Co-mutation Based Genetic Networks to Infer Temporal Mutation Dynamics in Ancient Human Mitochondrial Genomes

Abstract

The evolutionary history of Homo sapiens is marked by complex interactions between environmental, cultural, and genetic factors. To investigate the molecular signatures of these processes, we analyzed ancient mitochondrial DNA (mtDNA) across temporal and geographic contexts using principles of co-occurrence of minor alleles defined as co-mutation, through spatiotemporal co-mutation networks of variable sites. Haplogroup-based assessments of variable sites revealed a major transition from foraging to agrarian lifestyles during the Copper–Bronze Age. Genetic network analyses demonstrated that COX and CYB loci exhibited distinct temporal dynamics, with their interactions modulated by NADH dehydrogenase genes in a geological age-dependent manner. To complement the network approach, we constructed phylogeny-based gene interaction networks and assessed polymorphism-to-divergence from chimpanzee ratios. The tree-based networks displayed topologies consistent with co-mutation analyses but showed reduced gene–gene connectivity. Polymorphism/divergence analysis further indicated that the CYB gene has been under long-term purifying selection, whereas ATP6, COX, and NADH dehydrogenase genes experienced episodic purifying selection aligned with distinct historical phases. Collectively, our findings demonstrate that network-based analysis of ancient mtDNA provides insights into early human lifestyle transitions and haplogroup diversification, contributing to the evolutionary foundations of modern human populations.

Introduction

The mitochondrial genome provides a compact yet powerful model for studying evolutionary phenomena in eukaryotes. The human mitochondrial genome is far smaller than the nuclear genome, harboring only 37 genes, 13 of which are involved in oxidative phosphorylation. It is also present in high copy numbers making it a key tool for complimenting paleontological investigations of adaptive variation, metabolic constraints and human evolutionary dynamics (Merheb et al. 2019). The fossil record suggests that archaic humans (Homo erectus) first walked on Africa between 2 and 1.8 million years ago and likely migrated out of Africa for the first time between 300,000 and 280,000 years ago (Leakey et al. 1964; Pakendorf and Stoneking 2005; Carotenuto et al. 2016). Genetic studies provide further, confirmatory evidence of this biogeographical history of humans, especially when DNA collected from the archaeological sites is coupled with radiometric dating of the samples it was extracted from (Llamas et al. 2017). However, it has traditionally been difficult to extract and purify DNA samples from fossils due to (i) chemical degradation of the DNA sequence, (ii) minute quantities, and (iii) organic contamination (Binladen et al. 2006). Recent technical advances, however, have facilitated retrieval of much higher quality DNA sequences from archaeological sites (Orlando et al. 2021), which unsurprisingly, have mostly been mitochondrial DNA. In fact, it is often the only genetic marker that can be recovered from poorly preserved museum specimens and fossil remains. Ancient mtDNA therefore plays a critical role in the reconstruction of human history, including early human migration and admixture events (Nielsen 2017).

The mitochondrial genome, as is the case with the nuclear genome, has acquired genetic adaptations to different environments which can shed light on temporal and biogeographical evolutionary processes (Friedrich et al. 2022). The central role that mitochondrial genes play in metabolism, furthermore, allows us to use ancient mtDNA variation to infer how the availability of nutrients has influenced human migration and subsequent evolution (Kivisild et al. 2006; Luca et al. 2010). The mitochondria are the site of essential metabolic processes necessary to generate energy from proteins, carbohydrates, and fats (Rose et al. 1992; Juretic 2021). The availability of food, and its nutritional content, has been strongly linked with human population diversity (Enattah et al. 2002; Perry et al. 2007; Tishkoff et al. 2007) and mitochondrial gene expression patterns(Gupta and Bhardwaj 2013). The protein-carbohydrate makeup of a human’s diet influences the function of mitochondrial complex I both qualitatively and quantitatively (Juretic 2021). Recent studies have even shown that mtDNA variation can also have diet-dependent effects on longevity (Nagarajan-Radha et al. 2019; Camus et al. 2020).

Human subpopulations can be classified by multiple mitochondrial haplogroups that represent major branching points in the human mitochondrial phylogenetic tree from the Mesolithic (Meso) through the Neolithic (Neo), Copper (Cop), Bronze (Bro), Iron and Middle (Mid) ages (Fig. 1). During this period (9755 BCE to 1264 CE; Table 1) human populations went through cultural and dietary shifts mediated by both climate changes and the accelerating development of agriculture (Milisauskas 2013). The Eurasian samples we study here span from early cultures gathering wild game, fish and seafood in an environment exiting an ice age to agrarian societies in warm, mild climates with diets characterized by grain, livestock and dairy products.

