DISCUSSION
The urban sprawl is a worldwide phenomenon deeply affecting the environment and thus requiring fast adaptive responses in city dwellers. While a large body of literature already describes a myriad of phenotypic shifts in urban populations of numerous species (Chamberlain et al., 2009; Lowry, Lill, & Wong, 2013; Sepp, Mcgraw, Kaasik, & Giraudeau, 2017), the molecular bases of these shifts and their evolutionary implications remain yet to be documented and understood. This study uses genomic and epigenomic analyses to decipher the potential molecular bases implicated in phenotypic shifts and adaptation in several urban populations of a passerine bird, the Great tit. Note that this species shows largely parallel phenotypic shifts across its range in terms of morphology and life history (Biard et al., 2017, Thompson et al., sub ). Genomic analyses revealed weak yet significant average differentiation between urban and forest populations, suggesting ongoing gene flow and limited drift in urban populations. These analyses also identified a limited number of loci putatively under strong selection, non-repeated between pairs, and numerous loci supposedly under weaker selection, compatible with a polygenic model of evolution. On the epigenomic side, we found weak average differentiation of the methylome between urban and forest birds, suggesting an absence of genome-wide epigenetic deregulations, which is notably in line with an absence of strong genetic differentiation. In turn, we identified several strongly differentially methylated regions between urban and forest birds, mostly non-repeated between pairs and hence potentially implicated with local evolution of urban populations. Genes associated with either genomic footprints of divergent selection or differentially methylated regions had relatively similar functions, related to the nervous system, metabolism, immunity and behaviour, that have been repeatedly convicted in other studies (Riyahi, Björklund, Mateos-Gonzalez, & Senar, 2017). Hence, by identifying non-repeated genetic and epigenetic responses among replicated forest-urban population pairs, our findings support the hypothesis of mostly non-parallel rapid de novo adaptation to similar environments via both genetic and epigenetic mechanisms. Our results are in line with accumulating evidence that polygenic adaptation and epigenetic reprogramming may be involved in quick phenotypic shifts in response to rapidly emerging constraints such as urbanization.
Overall genetic differentiation between populations was relatively low (FST ranging from 0.009 to 0.034), although higher than what has been found at a much larger scale across the species distribution (e.g. FST around 0.01 between UK and Spanish or between French and Spanish populations (Laine et al., 2016)). Low but significant differentiation levels are in line with previously documented genetic divergences between city and forest great tit populations (Perrier et al., 2018; Senar & Björklund, 2020), and altogether suggests important gene flow, large effective population sizes and limited genetic drift at multiple spatial scales (Kvist et al., 2003). This overall genetic context is particularly suitable to search for genomic footprints of divergent selection between urban and forest populations, which would easily be identifiable above the neutral level of genetic differentiation.
We found a limited number of strong footprints of divergent selection, which is in line with previous results in Montpellier (Perrier et al., 2018), and across nine European cities (Salmón et al., 2020). Similarly to low levels of parallelism in allele frequency changes between cities observed by previous studies (Reid et al., 2016; Salmón et al., 2020), and despite similarities in phenotypic shifts, none of these outliers were shared between cities. This result suggests limited parallel evolution, supporting a scenario of independent de novo evolution between cities and/or different selection pressures between cities. Indeed, there may be multiple evolutionary solutions to the same environmental challenges (Losos, 2011) and multiple traits are linked to the same functional outcome (Thompson et al., 2017). Besides, the identification of numerous outliers by the multivariate framework applied at the scale of all six sampling sites supports a model of polygenic urban adaptation implicating multiple genes, biological pathways and phenotypic traits (Boyle, Li, & Pritchard, 2017). Polygenic adaptation is a reasonable expectation in urban habitats since the multiple new environmental conditions in cities most probably result in many novel selective pressures acting on a multitude of functional traits (Shochat, Warren, Faeth, McIntyre, & Hope, 2006), and because many of these traits may be quantitative, and  genetically correlated (Lande, 1979). Further polygenic analyses on more samples and more markers (i.e. whole genome data) are however required in order to estimate the potential effect of each genetic variant implicated in the adaptation to life in the city (Robinson, Wray, & Visscher, 2014; Zhou & Stephens, 2012).
