Julien Yann Dutheil

The heterogeneity of expression levels among genetically identical cells, termed gene expression noise, is a property of the gene expression process whose importance in the biology of organisms and their evolution is increasingly recognized. Measuring gene expression noise requires single-cell expression data, as obtained from single-cell RNA sequencing (scRNASeq). Its estimation, however, is challenging owing to (i) the presence of technical noise in addition to biological noise, and (ii) the heterogeneity of cell types in the sampled population. We propose a maximum-likelihood framework to infer biological noise from scRNASeq data, while accounting for technical noise, dropout probabilities, and distinct cell sequencing depths. We demonstrate the parameter identifiability using simulations and that the resulting noise estimates are uncorrelated from the mean gene expression, and therefore do not need extra correction in downstream analyses, easing intra- and inter- genome comparisons. Using two technical replicates of scRNASeq data from the wild yeast Saccharomyces paradoxus, we show that expression noise can be inferred in a reproducible manner.

Small regulatory RNAs (sRNAs) are key drivers of bacterial adaptation to environmental fluctuations, including iron and manganese restriction imposed by the host. This study explored the repertoire of sRNAs produced by the human pathogen Staphylococcus aureus in response to metal limitation. Two sRNAs were identified, S1077 and ZinS, that were regulated by zinc (Zn) availability. Further investigations revealed that, similar to the cnt operon from which it derives, S1077 synthesis is controlled by the transcription factors Zur and Fur. In contrast, zinS transcription is solely repressed by Zur. RNA affinity purification identified several ZinS-associated mRNAs encoding Zn-dependent enzymes, including the alcohol dehydrogenase Adh, whose synthesis is negatively regulated by ZinS. Loss of ZinS does not alter staphylococcal metal accumulation, suggesting a role in a Zn-sparing response. Remarkably, zinS also encodes a small peptide, ZinP. Genomic analysis shows that the regulatory portion of zinS emerged from the 3′ untranslated region of zinP and is conserved in S. aureus and closely related species, whereas zinP is also found in phylogenetically distant organisms. Our findings demonstrate that sRNAs facilitate bacterial adaptation to Zn limitation, and that genetic exchange and subsequent neofunctionalization have enabled S. aureus to adapt to metal-restricted environments.

The heterogeneity of expression levels among genetically identical cells, termed gene expression noise, is a property of the gene expression process whose importance in the biology of organisms and their evolution is increasingly recognized. Measuring gene expression noise requires single-cell expression data, as obtained from single-cell RNA sequencing (scRNASeq). Its estimation, however, is challenging owing to (i) the presence of technical noise in addition to biological noise, and (ii) the heterogeneity of cell types in the sampled population. We propose a maximum-likelihood framework to infer biological noise from scRNASeq data, while accounting for technical noise, dropout probabilities, and distinct cell sequencing depths. We demonstrate the parameter identifiability using simulations and that the resulting noise estimates are uncorrelated from the mean gene expression, and therefore do not need extra correction in downstream analyses, easing intra- and inter- genome comparisons. Using two technical replicates of scRNASeq data from the wild yeast *Saccharomyces paradoxus*, we show that expression noise can be inferred in a reproducible manner.

Evolutionary rates vary widely among sites in protein sequences, reflecting differences in functional constraints across residues. Individual sites can also experience lineage-specific shifts in substitution rates—a process known as heterotachy—when selective pressures change during evolution. Although such temporal variation has long been recognized, the mechanisms underlying lineage-specific rate shifts and the factors shaping their distribution across protein families remain poorly understood. Here we map site-specific rate shifts across thousands of orthologous protein families spanning the tree of life. Among more than 1.8 million aligned amino-acid sites, over one quarter show evidence of lineage-specific rate changes. Rate shifts are more frequent in families with deep evolutionary origins, including those tracing back to LUCA, whereas younger clade-specific families generally show lower proportions of rate-shifting sites. We next examined whether local structural features predict where rate shifts occur. Residue burial shows only a weak association with rate-shift probability, and its direction differs across domains, with buried residues enriched for rate shifts in Archaea but surface-exposed residues showing slightly higher probabilities in Eukaryota. Moreover, rate-shifting sites rarely form spatial clusters within protein structures, indicating that structural constraints do not globally determine their locations. Despite their widespread occurrence, rate-shifting sites have limited impact on phylogenetic reconstruction beyond random site variation. Together, these results show that lineage-specific rate shifts are a pervasive feature of protein evolution shaped primarily by evolutionary ancestry and phylogenetic depth.