Fig. 1.

The haplogroup composition (%) of mitochondrial samples from Homo sapiens populations sampled throughout Eurasia in each age. Only major haplogroups are shown here

Table 1.

Number of samples and the time-period (BCE = Before Common Era, CE–Common Era) of each age

Age	Time period	Sample size
Mesolithic	9755 BCE to 1850 BCE	82
Neolithic	8250 BCE to 1519 BCE	206
Copper	4725 BCE to 1690 BCE	108
Bronze	4452 BCE to 230 BCE	268
Iron	1390 BCE to 615 CE	76
Middle	10 CE to 1264 CE	72

These haplogroups are defined as sets of mutations in mitochondrial DNA inherited from a common ancestor that typically vary with geographic region (Ruiz-Pesini et al. 2004; Wallace 2015). Studies of ancient mtDNA have also revealed temporal variation in the prevalence of these haplogroups (Mishmar et al. 2003). A critical distinction, that will help us to understand the impact of early lifestyle changes on modern human settlement, is if shifts in haplogroup frequencies between the different ages of recent human evolution reflect selective effects of diet, driven in part by the transition to an agrarian lifestyle, or if they simply reflect genetic drift due to effective population size fluctuations. Currently, haplogroup H is the most frequent globally and is believed to have originated in Southwest Asia nearly 25,000 years ago, descending from the HV clade with two defining synonymous mutations, G2706A (16srRNA) and T7028C (CO1) (Achilli et al. 2004; van Oven and Kayser 2009). On the European continent, haplogroup U is the oldest maternal haplogroup, dating to pre-agricultural populations(Fu et al. 2013) where its dominance at that time coincided with the rise of early farmers in the Neolithic Age (Brotherton et al. 2013). Haplogroup U descended from haplogroup R through the defining mutations A11467G (ND4), A12308G (tRNA-Leu), and G12372A (ND5) nearly 43,000 to 50,000 years ago [24]. The shift from dominance of haplogroup U in Europe to the currently dominating haplogroup H coincides with transitions from hunter-gatherer societies to the early farmers of ancient humans(Haak et al. 2005, 2010; Bramanti et al. 2009) but it is not known if this shift was driven by agriculture-mediated selection on human mitochondria.

In the past two decades, a high number of complete mtDNA assemblies has been organized and made available through the ancient human mitochondrial genome database (AmtDB) which contains complete information regarding the source, age and haplogroup of the samples (Ehler et al. 2019). Here, we use a combination of novel, unconventional co-mutation network analysis, along with conventional phylogenetic reconstruction and within human polymorphism to human-chimp divergence ratio analysis to investigate patterns of mtDNA variation over the last ~10,000 years of human history. The co-mutation networks capture the co-occurrence of variable sites based on the minor allele frequency. The network comprises of variable sites being connected based on the co-occurrence frequencies. The detailed model is explained in the Methods section. We constructed and analyzed co-mutation-based genetic networks for ancient mtDNA covering six geological ages of human evolution. We identified a set of highly connected, in terms of co-mutation, genes for each age and found that NADH dehydrogenase complex genes co-mutate with cytochrome B and cytochrome oxidase complex genes throughout each age, suggesting a functional, modular role for these genes in defining the nutritional habits throughout recent human evolution. The co-mutating mitochondrial genes in each age also clustered as per the temporal sequence of the ages suggesting a role of demographic factors. We constructed gene trees using a maximum likelihood (ML) algorithm for each age and upon further analysis, found that phylogenetic concordance and discordance in these populations correlates with reduced/enhanced genetic interaction networks, respectively. Lastly, we determined the Dn to Ds ratios within humans per protein coding mitochondrial gene for each age (ωpolymorphism) and the divergence Dn to Ds ratios through alignment to the panTro6 chimpanzee reference mitochondria (ωdivergence) and used these to perform McDonald-Kreitman tests that we interpreted with the neutrality index (NI = ωpolymorphism/ωdivergence).