Several genomic footprints of divergent selection between urban and forest environments were in, or in the vicinity of, genes that have already been described as playing a role in neuronal development, behaviour or cognitive abilities. In particular, the NR4A2 gene plays an important role in recognition of novel objects and memory in mice (McNulty et al., 2012). Reaction to novel objects and novel food is known as one of the main factors determining the capacity of a species to thrive in an urban environment (Lowry et al., 2013). The DCX gene is related to neuronal plasticity (Kim, Peregrine, & Arnold, 2006) and experimental approaches revealed that artificial light at night induces an overexpression of this gene linked to a change in behaviour and expression of depressive-like behaviour in crows (Taufique, Prabhat, & Kumar, 2018). Finally, the CHRNA1 gene is associated with aggressive behaviour in chicken (Buitenhuis, Hedegaard, Janss, & Sørensen, 2009), and higher aggressiveness is commonly observed and hypothesized as adaptive in urban bird populations (Sprau & Dingemanse, 2017). Besides, the gene ontology enrichment analysis, performed on the entire set of genes identified via the outlier genome scan, reinforced these findings since multiple enriched GO terms were associated with the nervous system and stress response as well as hormonal response (Table 2). These results are informative on the type of traits involved in avian urban adaptation in cities and corroborate previous results from (Salmón et al., 2020; Sih & Del Giudice, 2012; Sol, Lapiedra, & González-Lagos, 2013) suggesting that natural selection repeatedly acted on neuronal, behavioural and cognitive traits that could contribute to the phenotypic shifts described in urban great tits (i.e. more aggressive and exploratory birds, with higher breath rate (Senar et al., 2017; Torné-Noguera, Pagani-Núñez, & Senar, 2014 & Caizergues et al. in prep ).
Contrary to the common prediction that living in cities is likely to influence epigenomes (McNew et al., 2017; Watson et al., 2021), no genome wide pattern of differentiation in methylation between urban and forest great tits was detected. However, we observed a difference in mean methylation level between birds from Warsaw and Barcelona on their autosomes, as well as between males and females on the Z chromosome, showing that methylation differences were identifiable. In addition, we found no mean difference in methylation level between habitats, revealing that urbanization did not strongly affect overall methylation levels in a specific direction. This overall low differentiated methylation context is perfectly suitable to investigate more localized zones that could differ in their levels of methylation. Note that the strong methylation contrast between males and females on sex chromosomes (Figure 2), is in line with previous reports in vertebrates (Teranishi et al., 2001; Waters, Capraro, McIntyre, Graves, & Waters, 2018) showing that methylation plays a major role in sex differentiation via regulation of gene expression and genetic imprinting.
Despite the non-significant effect of habitat on overall methylation levels, we found a large number of DMRs between pairs of forest and urban populations, suggesting that urbanization did affect particular regions of the genome. DMR were less likely to occur within a gene body than by chance, but it was not the case for promoter or TSS regions. This latter result contrasts with Watson and collaborators (Watson et al., 2021) who recently found an under-representation of DMR in both gene body and regulatory regions in urban great tits from Malmö (Sweden). Across the three cities, 35.3% of DMR were directly localized in gene bodies, and 47.3% in TSS or promoter regions, suggesting a potential role in gene expression modulation. Direction of methylation patterns did not follow any consistent pattern (no over-representation of hypo- or hypermethylated DMR in urban birds, Figure 5), in line with Watson and collaborators’ analyses on blood sample. However, birds in Barcelona presented significantly more hypomethylated DMR than hypermethylated ones, a result that warrants further investigation.
Only a limited number of urbanization-linked DMR were shared between two or more locations. In contrast, three times more sex linked DMRs were found in two locations or more. This comparison suggests that urbanization-linked epigenetic modifications most probably do not occur in a parallel way across cities, but rather that each city might have its particular epigenetic response. Indeed, in the emerging field of urban evolutionary biology, cities are often regarded as valuable replicates of human-altered habitats (Donihue & Lambert, 2015; Santangelo et al., 2020), and it is often expected that parallel adaptive responses to similar selective pressures will occur. This expectation is particularly strong when phenotypes show parallel changes, as is the case for the Great tit, which is consistently smaller and lays earlier and smaller broods in the various cities where it has been studied, compared to forest habitats (Caizergues, Grégoire, & Charmantier, 2018; Seress, Sándor, Evans, & Liker, 2020). However, as discussed above, parallel adaptation to similar environmental conditions should not be expected in the case of independent evolution, especially for multilocus traits. Additionally, cities are different from each other because of a wide array of climatic, cultural, historical and socioeconomic factors (Szulkin et al., 2020). In fact, besides the obvious differences in cities’ climatic conditions depending on their position on the globe, land use, fragmentation and pollutants levels can also largely vary across cities (Cárdenas Rodríguez, Dupont-Courtade, & Oueslati, 2016). In a general way, pollutants are known to affect DNA methylation and result in both hypo and hypermethylation, but the patterns of change observed largely rely on the pollutant involved (Head, 2014). Hence differences in cohorts of pollutants present in cities could be responsible for differences in patterns of methylations. Taken together, the results of the present study highlight the importance of questioning the assumption that cities are replicated environments that can be considered similar, and parallel evolution across cities may be the exception rather than the norm.
