the ANeR-defined oocyte fingerprint to investigate the regulatory mechanisms underlying meiotic translation. This analysis established the penetrance of canonical CPE/PAS rules, revealed polyadenylation- and CPE-independent modes of translational activation, and identified GU-rich hexamer motifs as additional regulators of translation. More broadly, ANeR can be applied to a wide range of omics datasets that quantify molecular changes between two biological conditions and used to generate reliable predictions of transcripts that are co-regulated with a curated training group. Materials and methods Attribute Network-based Ranking ANeR is a multi-objective approach to ranking transcripts based on their network attributes. It requires a list of transcripts and their expression levels obtained from different conditions and/or organisms (Fig. 1E). Transcripts are split into two categories: candidates and validated (here referred to as the training set). The main goal is to rank candidate transcripts according to their similarity with validated ones. For that, we create a graph/network in which nodes represent transcripts and weighted edges represent their distances. We first placed the validated transcripts and their N closest neighbours (N = 100). The distance thresholds that define the neighbourhood are computed dynamically, considering the mean and standard deviation of each validated transcript. ANeR implements several methods for computing the distance between transcripts; Euclidean distance is used by default. Then, to rank candidate transcripts, we extract four attributes from the graph: the number of neighbours, the number of cliques, the size of the maximum clique, and the average distance to its neighbours, where a clique is a subset of nodes such that an edge connects every two distinct nodes in the clique. Finally, we used a multi-objective optimisation strategy based on the Pareto frontier to rank candidate transcripts. The Pareto frontier is a set of solutions that represent the best trade-offs among all objective functions. A solution not dominated by any other solution is placed in the first frontier. Next, we update the solution space by removing elements from the first frontier and find non-dominated solutions to form the second frontier. We proceed until we cover the entire solution space or reach a predetermined number of frontiers (Fig. 1E). Hierarchical clustering of translation datasets indicates two signatures of translation. (A) Pearson correlation matrix and hierarchical clustering of normalized datasets (CAT1–CAT39) measuring translation. Colours in the matrix indicate Pearson correlation coefficients (blue, positive; red, negative). The dendrogram represents similarity relationships between datasets. Dataset classes (mitotic cells, mouse, human, or Xenopus oocytes) are indicated by coloured branches in the dendrogram. (B) Ribosome immunoprecipitation (RPL0-IP) followed by qPCR for Ccnb1 mRNA in Xenopus PRO and MII oocytes. Levels of Ccnb1 mRNA were quantified in RPL0-IP and expressed as log2 (2–ΔΔCt) relative to PRO. Error bars represent mean ± SD; * indicates P < 0.05 (one sample t-test). (C) PCA of the ribosome-IP dataset (CAT39) and the corresponding total RNA (input). (D) Distribution of log2 (MII/PRO) mRNA fold changes in translation for all quantified mRNAs in dataset CAT39 after standardization. The distributions of log2 (MII/PRO) mRNA fold changes in translation for mRNAs belonging to the general M-phase and oocyte-specific training groups are also shown. (E) Schematic of the ANeR workflow. Two inputs are required: a set of datasets measuring features across two conditions, and a validated training group of transcripts (training set). For each dataset, a feature is calculated for every transcript, representing the change between the two conditions. A graph/network is then built, where nodes represent transcripts and weighted edges represent their Euclidean distance in the feature space. For each node n, the following network attributes are computed: number of neighbours (connected nodes to n), number of cliques (the size of a set composed by a subset of nodes, such that every two distinct nodes are connected), size of the maximum clique, and average distance to n neighbours. For node G, the number of neighbours is 3 since it is connected to A, F, and I; the number of cliques is 5 = |{G, F}, {G, A}, {G, I}, {G, F, I}, {G, F, I, A}|; the size of max clique is 4 = |{G, F, I, A}|; and the average distance is (d (G, F) + d (G, I) + d (G, A)) / 3.Transcripts are ranked using a Pareto frontier strategy, which provides a set of solutions representing the best trade-off among all network attributes. Principal component analysis-based ranking We implemented a ranking strategy based on principal component analysis (PCA). First, we computed the two principal components of the entire dataset. Next, we calculated the Euclidean distances between the transcripts in the training set and the candidate transcripts. Finally, we sorted the candidate transcripts by their nearest-neighbour distances to the training set to yield the top-ranked results. Features calculation from the available datasets Data from 38 publicly available datasets were obtained for this study (Table 1). For each transcript in the dataset, a specific feature was calculated that represents the change in translation accumulation level of the corresponding protein during cell divisions. In addition we compiled an additional Xenopus dataset using a ribosome-immunoprecipitation method and RNA-seq, as outlined below. For the datasets from somatic cells, we compared M-phase to G2 in CAT4 and 8; M- to S-phase in CAT1, 3, 5, and 6; M-phase to pro-metaphase in CAT9, mitotic to asynchronous in CAT36, G2 to interphase for CAT37. For the datasets from oocytes, we compared MII-arrested to PRO-arrested. For the mass spectrometry datasets, the log2 fold change of the protein level between the two conditions was used. For the translatome datasets, the log2 fold change of the normalized counts (TPM or CPM) from polysomes or ribosomes was used. For the polyadenylation datasets, the difference in poly (A)-tail length normalized by the average poly (A)-tail length between the two conditions was used. Table 1 lists all extracted features. Data were then standardized using the formula: (X-Average of dataset)/Standard deviation of dataset, where X corresponds to an individual mRNA/protein. mRNAs/proteins were only retained if they were present in at least 60% of the datasets tested. Dataset 1 contains the step by step processing of the data. Table 1.Datasets measuring protein translation/accumulation during M-Phase. Overview of the 39 two-condition omics datasets (CAT1–CAT39). For each dataset, the table reports the year of publication, experimental technique (proteomics, RNA-seq, or poly(A)-tail profiling), model organism (human, mouse, Xenopus), biological system (somatic cells or oocytes), PubMed ID, and the calculated feature tracking translation/protein accumulationCATDateTechniqueModelSystemCitationPMIDCalculated feature12013ProteomicHumanCellAviner et al., 201323934657log2(FC) M/S - protein de novo22013RNAseqHumanCellStumpf et al., 201324120665log2(TE in M) - Riboseq32014ProteomicHumanCellLy et al., 201424596151log2(FC) M/S - protein de novo42015RNAseqHumanCellTanenbaum et al., 201526305499log2(FC) M/G2 - Riboseq52016RNAseqMouseCellPark et al., 201627153541log2(FC) M/S - Riboseq62016polyAMouseCellPark et al., 201627153541(M-S)/average(S, M) - length poly(A) tail72016RNAseq Xenopus OocyteSession et al., 201627762356log2(FC) MII/GV - total mRNAs82017ProteomicHumanCellLy et al., 201729052541log2(FC) M/G2 - protein de novo92017ProteomicHumanCellLy et al., 201729052541log2(FC) M/promet - protein de novo102017Proteomic Xenopus OocytePeuchen et al., 201729142207log2(FC) MII/GV - protein level112018RNAseqMouseOocyteSha et al., 201830478191log2(FC) MII/GV - polysome122020RNAseq Xenopus OocyteYang et al., 202031896558log2(FC) MII/GV - polysome132020PolyA Xenopus OocyteYang et al., 202031896558(MII-GV)/average (GV, MII) - lengthPoly(A) tail142020RNAseqMouseOocyteLuong et al., 202031970406log2(FC) MII/GV - Ribotag-IP152020ProteomicMouseOocyteLi et al., 202033068521log2(FC) MII/GV - protein level162021RNAseqHumanCellVanInsberghe et al., 202134497418log2(FC) M/interphase - Riboseq172022RNAseqMouseOocyteXiong et al., 202235697785log2(FC) MII/GV - Riboseq182022PolyAMouseOocyteXiong et al., 202235697785(MII-GV)/average(GV, MII) - lengthPoly(A) tail192022RNAseqMouseOocyteXiong et al., 202235697785log2(FC) MII/GV - Riboseq (single oocyte)202022ProteomicHumanOocyteGuo et al., 202235809850log2(FC) MII/GV - protein level212022RNAseqMouseOocyteHu et al., 202236042231log2(FC) MII/GV - Riboseq (miniRibo-seq)222022RNAseqMouseOocyteHu et al., 202236042231log2(FC) MII/GV - Riboseq (pools)232022RNAseqHumanOocyteHu et al., 202236042231log2(FC) MII/GV - Riboseq242022RNAseqHumanOocyteZou et al., 202236074823log2(FC) MII/GV - Riboseq252022RNAseq Xenopus OocyteHwang et al., 202236516762log2(FC) MII/GV - Pellet fraction262023ProteomicMouseOocyteSun et al., 202336496143log2(FC) MII/GV - protein level272023RNAseqMouseOocyteOzadam et al., 202337344592log2(FC) MII/GV - Riboseq282023ProteomicMouseOocyteZhang et al., 202337443062log2(FC) MII/GV - protein level292023RNAseqMouseOocyteIyyappan et al., 202337950888log2(FC) MII/GV - polysome302023PolyAHumanOocyteLiu et al., 202336646905(MII-GV)/average(GV, MII) - lengthPoly(A) tail