Materials and Methods

Data Acquisition and Variable Site Determination

Complete mtDNA FASTA sequences were downloaded from the Ancient Mitochondrial Database (AmtDB) (Ehler et al. 2019) for which C14 dating was available and through molecular techniques and bioinformatics analyses, DNA damage has been accounted for and corrected prior to uploading to the database. We grouped these sequences according to the available metadata for each geological age (Table 1; Fig. 2). We are specifically keeping the annotation of the ages and associated sequences as is mentioned in the AmtDB.

Fig. 2.

The chronological timeline map as described in Table 1

We aligned the sequences using Clustal Omega (Sievers et al. 2011) with default parameters. Any sequences with insertions/deletions were removed to keep the length of aligned sequences the same as the human mitochondrial reference sequence (rCRS). Any position exhibiting more than one variant, present at least twice within the samples, was considered a variable site. These variable sites were extracted from the aligned sequences for each age separately for constructing co-mutation networks. The ambiguous nucleotides or the missing nucleotides represented as ‘X’, ‘M’, ‘Y’, etc. were replaced with ‘N’ for all the sequences. Since these sequences were extracted from fossil remains, there were various positions where the nucleotide was unknown and denoted as ‘N’. Therefore, we incorporated the presence of an unknown nucleotide for considering a position as a variable site with the threshold of 5%, which was set empirically. For example, for variable site positions where ‘N’ was present < 5%, we still considered it as a variable site but did not include the sequences carrying ‘N’ for further calculations. We also determined which of the variable positions in the data used here are ancestral human SNPs using Mitomap (Lott et al. 2013).

Co-mutation Network Construction

For the construction of a network, nodes were represented by variable sites, and edges were represented by co-mutation frequencies between given node pairs. We defined co-mutation frequency for a pair of variable sites, C_(ij) as,

$$C_{ij}=\frac{{\left(m_{ij}\right)}^2}{\left(m_i\right)\left(m_j\right)}$$ (1)

where m_ij is number of times minor alleles co-occurred (defined as a co-mutation pair) at the i^th and j^th positions, respectively; m_i is the total number of times a minor allele occurs at the i^th position and m_j is the total number of times the minor allele occurs at the j^th position individually. For evaluating the statistical significance of any co-mutation pair, we defined and calculated P_ij as,

$$P_{ij}=\frac{\left[\left(C_{ij}^r\right)\ge \left(C_{ij}\right)\right]}{10000}$$ (2)

where ($C_{ij}^r$) is the number of times two nodes co-mutate randomly. The total number of random realizations was set to 10,000. To generate random simulations, 10,000 permutations were performed for nucleotides at each variable position, and we considered only those realizations where co-mutation occurred more often than the observed one. Moving forward, we took only those co-mutation pairs where P_ij is < 0.05.

Next, we mapped nodes to their corresponding genes for each co-mutation network to achieve one weighted gene-gene interaction (GGI) network for each age (Fig. 5). Since two or more co-mutation pairs may belong to the same gene or a pair of genes, each link was counted as many times as it was found, and this number was considered as a weighting for that pair of genes. For example, the co-mutation pairs (3352–7623) and (4125–8054) map to the ND1-CO2 gene-pair, so this edge was counted twice. Similarly, the co-mutation pairs (3352–3489) map to a self-loop for the ND1 gene. Since the number of variable sites contributed by each gene was proportional to the gene’s length (except for control regions), the weight of each gene pair was normalized by the sum of gene lengths for that gene pair.

Fig. 5.

The degree of the genes of first neighbors of common nodes

To quantify the similarities between ages with respect to the genes participating in the co-mutation networks we calculated the Jaccard distances between the ages to construct a dendrogram based on the variable sites and corresponding genes from the co-mutation networks as:

$$J(A,B)=\frac{|A\cap B|}{|A\cup B|}$$ (3)

where, J(A, B) is the Jaccard similarity coefficient, and A and B are sets of genes for any given pair of ages.

Calculating Network Metrices

Average Degree (< k>)

The average degree is calculated as,

$$<k>=\frac{2E}{N}$$ (4)

where E is the number of edges and N is the number of nodes. The degree of a node in context to co-mutation networks provides the frequency of co-mutation.

Average Clustering Coefficient

This metric tells us about the tendency for clustering by nodes in the network and is calculated as the mean of local clustering coefficients for all the nodes. The local clustering coefficient is calculated as,

$$cc=\frac{e_j}{\left(k_i\left(k_i-1\right)\right)}$$ (5)

where e_i is the number of edges between the neighbors of node i and k_i is the degree of node i. In context to the co-mutation networks, a high clustering coefficient suggests that variable sites tend to co-mutate together for most of the genes except for tRNAs since tRNA genes are only 60–90 bp long and contribute least number of variations (Cann et al. 1987).