The integrative Sequentially Markovian Coalescent (iSMC) is an extension of the sequentially Markovian Coalescent (SMC) model allowing for parameter heterogeneity along the genome, such as recombination and mutation rates. Heterogeneous parameters follow an autocorrelation process that modulates the genealogical process, extending the hidden state space and adding as few as two extra parameters per heterogeneous rate. Classical hidden Markov chain methodology is used to infer the posterior estimate of the rate landscape. In this chapter, we demonstrate the use of iSMC to infer both recombination and mutation landscapes, using data from Homo sapiens and Homo neanderthalensis genomes. We further indicate how to use simulations to assess statistical power and investigate possible sources of inference noise.

Genomic imprinting is a form of gene regulation leading to the unequal expression of maternal and paternal alleles. The main hypothesis invoked to explain the evolution of imprinted genes is the kinship theory, which posits a conflict between parental genomes over resource allocation in progeny. According to this theory, such conflicts select for parent-of-origin–dependent expression of genes involved in resource allocation. How such conflicts translate into signatures of selection at coding or regulatory sequences remains model-dependent and is not explicitly predicted by the kinship theory. However, most studies addressing selection in imprinted genes in flowering plants, particularly those based on population-genomic or phylogenetic analyses, have focused on self-fertilizing species, where conflicts over resource allocation are predicted to be weak. Consequently, the impact of the kinship theory on the evolution of imprinted genes remains largely unexplored in systems where parental conflict is expected to be strong. Furthermore, potential coevolution between antagonistically acting imprinted genes, as proposed in extensions of parental conflict models, has not yet been tested empirically. Using combined phylogenetic and population genomic approaches, we investigated signatures of selection on imprinted genes across the Brassicaceae family and in autogamous and allogamous populations of Arabidopsis lyrata, and searched for evidence of coevolution among imprinted genes. We found that endosperm-expressed genes exhibited signals of balancing selection across Brassicaceae and within allogamous populations, consistent with models of unresolved intralocus conflict. These population-level signals varied with the mating system, in line with expectations that parental conflict is reduced under self-fertilization. Moreover, phylogenetic analyses indicated signatures of purifying (negative) selection acting on imprinted genes. However, the population-level signatures of selection were independent of the mating system and showed limited concordance with kinship predictions, possibly due to stronger selection acting on expression than on coding sequences. Finally, we identified coevolution between imprinted genes, although not at specific sites, suggesting that interactions beyond protein sequence may contribute to this process.

Asserting which allele is ancestral or derived, known as polarisation, is a prerequisite of many population and quantitative genetic methods. One important application is the inference of the unfolded site‐frequency spectrum (uSFS). The most widely used approaches are based on outgroup data. However, for studies on species with only distantly related outgroups, large divergence between the ingroup and outgroup can result in alignment difficulties and substantial missing data, causing many sites of interest to be lost. Here, we present PolarBEAR (Polarisation By Estimation of the Ancestral Recombination graph), a method that uses the local genealogies from the ancestral recombination graph (ARG) to infer ancestral states. We show that PolarBEAR reaches high accuracy in polarisation and uSFS estimation using simulations under several scenarios. This accuracy, however, heavily depends on the ARG reconstruction method employed. We also applied our method to human population data and compared it with the outgroup‐based method est‐sfs. Although PolarBEAR could not infer the ancestral state with high confidence at certain positions, it obtained results for positions that est‐sfs could not polarise due to missing outgroup data. The polarisation results of the two methods were highly consistent at positions inferred by both methods. The two methods inferred similar uSFS, with PolarBEAR estimating slightly fewer high‐frequency derived alleles. Furthermore, we demonstrate that PolarBEAR is robust across different mutation models in our simulations, while est‐sfs exhibits a bias in the presence of heterogeneous base composition. PolarBEAR can complement outgroup‐based methods, or replace them when no appropriate outgroup sequence is available.