PAIso1312023PolyAHumanOocyteLiu et al., 202336646905(MII-GV)/average(GV, MII) - length Poly(A) tail PAIso2322024PolyAMouseOocyteLee et al., 202438306272(MII-GV)/average(GV, MII) - length Poly(A) tail332024PolyAMouseOocyteXiang et al., 202438460509(MII-GV)/average(GV, MII) - length Poly(A) tail342024PolyA Xenopus OocyteXiang et al., 202438460509(MII-GV)/average(GV, MII) - length Poly(A) tail352024ProteomicMouseOocyteSun et al., 202439256562log2(FC) MII/GV - protein level362024RNAseqHumanCellLy et al., 202439443796log2(FC) mitotic/asynchronous - Riboseq372024ProteomicHumanCellLy et al., 202439443796log2(FC) G2/interphase - protein level382024Proteomic Xenopus OocytePeshkin et al., 202440674131log2(FC) MII/GV - protein level392025RNAseq Xenopus OocyteThis paperlog2(FC) MII/GV - sucrose cushion Datasets measuring protein translation/accumulation during M-Phase. Overview of the 39 two-condition omics datasets (CAT1–CAT39). For each dataset, the table reports the year of publication, experimental technique (proteomics, RNA-seq, or poly(A)-tail profiling), model organism (human, mouse, Xenopus), biological system (somatic cells or oocytes), PubMed ID, and the calculated feature tracking translation/protein accumulation Xenopus oocyte handling Xenopus laevis adult females (Centre de Ressources Biologiques Xenopes, CNRS, France) were bred and maintained according to current French guidelines in the IBPS aquatic animal facility (Animal Facility Agreement A75-05-25). All Xenopus experiments were subject to ethical review and approved by the French Ministry of Higher Education and Research (APAFIS #45718-2025021111055055 v2). Xenopus females were anaesthetised (30 min bath in 1 g/l MS222; Sigma). Following surgical removal, ovarian lobes were dissected and incubated with dispase II (0.4 mg/ml, Roche) in buffer M (10 mM HEPES pH 7.8, 88 mM NaCl, 1 mM KCl, 0.33 mM Ca (NO3)2, 0.41 mM CaCl2, and 0.82 mM MgSO4) for 3 h. A further incubation for 1 h with collagenase (0.4 mg/ml; Sigma–Aldrich) was carried out before washing with 2 l of buffer M. Full-grown oocytes, called PRO oocytes, were selected for all experiments and were maintained in buffer M until harvest. To induce meiotic divisions, oocytes were incubated with 1 µM progesterone. Oocytes were scored as nuclear envelope breakdown (NEBD) when pigment rearrangement was first detected at the animal pole. They were collected 3 h after NEBD, when they reach metaphase II (MII). Plasmids and translational reporters Cloning of plasmids was performed using either Gibson assembly or overlap PCR. Plasmids generated for this study were V5-3′UTR-Aurka, V5-3′UTR-Smu1, V5-3′UTR-Cenpe, V5-3′UTR-Cenpa, V5-3′UTR-Cdca3, V5-3′UTR-Prc1, V5-3′UTR-Anln, V5-3′UTR-Kif23, V5-3′UTR-Pan2, V5-3′UTR-CenpjWT, V5-3′UTR-Cenpj∆. Translational mRNA reporters were generated using the mMESSAGE mMACHINE T3 Transcription kit (Invitrogen) and the Qiagen RNeasy kit. A constitutive probe, MBP-GST-flag-HA-V5, was in vitro polyadenylated using the Poly (A) Tailing Kit (Invitrogen, AM1350). Integrity of all reporters was confirmed by agarose gel electrophoresis and quantified using a NanoDrop One. For all translational reporter experiments, oocytes were co-injected with the reporter of interest (0.5 ng in a final volume of 50 µl) and MBP-GST-flag-HA-V5 (0.1 ng in a final volume of 50 nl). Oocytes were either incubated overnight or for at least 1 h before treating with progesterone and collected either in PRO or MII for further processing. Ribosome immunoprecipitation Oocytes were homogenised in Lysis buffer (10 µl/oocyte; 20 mM Tris pH 7.4, 150 mM NaCl, 5 mM MgCl2, 100 µg/ml cycloheximide, 1 mM dithiothreitol, RNase OUT, and Turbo DNase 25 U/ml). A second lysis was carried out on the supernatant by adding 1% Triton X-100 and 1% sodium deoxycholate. The lysate (2 oocytes/sample) was precleared by incubating with washed protein G beads for 1 h with rotation. An input sample was removed prior to starting the preclear step. The RPL0 antibody (MBL Life Science RN004M) was immobilised to protein G beads by incubating for 3 h with rotation. For immunoprecipitation, the precleared lysate was added to the antibody-immobilised beads and incubated overnight with rotation. Beads were washed four times the following day with Urea Wash buffer (20 mM Tris pH 7.5, 2.5 mM MgCl2, 150 mM NaCl, 0.05% (v/v) Tween 20, RNAse OUT, 100 µg/ml cycloheximide, 1.5 mM dithiothreitol, and 1 M urea), then resuspended in 200 µl of Urea Wash buffer. 20 µl of 10% SDS and 5 µl of Proteinase K were added and incubated at 37°C for 75 min. RNA was extracted from input and pellet samples using the Qiagen RNeasy kit. RNA was quantified using the NanoDrop One. Experiments were carried out on three independent biological samples. Library preparation and sequencing Libraries for sequencing were prepared using the NEBNext® Single Cell/Low Input RNA Library Prep Kit for Illumina® and NEBNext® Multiplex Oligos for Illumina® Index Primers Set 1 and Set 2 as per the manufacturer’s instructions. Libraries from both immunoprecipitated samples and inputs were prepared for sequencing. Library concentrations were determined using the Qubit. Next generation sequencing was carried out on a Novaseq X Plus at the iGenSeq platform, Institut du Cerveau, Paris. Quality control of reads was carried out using FastQC. Reads were aligned to the X. laevis genome v10.1 (obtained from Xenbase) using the Hisat2 aligner [15, 16] and counts obtained using htseq-count [17]. Sucrose cushion For sucrose cushion experiments, 50 PRO or MII oocytes were lysed in Lysis buffer (5 µl/oocyte; Incomplete polysome buffer 1X, cycloheximide 100 µg/ml, dithiothreitol 1 mM, RNAse OUT, Turbo DNase 25 U/ml, and RNAse-free water). To complete lysis, 1% Triton X-100 and 1% sodium deoxycholate were added to the sample. 100 µl of lysate was layered on top of a 37% sucrose cushion (cycloheximide 100 µg/ml, SUPERase-In 20 U/ml, 100 mM KCl, 5mM MgCl2, 1 M sucrose, and 20 mM Tris pH 7.5). Centrifugation was carried out using a Beckman TL-100 Ultracentrifuge and TLA 100.2 rotor at 100 000 rpm for 1 h and 11 min, 4°C. After ultracentrifugation, layers were carefully removed, so as not to disturb the pellet. Pellets were resuspended in RLT buffer (Qiagen). RNA was extracted from pellet samples using the Qiagen RNeasy kit as per the manufacturer’s instructions and concentrations quantified using the NanoDrop One (ThermoFisher Scientific). Experiments were carried out on seven independent biological samples. Quantitative PCR For quantitative PCR (qPCR) analysis, RNA was extracted as above and converted to cDNA using the Superscript VI Kit (ThermoFisher Scientific). All primer pairs were designed for targets using the Primer Blast program. The efficiency of each primer pair used was tested using consecutive dilutions of cDNA. For efficiency experiments, a standard curve was generated by plotting the average Ct value obtained against the log of the dilution, and the slope of the line obtained. Efficiency was calculated using the formula: 10^ [(−1/slope) − 1] * 100. Primer sequences and efficiency tests are listed in the supplementary files. For sucrose cushion experiments, cDNA was diluted as 1 in 10, and 1 µl used per reaction. Primers were used at a final concentration of 0.25 µM per reaction. All reactions were carried out in triplicate using the Luna Universal qPCR mix (New England Biolabs) and a CFX Opus 384 Real-Time PCR System (Bio-Rad). The following thermocycling program was used: initial denaturation at 95°C for 60 s, then 40 cycles of 95°C for 15 s and 60°C for 30 s. Melt curve analysis was carried out to assess PCR product specificity. Data were analysed using Bio-Rad CFX Maestro software (v4.1.2433.1219). Relative mRNA expression was calculated using the 2−ΔΔCt method, where the Ct obtained for dctpp1 was used as the housekeeping gene for normalization. SDS–PAGE and western blotting Oocytes were lysed in extraction buffer (80 mM b-glycerophosphate pH 7.3, 20 mM EGTA, 15 mM MgCl2, and 10 µl/oocyte). Lysates were loaded (equivalent of 0.8 oocytes per lane) on 12% polyacrylamide gels and SDS–PAGE carried out as per standard procedures. Gels were transferred to nitrocellulose membranes using a semi-dry transfer system. Primary antibodies and dilutions used are as follows: V5, Invitrogen 46-0705, 1/10 000; Karyopherin, Santa Cruz sc-1863, 1/3000; pT320-PP1, Abcam ab62334, 1/10 000; pMAPK, Cell Signaling 9106, 1/3000; pCdc25, Cell Signalling S216, 1/20 000; Mos, Santa Cruz sc-86, 1/1000; pY15-Cdk1, Cell Signalling 9111S, 1/500. After overnight incubation at 4°C in primary antibodies, nitrocellulose membranes were incubated 1 h at room temperature in corresponding horseradish peroxidase secondary antibodies (Jackson Immunoresearch) used at a dilution of 1/10 000. Nitrocellulose membranes were revealed by chemiluminescence (Pierce). All western blots are representative of at least three different experiments. Antisense oligonucleotide design and microinjection Specific antisense oligonucleotides were designed to target both S and L forms of candidate mRNAs in X. laevis by aligning the two sequences. Two independent antisense sequences were chosen adjacent to the ATG for each candidate mRNA as follows: Fam76b AS1 ACTTGGTGCAGGCGTACA; Fam76b AS2 TCGAACGGGTATCGCT; Cdca3 AS1 TTAGAGTCGGCGGATC; Cdca3 AS2 GAAGGGGTTACTTGGGC. Antisense were microinjected into prophase oocytes (100–150 nM). Oocytes were incubated overnight and treated with progesterone