As mentioned earlier, increasing evidence suggests that DNA methylation can be associated with environmental and stress factors (env: (Foust et al., 2016), stress: (Sun et al., 2018)) especially during early development (Meaney & Szyf, 2005). Here, we found four genes (POMC, ADAMTS3, PAPD4 et GCC1), associated with DMR that were previously described in great tits as undergoing major changes in methylation levels in case of exposure to higher levels of pollutants (Mäkinen, van Oers, Eeva, Laine, & Ruuskanen, 2019). Notably, the functions of these genes remain to be determined, and they could thus be interesting to target in future studies. In addition, the past literature has repeatedly found SERT and DRD4 as two major  genes involved in urban-specific avian human avoidance (or wariness) behaviours (see for example in the Great tit (Riyahi et al., 2015), in the blackbird (Garroway & Sheldon, 2013), in the black swan (Van Dongen, Robinson, Weston, Mulder, & Guay, 2015) & in the burrowing owl (Mueller et al., 2020) ; see SI Figure S7 to S11 to see patterns of methylation associated with 6 classical great tit linked candidate genes). In this study, while urban great tits show higher levels of aggressiveness in at least two of the cities (Riyahi et al., 2017 & Caizergues et al. in prep ) we found no DMR associated with these two genes in either of the three forest-city pairs. However, we found a significant urban-related change in methylation linked to the DRD3 gene, belonging to the same gene family as DRD4 and known to be similarly involved in chicken aggressive behaviour (Z. Li et al., 2016). In line with these results, GO analyses revealed enrichment in genes associated with neuronal functions, behaviour, but also blood, immune and endocrine systems (Table 2, Figure S5), revealing the potential need of physiological adjustments in urban habitats. Surprisingly, a recent study on great tit differences of methylation between city and forest habitats in another European city found no GO enrichment in blood, while some in liver tissue (Watson et al., 2021) (note that they investigated DMSs, Differentially Methylated Sites, which differs from DMRs identified here). These contrasted results highlight the fact that methylation patterns highly depend on the analysed tissues (Derks et al., 2016), and show, once more, that urban linked methylation might not be similar from one city to another. In addition, it has been demonstrated that DNA methylation can undergo seasonal variation (Viitaniemi et al., 2019). Hence, analyses on multiple tissues and life-stages replicated in multiple pairs of urban and forest populations are required, to draw a broader view of the impact of urbanization on global methylation patterns and to understand replicated parallel occurrence across cities. However, tissue-specific and age-specific analyses in multiple individuals across several pairs of urban and forest environments poses major technical, budget and ethical limitations and should be coordinated very carefully. Additionally, specific drivers of shifts in methylation remain to be disentangled to understand which environmental factors are responsible for which change in methylation. To do so, experimental settings manipulating environmental factors such as performed by Mäkinen and collaborators (Mäkinen et al., 2019) would be particularly useful. More integratively, information on how shifts in methylation patterns affect phenotype, fitness, and adaptation, often remain elusive (Sepers et al., 2019). To our knowledge, a limited number of studies attempted to link methylation and expression levels in natural population contexts (Derks et al., 2016; Laine et al., 2016), and even fewer in urban habitats (but see e.g. McNew et al., 2017; Watson et al., 2020). Hence, future work might need to tackle the question of the origin and adaptive significance of these variations in a controlled framework.
This study identified both genomic footprints of selection and differentially methylated regions between urban and forest populations, suggesting that both genetic and epigenetic processes could play an important role in rapid adaptation to urban habitat. To our knowledge, our study is the first to use replicated pairs of cities and forest populations to study modifications of methylation in urban habitats. This study design revealed limited evidence for parallelism between cities both at the genetic and epigenetic levels, suggesting that cities might not present exactly similar environmental conditions or that different genetic and epigenetic pathways are involved in adaptation to urban environmental conditions, although possibly associated with similar biological functions. This study finally highlights the need to unravel both environmental origins and evolutionary implications of methylation shifts, in order to understand to which extent environmental induced methylation can contribute to adaptation.