Small regulatory RNAs (sRNAs) are key drivers of bacterial adaptation to environmental fluctuations, including iron and manganese restriction imposed by the host. This study explored the repertoire of sRNAs produced by the human pathogen Staphylococus aureus in response to metal limitation. Two sRNAs, S1077 and ZinS (RsaX20), regulated by zinc (Zn) availability, were identified. Further investigations revealed that, similar to the cnt operon from which it derives, S1077 synthesis is controlled by the transcription factors Zur and Fur. In contrast, zinS transcription is solely repressed by Zur. Amongst the ZinS targets are several Zn-dependent enzymes, such as the alcohol dehydrogenase Adh, whose synthesis is negatively regulated by ZinS. Loss of ZinS does not alter staphylococcal metal accumulation, suggesting a role in a Zn-sparing response. Remarkably, zinS also encodes a small peptide, ZinP. Genomic analysis suggests that the regulatory portion of ZinS emerged from the 3’ untranslated region of zinP in S. aureus and closely related species after horizontal gene transfer from phylogenetically distant organisms. All our findings demonstrate that sRNAs also facilitate bacterial adaptation to Zn limitation, and that genetic exchange and subsequent neofunctionalization have enabled S. aureus to adapt to metal-restricted environments.

Asserting which allele is ancestral or derived, known as polarization, is a prerequisite of many population and quantitative genetic methods. In particular, it allows inference of the unfolded site-frequency spectrum (uSFS). The most widely used approaches are based on outgroup data. However, for studies on species for which closely related outgroups are difficult to obtain, information on many sites of interest may be missed due to alignment problems. Here, we present PolarBEAR (Polarization By Estimation of the Ancestral recombination graph), a method that uses the local genealogies from the ancestral recombination graph (ARG) to infer ancestral states. We show that PolarBEAR can reach high accuracy in polarization and uSFS estimation using simulations under several scenarios. This accuracy, however, heavily depends on the ARG used as input. It is maximal when the true ARGs is used, but can be very low depending on the ARG reconstruction method employed. We also applied our method to human population data and compared it with the outgroup-based method est-sfs. Although our method could not infer the ancestral state with high confidence at certain positions, it obtained results for positions that est-sfs could not polarize due to missing outgroup data. The polarization results of the two methods were highly consistent at positions inferred by both methods. The two methods inferred similar uSFS, with PolarBEAR estimating slightly fewer high-frequency derived alleles. Furthermore, we demonstrate that PolarBEAR is robust to the mutation model used, while est-sfs exhibits a bias in the presence of heterogeneous base composition. PolarBEAR can complement outgroup-based methods, or replace them when no appropriate outgroup sequence is available.

Genomic imprinting is a form of gene regulation leading to the unequal expression of maternal and paternal alleles. The main hypothesis invoked to explain the evolution of imprinted genes is the kinship theory, which posits a conflict between parental genomes over resource allocation in progeny. According to this theory, such conflicts select for parent-of-origin–dependent expression of genes involved in resource allocation. How such conflicts translate into signatures of selection at coding or regulatory sequences remains model-dependent and is not explicitly predicted by the kinship theory. However, most studies addressing selection in imprinted genes in flowering plants, particularly those based on population-genomic or phylogenetic analyses, have focused on self-fertilizing species, where conflicts over resource allocation are predicted to be weak. Consequently, the impact of the kinship theory on the evolution of imprinted genes remains largely unexplored in systems where parental conflict is expected to be strong. Furthermore, potential coevolution between antagonistically acting imprinted genes, as proposed in extensions of parental conflict models, has not yet been tested empirically. Using combined phylogenetic and population genomic approaches, we investigated signatures of selection on imprinted genes across the Brassicaceae family and in autogamous and allogamous populations of Arabidopsis lyrata , and searched for evidence of coevolution among imprinted genes. We found that endosperm-expressed genes exhibited signals of balancing selection across Brassicaceae and within allogamous populations, consistent with models of unresolved intralocus conflict. These population-level signals varied with the mating system, in line with expectations that parental conflict is reduced under self-fertilization. Moreover, phylogenetic analyses indicated signatures of purifying (negative) selection acting on imprinted genes. However, the population-level signatures of selection were independent of the mating system and showed limited concordance with kinship predictions, possibly due to stronger selection acting on expression than on coding sequences. Finally, we identified coevolution between imprinted genes, although not at specific sites, suggesting that interactions beyond protein sequence may contribute to this process.