the following morning. Oocytes were then monitored for time to NEBD. For qPCR analysis, oocytes were lysed in Lysis buffer (Incomplete polysome buffer 1×, cycloheximide 100 µg/ml, dithiothreitol 1 mM, RNAse OUT, Turbo DNase 25 U/ml, and RNAse-free water), then with 1% Triton X-100 and 1% sodium deoxycholate. RNA was extracted using the Qiagen RNeasy kit and converted to cDNA using the Superscript VI Kit (ThermoFisher Scientific). qPCR was carried out using specific primers and the cycling conditions outlined above. For western blot analysis, oocytes were lysed with Extraction buffer and analysed as above. Poly (A)-tail length assay Oocytes were lysed as above, first with Lysis buffer, then with 1% Triton X-100 and 1% sodium deoxycholate. RNA was then extracted using the Qiagen RNeasy kit. G/I tailing, reverse transcription and PCR amplification were then carried out using the Poly (A) Tail-Length Assay Kit (Affymetrix), according to the kit instructions. Gene specific primers were designed to amplify between 150 and 170 bp directly upstream from the poly (A) tail. PCR amplicons were run on 2.5% agarose gels and imaged using a UV imager. Images were quantified using ImageJ and a Python script (Appendix A) to measure the average poly (A) tail length. Statistical analysis Statistical analysis was performed using GraphPad Prism (v10.5). Data are presented as the mean ± standard deviation (SD), where results were considered statistically significant if P < 0.05. Asterisks signify that *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001. Results Hierarchical clustering of translation-related datasets reveals meiotic and mitotic translation signatures Over the past few decades, the development of transcriptome-wide techniques to study translation and protein accumulation has resulted in the production of multiple datasets describing translation during cell divisions. We investigated whether these datasets could be combined to obtain a shared translational fingerprint of M-phase. We collected 38 publicly available datasets, now referred to as CAT1-38 (Table 1) that measure protein translation or accumulation during either the G2/M transition of mitosis or the transition from prophase I (PRO) to metaphase II (MII) of oocyte meiosis [18–45]. Datasets were collated from three species: Homo sapiens, Mus musculus, and Xenopus laevis. These high-throughput datasets were obtained using different techniques to measure translation (polysome profiling or ribosome profiling), mRNA polyadenylation (Tail-seq, PAL-seq), or protein abundance (mass spectrometry). Each of these approaches captures a distinct aspect of translation or protein accumulation. Translation-based methods are highly sensitive, but they cannot detect proteins that accumulate as a result of decreased turnover rather than increased translation. Polyadenylation is tightly connected to the control of translation in oocytes and during the first embryonic cell divisions [46]; however, not all aspects of translation are regulated solely by poly (A)-tail length. Additional mechanisms, often mediated by features of the 5′UTR, can act independently of polyadenylation [23, 47]. Measuring protein accumulation by mass spectrometry provides the most direct readout of protein levels, but these techniques are generally less sensitive than RNA-based approaches and may fail to detect low abundance proteins that nevertheless play critical roles in cell division. For each dataset, a feature representing the change in translation/accumulation of a given protein during cell division was calculated (see Materials and methods and Table 1). Additionally, all these methods have different detection ranges, and feature distributions therefore vary (Supplementary Fig. S1A). To enable direct comparison, the feature distribution was standardized as described in the Materials and methods section (Supplementary Fig. S1B). Additional cleaning steps were carried out as described in Materials and methods. CAT18 and 15 were excluded because too few mRNAs/proteins were detected (Supplementary Fig. S1C). CAT6, 9, 10, and 29 were also excluded because several proteins that are well established to accumulate during M-phase (Table 2) were not translated/accumulated as expected (Supplementary Figs S2–S4). Finally, 7089 mRNAs/proteins, which were measured in at least 60% of the selected datasets, were retained for the analysis. Table 2.General M-phase training group. List of proteins included in the General M-phase training set used by ANeR. For each mRNA table reports the functional role, the low-throughput techniques used to demonstrate the translation activation in M-phase, the biological systems in which this regulation was shown, and the species in which supporting evidence has been reported, with representative referencesNameRoleLow throughput techniquesBiological systemSpeciesWestern blotPolyadenylation/translational reporterMitosisMeiosisHumanMouse Xenopus General M-phase proteinsCyclin B1Direct activator of Cdk1Santoni et al., 2024 Frank-Vaillant et al., 1999 Han et al., 2017Santoni et al., 2024 Mendez et al., 2002Yang et al., 2017 Luong et al., 2020Toyoshima et al., 1998 Gong et al., 2010 Stewart and Fang, 2005Santoni et al., 2024 Luong et al., 2020 Frank-Vaillant et al., 1999Toyoshima et al., 1998Gong et al., 2010 Stewart and Fang, 2005Luong et al., 2020 Han et al., 2017 Yang et al., 2017Santoni et al., 2024 Frank-Vaillant et al., 1999 Mendez et al., 2002Cyclin A2Direct activator of Cdk1 and Cdk2Kobayashi et al., 1991Zhang et al. 2017Sheets et al., 1994Gong et al., 2010 Stewart and Fang, 2005Kobayashi et al., 1991Zhang et al. 2017Gong et al., 2010, Stewart and Fang, 2005Zhang et al. 2017Kobayashi et al., 1991Sheets et al. 1994Tpx2Spindle assembly factorBrunet et al., 2008 Bärenz et al., 2013 Chen et al. 2013Chen et al., 2013Stewart and Fang, 2005Brunet et al., 2008 Bärenz et al., 2013 Chen et al. 2013Stewart and Fang, 2005Brunet et al., 2008 Chen et al. 2013Bärenz et al., 2013Cdc20Activator of the APCTaieb et al. 2011 Adhikari et al., 2014 Fang et al., 1998Fang et al., 1998 Stewart and Fang, 2005Taieb et al. 2011 Adhikari et al., 2014Fang et al., 1998 Stewart and Fang, 2005Adhikari et al., 2014Taieb et al., 2011Aurora APlk1 activatorSantoni et al., 2024 Frank-Vaillant et al., 2000 Taguchi et al., 2002Paris et al., 1990Taguchi et al., 2002Santoni et al., 2024 Frank-Vaillant et al., 2000 Paris et al., 1990Taguchi et al., 2002Santoni et al., 2024 Frank-Vaillant et al., 2000 Paris et al., 1990 General M-phase training group. List of proteins included in the General M-phase training set used by ANeR. For each mRNA table reports the functional role, the low-throughput techniques used to demonstrate the translation activation in M-phase, the biological systems in which this regulation was shown, and the species in which supporting evidence has been reported, with representative references To assess whether the protein translation/accumulation values extracted from the collated datasets were sufficient to group datasets derived from similar biological samples, we carried out hierarchical clustering analysis. The datasets segregated into two main clusters: one corresponding to oocytes and the other to somatic cells (Fig. 1A). This distinction is consistent with the two overlapping translation programs known to operate during meiotic divisions. The first regulates progression through cell division, a program shared between mitosis and meiosis; the second controls meiosis-specific events, including the absence of DNA replication between the two meiotic divisions, chromosome rather than chromatid segregation during the first meiotic division, and preparation for early embryonic development through the degradation of maternal mRNAs and the accumulation of proteins [5]. Interestingly, within the oocyte cluster, datasets from mouse and Xenopus oocytes formed two distinct sub-clusters (Fig. 1A). Moreover, within the large cluster of mouse oocyte, datasets measuring translation and polyadenylation grouped closely together, suggesting a strong correlation between polyadenylation changes and translational regulation, while proteomic datasets (CAT26, 28, and 35) formed a separate subcluster (Fig. 1A). In addition to the publicly available datasets, we also carried out our own translatomic study using X. laevis oocytes (CAT39; Table 1). We performed an immunoprecipitation (IP)-based method in which an antibody specific to the ribosomal protein RPL0 was used to pull down ribosomes and associated transcripts in lysates from PRO and MII oocytes. To validate this approach, we performed qPCR in RPL0-IP samples to measure the translation levels of Cyclin B1 (Ccnb1). As expected, the translation of Ccnb1 significantly increased in MII oocytes compared to PRO oocytes (Fig. 1B). Therefore, we performed a RNA-seq analysis of RPL0-IP samples, along with their corresponding total RNA inputs. PCA confirmed that the sequenced RPL0-IP samples clustered into two distinct groups corresponding to PRO and MII stages (Fig. 1C). Importantly, the majority of the proteins known to accumulate during the PRO/MII transition (Tables 2 and 3) were correctly measured in our dataset (Fig. 1D). Furthermore, our CAT39 dataset grouped within the Xenopus oocyte cluster (Fig. 1A). Table 3.Average feature values for training-set members and ANeR candidates. For each gene in the General training group, the oocyte-specific training group, and the set of experimentally tested ANeR predictions, the table reports the average standardized feature value in the mitotic, Xenopus oocyte, mouse oocyte, and human oocyte dataset subcategories. Values correspond to the mean standardized log2 fold change across all CAT datasets within each subcategoryGene knownAverage mitoticAverage XenAverage mouseAverage humanGeneralCcnb12.0148462.0785161.8899470.947007Ccna21.3936470.6930020.9428920.963506Tpx20.6562790.6796561.5878440.418099Cdc201.8949320.9589761.44420.745577Aurka1.5919430.8905780.6301530.309633Oocyte-specificMos01.2036221.7413830.484908Cdc60.277871.5515760.36599-0.36633Slbp0.4856120.6655281.758130.386851Prc11.3326110.6439751.8834950.665025Btg401.8725162.4349840.447856Cnot7-0.860030.1738341.9541131.047793Cnot8-0.638591.019281.7462490.418036ANeR predictionsDcun1d50.8651750.2757451.124110.177385Smu10.4760870.5288521.291066-0.09972Ckap50.7686490.594540.8570040.823418Cenpe0.7262950.6979481.310490.767328Rab140.7622240.6086340.7520630.217016Cenpa1.5221530.2469630.8124090.388808Cdca81.6479760.3222921.2161411.169312Cdca31.5632960.4695361.3235060.444758Prc11.3326110.6439751.8834950.665025Anln1.3140380.9066111.2557860.085868 Average feature values for training-set members and ANeR candidates. For each gene in the General training group, the oocyte-specific training group, and the set of experimentally tested ANeR predictions, the table reports the average standardized feature value in the mitotic, Xenopus oocyte, mouse oocyte, and human oocyte dataset subcategories. Values correspond to the mean standardized log2 fold change across all CAT datasets within each subcategory Development of an ANeR method to identify a translation fingerprint conserved between mitosis and meiosis The clustering analysis indicated that each biological system has a particular translational signature. We therefore asked whether the heterogeneous datasets could nevertheless be integrated to identify conserved components of these programmes. Specifically, we sought to define two fingerprints: a general M-phase fingerprint shared between mitotic and meiotic cells, and an oocyte fingerprint conserved across vertebrate species. Therefore, we developed an ANeR method to identify and rank mRNA/proteins that behave similarly to a training group across multiple omics datasets (Fig. 1E). Our approach required a list of transcripts and their expression levels obtained from two different biological conditions. ANeR ranks candidate transcripts based on their similarity to validated transcripts within a training set (Fig. 1E). To achieve this, a graph/network was built from gene expression data, where nodes represent transcripts and weighted edges represent their Euclidean distance. Two outputs were produced: a first list of the genes ranked by number of neighbours and average Euclidean distance; a second list based on the Pareto frontier that ranks transcripts according to their extracted network attributes, including the number of neighbours, the number of cliques, the size of the maximum clique, and the average Euclidean distance. A first training set, corresponding to a general M-phase group, was defined by selecting proteins whose translation or abundance increases during M-phase in both mitotic and meiotic cells (Table 2). Proteins included in this training set were selected based on the following criteria: their increase in translation or accumulation was demonstrated using low-throughput techniques such as western blot, polyadenylation assays or translation reporters, in both meiosis and mitosis and in at least two of the three species analysed, human, mouse, and Xenopus. This first training set includes Cyclin B1 [7, 18, 48–52, 53] and Cyclin A2 [7, 51, 54–56], which are two direct activators of Cdk1 and Aurora-A [53, 57–59] (Aurka), which is part of the kinase network required for Cdk1 activation, Tpx2 [51, 60–62], which is involved in spindle formation and is also a binding partner of Aurora-A [63, 64, 65], and Cdc20 [51, 66–68], an activator of the anaphase promoting complex, which is required for exit from M-phase in both mitosis and meiosis. Herein, we sorted the features describing translation from the 30 datasets passing the quality selection (Table 1) into subcategories, pertaining to mitotic cells or oocytes. PCA of the different subcategories shows that, as expected, the members of our general M-phase training set clustered in both mitotic and oocyte subcategories (Supplementary Fig. S5A and B). The oocyte subcategory was further divided into mouse or Xenopus, where the training set showed similar clustering (Supplementary Fig. S5C and D). ANeR validation and candidate predictions To measure the robustness of the ANeR method, a permutation-based approach was taken whereby the program iterated 100 times. Each permutation contains features from a different dataset using the same general M-phase training group. To ensure that permutations were not over-sampled towards one subcategory of datasets (mitotic cell, or Xenopus, mouse, and human oocytes), the number of datasets from each subcategory present in each permutation was kept constant. The ranked output of the 100 iterations converged toward the same predictions, which are consistently found with the top 200 best ranked mRNA/proteins (Fig. 2A). Next, we demonstrated the robustness of our method by comparing it to a PCA-based ranking approach (see Materials and methods and Fig. 2B). The PCA-based ranking achieved higher medians and its top predictions include fewer experimentally validated transcripts, which are shown in green in Fig. 2A and B. Candidate prediction by ANeR. (A) Permutation analysis of ANeR rankings using the M-phase training group. For each transcript, its rank among the predicted list was recorded across 100 independent permutations of different combinations of datasets. The plot shows the distribution of ranks for each transcript over these permutations. When a transcript was not ranked among the top 200 candidates in a given permutation, a fixed rank value of 1000 was assigned. Transcripts are ordered on the x-axis according to their median rank across the 100 ANeR iterations. The translation of the transcripts highlighted in green was experimentally validated in this study. (B) Permutation analysis of PCA rankings using the M-phase training group. Transcripts are ordered on the x-axis according to their median rank across the 100 PCA iterations. (C) PCA of all datasets used in ANeR. The transcripts highlighted in green were experimentally validated for translation. (D) GO analysis of the 200 top-ranked ANeR predictions. Blue bars indicate the number of top-ranked predicted transcripts annotated to each GO category, and red bars indicate the expected number of transcripts in each category. Fold change (FC) of enrichment and false discovery rate (FDR) for the enrichment tests are indicated. Biological processes are ordered on the x-axis according to decreasing statistical significance (FDR). After demonstrating the quality of ANeR predictions, we ran it again using the general M-phase training group and the features from all published datasets that passed the quality criteria. PCA plot shows that the top 12 predictions clustered well with the M-phase training set (Fig. 2C). Of these top predictions, several also ranked highly in the validation iterations, as shown in green (Fig. 2A). Furthermore, GO (gene ontology) analysis of the top 200 predictions showed an enrichment on cell cycle-related processes, such as mitotic cell cycle, chromosome segregation, and cytoskeleton organization, indicating that the abundance of multiple important regulators of cell divisions is controlled at the translational level (Fig. 2D). Of the top predictions, we selected 10 to experimentally validate (Supplementary Fig. S6). These included the cell division cycle associated (Cdca) family members Cdca3/Tome-1, which is involved in Wee1 degradation [69], and Cdca8/Borealin, a member of the Chromosomal Passenger protein Complex [70]; the kinetochore-associated centromere proteins (CENP) Cenpe and Cenpa [71–73]; Smu1, a component of the spliceosome and predicted to be necessary for cell cycle progression [74]; Anillin (Anln), a regulator of cytokinesis interacting with myosin II [75, 76]; Prc1, the protein regulator of cytokinesis 1, involved in cytokinesis [77]; Dcun1d5, a regulator of cullin neddylation [78]; the small GTPase Rab14 [79, 80]; and Ckap5, also known as Xmap215 in Xenopus, a regulator of microtubule and spindle dynamics whose downregulation results in mitotic spindle collapse [81, 82], and whose translation has previously been shown to be upregulated during M-phase in Xenopus oocytes [28]. When the average normalized log2FC of the training set and the chosen predictions were compared across the mitotic cell and Xenopus, mouse and human oocyte subcategories, the majority of the candidates were upregulated across these dataset classes, indicating that their ANeR ranking was supported by information from all dataset subtypes (Table 3). Validation of the ANeR predictions in meiotic and mitotic divisions In order to experimentally validate the ANeR predictions, we optimized an independent method to track translation in Xenopus oocytes using the general M-phase training set as positive control. We isolated polysomes on a sucrose cushion and extracted the RNA from the pellet. The RNA was then reverse-transcribed into cDNA and analysed by qPCR (Fig. 3A). The levels of all training set members were significantly increased in MII pellet samples, as compared to PRO samples (Fig. 3B), confirming their translation activation during meiotic divisions and demonstrating the efficacy of this method in tracking changes in translation in Xenopus oocytes. We used Psmg1 as negative control since its translation is decreased during meiotic divisions in our translatomic dataset. Indeed, our experimental approach revealed that Psmg1 translation decreases in MII as compared to PRO (Fig. 3B). We then applied the same assay to the ANeR predictions and confirmed increased translation during meiotic divisions for 8 of the 10 candidates tested (Fig. 3B). Smu1 could not be assessed because suitable qPCR primers could not be optimized, whereas Dcun1d5 did not show a significant increase in translation. Experimental validation of ANeR. (A) Schematic of the sucrose cushion–qPCR workflow. (B) qPCR analysis of M-phase training group mRNAs (red), the negative control Psmg1 (grey), and ANeR predicted candidates (green) in sucrose cushion pellet samples. Graphs show log2 (2–ΔΔCt) values in PRO and MII oocytes after normalization to reference transcripts. Bars represent mean ± SD; asterisks indicate statistical significance (one sample t-test). (C) Schematic of translational reporter constructs. Candidate 3′UTRs are cloned downstream of a GST–V5 coding sequence (reporter). A constitutive probe encoding an MBP-GST-V5 fusion protein is in vitro polyadenylated. Both are co-injected into oocytes. (D–F) PRO oocytes were co-injected with the constitutive probe and a reporter bearing the 3′UTR of the indicated mRNAs. PRO (no progesterone) and MII (+ progesterone) oocytes were analysed by western blot with antibodies against V5 (expression of the two fusion proteins), karyopherin (loading control), pT320-PP1, and pMAPK (MII markers). Images are representative of at least three independent biological replicates. Quantification of relative V5 reporter and constitutive probe expression in PRO and MII oocytes was performed by densitometric analysis (n = 3 or 4). Bars represent mean ± SD; asterisks indicate statistical significance (two-tailed t-test). We next asked whether the 3′UTRs of selected candidates were sufficient to confer translational activation. We previously established a translational reporter assay consisting of the co-injection of two mRNAs, the first encoding a GST-V5 protein under the translational control of the 3′UTR of the target, and the second encoding a MBP-GST-V5 protein whose translation is maintained constant by in vitro polyadenylation [53] (Fig. 3C). We have used this method to show that Cyclin B1 translation is activated during meiotic divisions [53]. Thus, we generated a reporter incorporating the 3′UTR of Aurka downstream of the GST-V5 open reading frame. PRO-injected oocytes were stimulated with progesterone to trigger meiotic resumption. This led to an increase in reporter translation (Fig. 3D), confirming that Aurka translation increases during meiotic divisions and the information contained in the Aurka 3′UTR is sufficient to control its translation. The phosphorylation of PP1 at T320 (a Cdk1 substrate) and of MAPK was used to confirm Cdk1 and MAPK activation and the correct PRO-MII transition of the oocytes (Fig. 3D). We subsequently generated reporters for 6 ANeR predictions, all of which showed increased translation during the PRO to MII transition (Fig. 3E and F). Thus, two independent experimental approaches confirmed that ANeR efficiently prioritizes mRNAs whose translation is activated during meiotic maturation. We next evaluated whether the general M-phase fingerprint obtained by ANeR was supported by independent somatic cell-cycle proteomes. ANeR prediction tiers were compared with three published proteomic datasets that quantified protein abundance dynamics across the cell cycle in cultured somatic cells and were not used for training [83, 84, 85]. Proteins were assigned to the cell-cycle phase in which their abundance peaked using the annotations provided in the original studies (Supplementary Fig. S7). In the complete datasets, M-phase-peaking proteins accounted for 24% of proteins in Herr et al., 17% in Becher et al., and 25% in Rega et al. when G2-M and M/early-G1 windows were combined. This proportion increased across the ANeR prediction tiers to 36%–67%, 29%–100%, and 36%–50%, respectively. The enrichment was strongest in the highest-confidence tiers and remained detectable in the broadest ≥ 1-neighbour tier. Thus, independent somatic cell-cycle proteomes support the conservation of the ANeR general M-phase fingerprint in mitotic divisions and indicate that prediction quality extends beyond the top-ranked candidates. ANeR predicts an oocyte-specific signature of translation Oocyte meiotic maturation is accompanied by extensive translational reprogramming that controls both entry into M-phase and subsequent progression through the meiotic divisions. This programme includes proteins required for Cdk1 activation as well as factors involved in spindle assembly, maternal mRNA degradation, and preparation for embryonic development. Because this temporal control of protein synthesis is a deeply conserved feature of meiosis, we used ANeR to identify an evolutionarily conserved oocyte fingerprint of translation, characteristic of mRNAs whose translation is specifically activated during meiotic divisions. For this purpose, we designed an oocyte-specific training group of proteins accumulating specifically during meiotic divisions (Table 4). This second training set consists of proteins accumulating/translated only in oocytes in at least two of the three species analysed, human, mouse, and Xenopus. This group includes Mos [18, 19, 49, 50, 53], a protein essential for the second arrest of oocytes (in MII in vertebrates), and proteins like Btg4 [86–88], Cnot7 [86–89] and Cnot8 [89], which degrade maternal mRNAs as part of the preparatory program for embryonic development post-fertilization, Prc1, which was also present in the M-phase training group. Additionally, this group contains proteins required in early embryogenesis to regulate initial cell divisions in the absence of zygote transcription, such as Cdc6, a component of the pre-replicative complex [90–92, 93], and SLBP, a RNA-binding protein that controls maternal histone mRNA translation [94, 95]. ANeR was run using the oocyte-specific training group (Table 4) and the oocyte clusters identified in our hierarchical clustering analysis (Fig. 1A). Table 4.Oocyte-specific M-phase training group. List of proteins included in the oocyte-specific M-phase training set used by ANeR. For each mRNA the table reports the functional role, the low-throughput techniques used to demonstrate its translation activation during meiotic maturation, the biological system in which this regulation was observed, and the species in which supporting evidence has been reported, with representative referencesNameRoleLow throughput techniquesBiological systemSpeciesWestern BlotPolyadenylation/translational reporterMitosisMeiosisHumanMouse Xenopus Oocyte-specific M-phase proteinsMosOocyte specific activator of MAPK pathwaySantoni et al., 2024, Frank-Vaillant et al., 1999, Han et al., 2017Santoni et al., 2024, Mendez et al., 2002Luong et al., 2020Not expressedSantoni et al., 2024, Luong et al., 2020, Frank-Vaillant et al., 1999Luong et al., 2020, Han et al., 2017, Yang et al., 2017Santoni et al., 2024, Frank-Vaillant et al., 1999, Mendez et al., 2002Cdc6Component of the DNA replication machineryWhitmire et al., 2002, Lemaitre et al., 2002, Daldello et al., 2015Yan et al., 1998Whitmire et al., 2002, Lemaitre et al., 2002, Daldello et al., 2015, Lemaitre et al., 2004Yan et al., 1998Lemaitre et al., 2004Whitmire et al., 2002, Lemaitre et al., 2002, Daldello et al., 2015SLBPRNA-Binding proteinAllard et al., 2002, Yang et al., 2010, Wang et al., 1999Whitfield et al., 2000Allard et al., 2002, Yang et al., 2010Allard et al., 2002, Yang et al., 2010Wang et al., 1999Prc1Regulator of cytokinesisLi et al. 2021, Bärenz et al., 2013,Jiang et al., 1998Li et al. 2021, Bärenz et al., 2013,Jiang et al., 1998Li et al. 2021Bärenz et al., 2013,Btg4Regulator of mRNA degradationYu et al., 2016, Rong et al.2019Sha et al., 2017, Rong et al.2019Yu et al., 2016, Rong et al.2019, Sha et al., 2017Yu et al., 2016, Rong et al.2019, Sha et al., 2017maternal mRNA degradation is conservedCnot7Regulator of mRNA degradationYu et al., 2016, Rong et al.2019, Soeda et al., 2023Yu et al., 2016, Rong et al.2019, Sha et al., 2017, Soeda et al., 2023Yu et al., 2016, Rong et al.2019, Sha et al., 2017, Soeda et al., 2023maternal mRNA degradation is conservedCnot8Regulator of mRNA degradationSoeda et al., 2023Soeda et al., 2023Soeda et al., 2023maternal mRNA degradation is conserved Oocyte-specific M-phase training group. List of proteins included in the oocyte-specific M-phase training set used by ANeR. For each mRNA the table reports the functional role, the low-throughput techniques used to demonstrate its translation activation during meiotic maturation, the biological system in which this regulation was observed, and the species in which supporting evidence has been reported, with representative references To assess the stability of the model, we performed 100 iterations of ANeR, each using a different permutation of the training categories, and confirmed a strong convergence of the outputs across runs (Fig. 4A). Interestingly, many of the predictions using the oocyte training set were also either predicted using the general training set (Fig. 4A; green-Top100, yellow-Top 200), or members of the general training set (Fig. 4A, red). We compared the robustness of ANeR predictions with that of PCA and found that the median values were again higher for PCA (Fig. 4B). A final run was carried out using all oocyte datasets and the clustering of training and predicted transcripts was observed by PCA (Fig. 4C). Again, many of the top 200 predictions overlapped with validated candidate predictions from the general fingerprint, such as Smu1 and Cenpe, while 3 members of the general M-phase training group (Cyclin B1-Ccnb1, Cdc20, and Tpx2) were also predicted (Fig. 4C). Similarly, GO analysis revealed functional categories consistent with general cell cycle processes but also with meiotic progression and early embryonic development (Fig. 4D). ANeR predictions using the oocyte-specific training group. (A) Permutation analysis of ANeR rankings using the oocyte training group. For each transcript, its rank among the candidate list was recorded across 100 independent permutations of different dataset subgroups. The plot shows the distribution of ranks for each transcript over these 100 permutations. When a transcript was not ranked among the top 200 candidates in a given permutation, a fixed rank value of 1000 was assigned. Transcripts are ordered on the x-axis according to their median rank across the 100 ANeR iterations. Red: transcripts belonging to the M-phase training group; green or yellow: transcripts ranked in the top 100 or 200, respectively, in the M-phase general fingerprint. (B) Permutation analysis of PCA rankings using the oocyte training group. Transcripts are ordered on the x-axis according to their median rank across the 100 PCA iterations. Red: transcripts belonging to the M-phase training group; green, yellow, or blue: transcripts ranked in the top 100, 200, or >200, respectively, in the M-phase general PCA predictions. (C) PCA of the oocyte-cluster datasets used in ANeR. Transcripts belonging to the oocyte-specific training group and the corresponding ANeR-predicted candidates are labelled in blue and pink, respectively. (D) GO analysis of the top-ranked ANeR predictions obtained with the oocyte-specific training group. Bars indicate the number of genes in the prediction set and the expected number for each GO term. The fold change (FC) of enrichment and false discovery rate (FDR) for each GO term is indicated next to the corresponding bar. Biological processes are ordered on the x-axis according to decreasing statistical significance (FDR). ANeR identifies previously unrecognized regulators of meiotic M-phase entry To investigate whether the translation of ANeR predicted proteins control meiotic divisions, we selected two predicted candidates, Fam76b and Cdca3, from the oocyte fingerprint, for further analysis. Fam76b is a nuclear speckle-localized protein, highly conserved among species, whose biological function remains poorly characterized, although it has been implicated in inflammatory signalling [96] and mRNA stabilization [97]. Cdca3, also called Tome-1, is a F-box-like protein which is required for entry into mitosis through its participation to the E3 ligase complex that degrades the Cdk1 inhibitory kinase, Wee1 [69]. Because ANeR predicted strong upregulation of both proteins during the oocyte PRO-MII transition, we tested whether preventing their accumulation affected meiotic divisions. Two distinct antisense oligonucleotides were designed for each protein. Since most of the genes in X. laevis are present in two copies due to an event of genome duplication [25], the antisense were designed to target both genes (Supplementary Fig. S8A) We first assessed that the antisense oligonucleotide led to the degradation of the target mRNAs (Fig. 5A). Antisense-injected oocytes were then stimulated by progesterone and the PRO-MII transition was ascertained by the appearance of a white spot at the animal pole of the oocytes, corresponding to the nuclear envelope breakdown (NEBD; Fig. 5B) At the time when 50% of control oocytes completed NEBD, <20% of the antisense-injected oocytes underwent the nuclear envelope breakdown (Fig. 5C). This delay in meiotic maturation induced by the antisense injection was reproducible in four independent females (Fig. 5B and Supplementary Fig. S8B–D). We analysed the molecular markers of the two pathways controlling the PRO-MII transition, Cdk1, and MAPK: Y15 dephosphorylation of Cdk1, S287 dephosphorylation of the Cdk1 phosphatase, Cdc25, T320 phosphorylation of the Cdk1 substrate, PP1, accumulation of Mos and MAPK phosphorylation (Fig. 5D). This molecular analysis demonstrates that Cdk1 and MAPK were not activated in most of the antisense-injected oocytes at time when 50% of control oocytes completed NEBD (Fig. 5D). Therefore, inhibition of Fam76b and Cdca3 synthesis is able to strongly delay the activation of the Cdk1 and MAPK pathways, indicating that both proteins are involved in meiotic progression upstream of these signalling cascades. These findings demonstrate that ANeR can predict proteins whose translation increases and that control the activation of Cdk1 during meiotic divisions. Depletion of two ANeR candidate mRNAs in oocytes delays meiotic progression. (A) Antisense oligos (AS) complementary to Fam76b (left) and Cdca3 (right) mRNA were microinjected into Xenopus oocytes. After an overnight incubation, levels of depletion were quantified following qPCR analysis. Graphs show 2–ΔΔCt values in non-injected and AS-injected oocytes after normalization to reference transcripts. Bars represent mean ± SD; asterisks indicate statistical significance (one sample t-test). (B) PRO oocytes were injected or not with AS against Fam76b and Cdca3 and then treated with progesterone. NEBD time-course was scored. Graph shows nonlinear regression curve fit of time to NEBD following progesterone treatment (non-injected) and four representative experiments of AS injection. On the right, pictures of PRO and MII oocytes. (C) Percentages of NEBD oocytes in AS conditions were interpolated from the nonlinear regression curve fit at t50% (control), i.e. the time for 50% of control oocytes to reach NEBD. (D) Non-injected or AS-injected oocytes (left: Fam76b; right: Cdca3) were collected at PRO and t50% (control). If completed, NEBD is indicated. To monitor the PRO-MII transition, oocytes were analysed by western blot with antibodies against Cdk1 markers: pCdc25, pT320-PP1, pY15-Cdk1; and MAPK markers: pMAPK and Mos. Karyopherin is used as a loading control. Images are representative of at least three independent biological replicates. ANeR defines an oocyte fingerprint with conserved 3′UTR regulatory features A hallmark previously associated with mRNAs whose translation increases during oocyte meiotic progression is the presence of AU-rich motifs within the 3′UTR [28, 42]. We therefore first asked whether the ANeR-predicted oocyte fingerprint retained this established feature of mRNAs that are translationally activated during meiotic maturation. We quantified this property by calculating the density of every possible 6-mer motif across the 3′UTRs of Xenopus, mouse, and human transcriptomes. Comparing the top 200 ANeR predictions to the full transcriptome showed a striking enrichment of AU-rich 6-mers in the oocyte fingerprint mRNAs, not only in Xenopus, as previously observed [28], but also in mouse and human mRNAs (Fig. 6A). Two RNA-binding proteins, CPEB1 and Zar-family proteins, play central roles in the control of translation activation [98, 50, 99, 100]. CPEB1 recognizes cytoplasmic polyadenylation elements (CPEs) and promotes poly (A)-tail elongation and translation activation of key meiotic regulators such as Cyclin B1 and Mos [101]. Zar-family proteins, on the other hand, bind AU-rich sequences of distinct composition and have been implicated in translational repression and activation during meiosis resumption [87, 99, 100]. Consistent with these known functions, both CPE and Zar-binding motifs were significantly enriched in the 3′UTRs of mRNAs belonging to the oocyte fingerprint (Fig. 6B and Supplementary Fig. S9A–F). Notably, the density of CPE motifs was higher than that of Zar-binding sites, supporting the model that CPEB1 acts through clustered cis-elements [102, 103, 19]. As a control, no enrichment of the splicing factor ESRP1, which binds to UG-rich exonic motifs [104], was detected in the ANeR-predicted mRNAs compared to the general transcriptome (Supplementary Fig. S9G–L), confirming the specificity of the identified regulatory signatures. These results confirm that the ANeR predictions recover known 3′UTR features and elements controlling translation during meiotic divisions, demonstrating the quality of our predictions and highlighting the evolutionary conservation of this regulatory logic. 3′UTR motif analysis in ANeR-predicted transcripts. (A) Hexamer enrichment analysis in 3′UTRs of ANeR top predictions. Volcano plots show hexamer (6-mer) enrichment in 3′UTRs from the top ANeR-predicted transcripts compared with all 3′UTRs for Xenopus, mouse and human. Each dot represents a 6-mer. The x-axis shows log2 fold change of 6-mer frequency (top predictions/all 3′UTRs), and the y-axis shows –log10 of the FDR. (B) Density of CPE motifs (left) and Zar-binding motifs (right) in 3′UTRs. Bar plots show the difference in the percentage of transcripts between the top-prediction set and all 3′UTRs as a function of the density of CPE or Zar-binding motifs (number per 1000 nt) of 3′UTR (0, 1, 2, 3, 4, or ≥ 5) for Xenopus, mouse, and human. The P-values for the comparison between the two transcript sets are indicated (Chi-square test for trend). (C) Positional arrangement of CPE and PAS motifs. The schematics define the positional windows used to score two motif arrangements proposed to repress translation in PRO (rule 1) and activate translation in MII (rule 2). Bar graphs show the percentage of transcripts in all 3′UTRs (black bars) and in the top-prediction set (red bars) that satisfy rule 1, rule 2, or both (rule 1 + 2) for Xenopus, mouse, and human. The total number of transcripts analysed in each group is indicated above the graphs; P-values for the comparisons between all 3′UTRs and top predictions are shown (Fisher’s exact test, two-sided). The position and number of CPE are also crucial in controlling the timely activation of translation [102]. Analysing the CPE distribution in Cyclin B mRNAs, two rules for the regulation of translation during meiotic divisions were proposed [102]. Translational repression in PRO oocytes requires two CPEs separated by no more than 50 nucleotides (rule 1), whereas translational activation is promoted by a CPE located within 100 nucleotides of the polyadenylation signal (PAS) (rule 2) (Fig. 6C) [102]. At the time when these rules were defined, there were no genome wide analyses of translation in oocytes, making it difficult to verify how widely these rules were applicable to the oocyte translational program. Using the transcripts predicted to be translated during meiotic divisions by ANeR, we showed that both rules are better represented among the predicted mRNAs, as compared to all mRNAs in the transcriptome (Fig. 6C). Hence, the use of ANeR allowed us to estimate the penetrance of the CPE code in the transcriptome, and demonstrate that this regulatory code is well conserved among vertebrates. While the enrichment of canonical CPE/PAS configurations validated the overall fingerprint, ANeR also identified translationally activated mRNAs that do not conform to these rules. Pan2 lacks a CPE in its 3′UTR, yet its 3′UTR was sufficient to drive increased reporter translation during meiotic maturation (Fig. 7A), providing experimental evidence for CPE-independent translational activation. Cenpj 3′UTR contains only one atypical CPE whose position does not satisfy the canonical CPE/PAS configurations (Fig. 7B). However, its 3′UTR contains two GU-rich 6-mer motifs that are found enriched in the oocyte fingerprint, particularly in human (Fig. 6A and B). Deletion of these motifs significantly reduced the increase in reporter translation during the PRO-MII transition (Fig. 7B), indicating that GU-rich elements contribute to Cenpj translational activation. Translation activation of Pan2 and Cenpj is independent of the CPE rules. PRO oocytes were co-injected with two mRNAs: a constitutive control reporter and a reporter bearing the 3′UTR of Pan2 (A) or Cenpj (B), either wild-type or mutated. Oocytes were then treated or not with progesterone. Protein extracts of PRO and MII oocytes were analysed by western blot using antibodies against V5 to detect the constitutive probe and each 3′UTR reporter, karyopherin as a loading control, and pT320-PP1 and pMAPK as MII markers. Images are representative of at least three independent biological replicates. Reporter levels were measured by densitometry, normalized to the constitutive probe. Bars represent mean ± SD; asterisks indicate statistical significance between PRO and MII (two-tailed t-test). These findings indicate that CPE-based regulation alone cannot fully explain translational control during meiotic divisions, and that GU-rich 6-mer motifs contribute to the translational activation of a subset of ANeR-predicted mRNAs. Translation activation occurs through polyadenylation-dependent or -independent mechanisms during meiotic divisions Cytoplasmic mRNA polyadenylation is the main known mechanism controlling translation in oocytes and during the early embryo development. Interestingly, changes in poly (A)-tail length have been shown to strongly correlate with translational efficiency during the early embryonic cell divisions of both Xenopus and zebrafish [46]. We compared the changes in translation of mRNAs that were measured in all four translation datasets available (CAT7, 12, 25, and 39) with changes in the length of poly (A)-tail measured in both Xenopus CAT13 and 34 (Fig. 8A). Of note, while a highly significant correlation was observed between the two measurements (Pearson coefficient: 0.712, P-value < 0.0001), the translation of several mRNAs is activated without apparent changes in their poly (A)-tail length (Fig. 8A). A binning analysis revealed that a 25 nt poly (A) extension is sufficient to increase the translation of mRNAs (Supplementary Fig. S10A). Similar results were obtained in mouse oocytes (Supplementary Fig. S10B and C). Polyadenylation-dependent and -independent activation of translation. (A) Scatter plot comparing poly (A)-tail length changes and normalized translation changes for Xenopus mRNAs. For each mRNA detected in all datasets, polyadenylation change (x-axis; MII–PRO) was calculated as the mean change across the two poly (A)-tail datasets (CAT13 and CAT34). Normalized translation change (y-axis; MII/PRO) was calculated as the mean standardized log2 fold-change across the four translation datasets (CAT7, CAT12, CAT25, and CAT39). Each grey dot represents one mRNA. The red line shows the linear regression with its interval of confidence (black dashed lines). (B) Schematic of the poly (A)-tail length assay. PRO and MII oocytes were lysed and total RNA was extracted. Poly (A)-tails were labelled by G/I tailing, then reverse-transcribed using a universal primer, and PCR amplified with a gene-specific forward primer and the universal reverse primer. PCR products were resolved on agarose gels and quantified to obtain poly (A)-tail length distributions. (C, E) Representative agarose gels showing PCR products from the poly (A)-tail assay for Ccnb1 (C) and Tpx2 (E) in PRO and MII oocytes. Poly (A)-tail length distributions for Ccnb1 and Tpx2 are represented by curves showing the relative frequency of molecules as a function of poly (A)-tail length obtained from gel quantification. (D, F, G) Quantification of average poly (A)-tail length for Ccnb1 (D), Tpx2 (F) and Ccna2, Cdc20, Aurka, Kif23, and Prc1 (G) in PRO and MII oocytes. Average tail lengths were calculated from the poly (A)-tail distributions derived from independent biological replicates. Bars represent mean ± SD; asterisks indicate statistical significance (two-tailed t-test). This multi-dataset comparison allowed us to position individual ANeR candidates within the global relationship between translation changes and poly (A)-tail length changes and assess whether they are regulated by polyadenylation (Fig. 8A). While polyadenylation of Cyclins B1 and A2, Cdc20, Aurka, Cenpe, and Cdca3 increased by around 25 nt or more, the length of the poly (A)-tail of Tpx2 and Prc1 did not change. Another protein that shows similar patterns of increased translation without any change in polyadenylation is the kinesin family member, Kif23 (Fig. 8A). Kif23 was among the top predictions of the mitotic fingerprint (Fig. 2). We experimentally confirmed that Kif23 translation was activated during meiotic divisions using two approaches, the sucrose cushion-qPCR method (Supplementary Fig. S10D) and the translation reporter assay (Supplementary Fig. S10E). We next experimentally assessed poly (A)-tail length using a G/I tailing assay on mRNAs from the two transcript classes predicted to undergo poly (A)-tail extension or not (Fig. 8B). We implemented a python-based pipeline to quantify the distribution of lengths of the poly (A)-tails from the quantification of DNA electrophoresis images, and calculated the average length for each condition (see Appendix B—Step-by-Step process of poly (A)-tail quantification). We first applied our pipeline to the Cyclin B1 (Ccnb1) mRNA, whose extension of poly (A)-tail during meiotic divisions was previously demonstrated in Xenopus and mouse oocytes [52, 55, 105]. As expected, Cyclin B1 showed a large and reproducible increase in poly (A)-tail length of around 90 nt (Fig. 8C and D). Using the same method, we confirmed that the average length of the poly (A)-tail of Tpx2 did not change (Fig. 8E and F). Similarly, and in agreement with our prediction, the poly (A)-tail length of Cdc20, Aurka, and Cyclin A2 (Ccna2) increased by around 55, 50, and 65 nt, respectively, during the PRO-MII transition (Fig. 8G). Importantly, the length of the poly (A)-tails of Prc1 and Kif23 was not significantly altered during meiotic divisions (Fig. 8G), even though the translation of Tpx2 (Fig. 3B), Prc1 (Fig. 3B, F), and Kif23 (Supplementary Fig. S10D and E) increased during the PRO-MII transition. Together, these results demonstrate that ANeR method identifies M-phase-regulated transcripts whose translational activation can occur either with or without detectable poly (A)-tail extension. Discussion Here, we developed ANeR that ranks transcripts according to their proximity to a curated training group across heterogeneous omics datasets. We first applied ANeR to mitotic and meiotic datasets to define a conserved M-phase translational fingerprint shared between the two types of division. We then applied ANeR specifically to vertebrate oocyte datasets to define a conserved meiotic translational fingerprint, which led to the identification of previously uncharacterized regulators required for timely Cdk1 activation and to further characterization of canonical and alternative mechanisms of translational control. Because mitosis and meiosis share a common evolutionary origin and use Cdk1 as a master M-phase regulator, we reasoned that part of the translational program should be shared between them. In line with this idea, ANeR identified a conserved M-phase translational fingerprint shared by mitotic cells and meiotically maturing oocytes. This fingerprint was supported by experimental validation in Xenopus oocytes and by independent benchmarking in somatic cell-cycle proteomic datasets. An important question raised by these observations is whether translational activation during mitosis and meiosis relies on common regulatory mechanisms or on distinct mechanisms that converge on a similar M-phase translational output. These transitions occur in different cellular contexts. In proliferating somatic cells, mitotic entry is embedded within a continuous cell cycle, whereas oocytes enter M-phase after a prolonged prophase arrest during which maternal mRNAs are stored in a translationally repressed state [5]. Once meiosis resumes, however, the transition to M-phase occurs over a time scale of hours, providing a window for extensive remodeling of pre-existing mRNPs and translationally silent transcripts. In oocytes, this remodeling involves the release of maternal mRNAs from repressive granules [106–109]. Interestingly, somatic cells also contain repressive RNA granules, such as P-bodies, which are dissolved during the G2/M transition [110]. This parallel suggests that, despite potential differences in granule composition, mitotic and meiotic cells may share a common regulatory principle based on the remodeling of repressive mRNP assemblies, thereby promoting translation of cell-division regulators. Our work pinpoints mRNAs whose translation is upregulated during both mitotic and meiotic divisions, providing a framework to investigate whether their mechanisms of translational activation are common or divergent in the two contexts. We next applied ANeR to a more focused question: translation during meiotic divisions in vertebrate oocytes. Using an oocyte-specific training group and meiotic datasets from Xenopus, mouse, and human oocytes, ANeR defined a conserved meiotic translational fingerprint. We functionally exploited this meiotic fingerprint by selecting two candidates: Cdca3 and Fam76b. Depletion of either Cdca3 or Fam76b mRNA delayed meiotic resumption and activation of Cdk1 in Xenopus oocytes. These results extend the analysis beyond the identification of translationally regulated candidates and show that ANeR can pinpoint previously uncharacterized factors required for the timely activation of the meiotic G2/M transition. Future work will be required to define the molecular mechanisms by which Cdca3 and Fam76b contribute to the control of Cdk1 activation. Cdca3 is an F-box protein previously shown to control Wee1 degradation and to regulate the G2/M transition in somatic cells [69]. However, the meiotic function of Cdca3 is unlikely to depend on Wee1, which is not expressed during the G2 arrest in Xenopus oocytes [111]. Cdca3 may control the stability of other Cdk1 regulators during this transition, emphasizing the importance of proteome remodeling during the meiotic G2/M transition. Fam76b function remains largely unknown, but recent evidence supports a role in stabilizing specific mRNAs [96]. Hence, Fam76b may contribute to Cdk1 activation by controlling the stability or availability of maternal transcripts whose translation is required for meiotic progression. Meiotic translation in vertebrate oocytes is largely controlled by 3′UTR elements and cytoplasmic polyadenylation. Accordingly, the CPE/PAS rules proposed by Méndez et al. [102] accounted for a substantial fraction of the ANeR-predicted meiotic translational program in Xenopus and mouse. In humans, these rules remained enriched among ANeR predictions, but their absolute penetrance was lower, indicating that they account for a smaller fraction of the predicted translational program and demonstrating that CPE-independent mechanisms also contribute to its regulation. Similarly, during meiotic divisions, the extent of polyadenylation of an mRNA usually dictates its translational activity, and changes in poly (A)-tail length are strongly correlated with changes in translation in Xenopus and mouse oocytes. However, neither CPE/PAS configuration nor poly (A)-tail extension fully explain the ANeR-predicted meiotic program. Our analysis pinpointed candidates that can be used to investigate these exceptions. Indeed, we show that Pan2 is translationally activated despite lacking canonical CPEs and that GU-rich elements contribute to Cenpj translational activation. We also identified several ANeR-predicted mRNAs, such as Tpx2, Kif23, and Prc1, which are robustly translated without a concomitant poly (A)-tail extension. These observations, together with the recent identification of poly (A)-independent translational repression in prophase oocytes, support the idea that poly (A)-tail length and canonical CPE/PAS rules are major but not exclusive determinants of translational status during oocyte maturation [42]. ANeR provides a rich source of candidates whose investigation is likely to uncover previously unknown principles governing translational regulation that are independent of polyadenylation and CPE/PAS motifs. Together, these findings establish ANeR as a strategy for extracting conserved biological information from heterogeneous datasets and converting it into experimentally testable predictions. Beyond M-phase translation, ANeR can be applied to other biological questions, provided that a curated training group and datasets measuring changes between two conditions are available. Thus, ANeR offers a flexible approach to identify co-regulated molecules across diverse experimental systems and datasets. Supplementary Material Acknowledgements Author contributions: Sarah B. Eivers (Data curation [equal], Formal analysis [equal], Investigation [equal], Methodology [equal], Supervision [equal], Writing – original draft [equal]), Axelle Baumgartner (Formal analysis [supporting], Investigation [supporting]), Marcio Dorn (Supervision [supporting]), Catherine Jessus (Writing – original draft [equal], Writing – review & editing [equal]), Juliana Silva Bernardes (Conceptualization [equal], Funding acquisition [equal], Methodology [equal], Software [equal], Supervision [equal], Writing – original draft [equal]), and Enrico Maria Daldello (Conceptualization [equal], Data curation [equal], Formal analysis [equal], Funding acquisition [equal], Methodology [equal], Project administration [equal], Supervision [equal], Writing – original draft [equal]) Contributor Information Sarah B Eivers, Sorbonne Université, CNRS, Development, Adaptation and Ageing, Dev2A, 75005 Paris, France; Sorbonne Université, CNRS, Institut de Biologie Paris-Seine, IBPS, 75005 Paris, France. Axelle Baumgartner, Sorbonne Université, CNRS, Development, Adaptation and Ageing, Dev2A, 75005 Paris, France; Sorbonne Université, CNRS, Institut de Biologie Paris-Seine, IBPS, 75005 Paris, France. Marcio Dorn, Institute of Informatics and Centre for Biotechnology, Federal University of Rio Grande do Sul, Porto Alegre 91501-970, Brazil. Catherine Jessus, Sorbonne Université, CNRS, Development, Adaptation and Ageing, Dev2A, 75005 Paris, France; Sorbonne Université, CNRS, Institut de Biologie Paris-Seine, IBPS, 75005 Paris, France. Juliana Silva-Bernardes, Sorbonne Université, CNRS, AD2M-UMR7144, Station Biologique de Roscoff, 29680 Roscoff, France. Enrico Maria Daldello, Sorbonne Université, CNRS, Development, Adaptation and Ageing, Dev2A, 75005 Paris, France; Sorbonne Université, CNRS, Institut de Biologie Paris-Seine, IBPS, 75005 Paris, France. Supplementary data Supplementary data is available at NAR online. Conflict of interest The authors declare no conflict of interest. Funding This work was supported by the National Centre for Scientific Research (CNRS) and Sorbonne University, the National Research Agency (ANR) (ANR-23-CE12-0045-01 to EMD, Sorbonne University Emergence grant to EMD), the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - STICAMSUD (88881.522073/2020–01) to M.D. and J.B., the ARC foundation grant ARCPJA2023080006901 to E.M.D., I-Bio postdoctoral fellowship to E.M.D. and J.B. Funding to pay the Open Access publication charges for this article was provided by ANR-23-CE12-0045-01. Data availability Supplementary raw data are deposited on Zenodo (DOI: 10.5281/zenodo.22125232). The raw RNA-seq data generated in this article are available on the NCBI Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra/) under the BioProject ID PRJNA1375765. ANeR code has been uploaded to Zenodo (DOI: 10.5281/zenodo.22079137). References Associated Data Supplementary Materials Data Availability Statement Supplementary raw data are deposited on Zenodo (DOI: 10.5281/zenodo.22125232). The raw RNA-seq data generated in this article are available on the NCBI Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra/) under the BioProject ID PRJNA1375765. ANeR code has been uploaded to Zenodo (DOI: 10.5281/zenodo.22079137).