Modularity

This measure tells us how well the network is subdivided into small densely connected group of nodes known as modules or communities. We calculated modularity using the Louvain algorithm (Blondel et al. 2008). High modularity suggests that the co-mutation networks are distinctly divided into well-formed communities. Communities usually represent the specific functions or pathways of broader biological function. In the context of co-mutation networks, communities represent the formation of mutational cohorts at the whole mitochondrial genome level.

The Ratio of co-mutating to Variable Sites

We calculated the ratio of variable sites that are taking part in network formation to total variable sites present in a gene across all the ages. It is calculated as,

$$R_g=1-\frac{V_n}{C_n}$$ (6)

where Rg is the ratio for a given gene g, V_n is the total variable sites present in g and C_n is the significant variable sites forming network from g.

Identification of Hub/Global Nodes

High-degree nodes, or hub nodes, are of particular interest as they represent variable sites with extensive connections that enhance network stability and robustness. Such nodes may correspond to evolutionary “hot spots” that shape interaction patterns across the mitochondrial genome. Nodes with abnormally high average degrees were designated hub nodes. To examine the overall dynamics of variable sites across ages, we identified (global) nodes present in all networks. These global nodes served as reference points to track changes in gene interactions over time.

Phylogeny-Based Gene Networks

To identify whether co-mutation interactions (with either high or low weights) were reflecting patterns of common ancestry, we contrasted them with analogous networks that are informative of phylogenetic congruence by constructing thirteen gene trees of protein-coding genes for each age. The nucleotide substitution models incorporated various combinations of substitution frequencies and rates, fraction of invariant sites, and rate heterogeneity under a gamma distribution with five categories (Schwarz 1978). Model fit was assessed under Maximum Likelihood, and the candidates were compared under the Bayesian Information Criterion (Schwarz 1978). The preferred candidates from this comparison (i.e. those reporting the lowest scores) were used as substitution models to infer gene trees for each age using Maximum Likelihood (Penny 2004). We conducted a heuristic search using Neighbor-Joining as the initial tree which was subsequently rearranged under Subtree-Pruning-Regrafting. These analyses were conducted in MEGA11 (Robinson and Foulds 1981). To construct networks of protein-coding genes for each period, we computed weighted Robinson-Foulds distances (Robinson and Foulds 1981) among all the resulting trees using the R (R Core Team 2021) package phangorn v2.11.1 (Schliep 2011). The weighted Robinson-Foulds distance accounts for both tree topology and branch lengths, which allowed us to contrast the co-mutation networks in terms of tree shape as well as accumulated molecular diversification. It is a commonly used distance metric (McDonald and Kreitman 1991) that has shown good performance under similar phylogenies (Stoletzki and Eyre-Walker 2011), which is expected for inferred gene trees that are based on the same individuals and respective ages.

Comparing Phylogeny Based to Co-mutation Based Networks

To quantify the degree of similarity between the phylogeny-based gene networks to the (i) co-mutation enhanced and (ii) co-mutation reduced GGI networks, network similarity indices were calculated as follows:

NetSimile Similarity Node level structural features were used to calculate aggregate statistics for a graph signature vector as.

$$\mathrm{N}\mathrm{e}\mathrm{t}\mathrm{S}\mathrm{i}\mathrm{m}\mathrm{i}\mathrm{l}\mathrm{e}\mathrm{S}\mathrm{i}\mathrm{m}\mathrm{i}\mathrm{l}\mathrm{a}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{y}=\mathrm{c}\mathrm{o}\mathrm{s}\mathrm{i}\mathrm{n}\mathrm{e}(\mathrm{s}\mathrm{i}\mathrm{m}\mathrm{i}\mathrm{l}\mathrm{a}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{y}(\mathrm{V}1,\mathrm{V}2)),$$

where V1 and V2 are the feature signature vectors of the two networks being compared. Values closer to 1 indicate higher similarity.

Spectral Similarity was calculated using the eigenvalue spectra of the graph Laplacian matrices. For each network G, the normalized Laplacian was computed as:

$$L=D-A$$

where A is the weighted adjacency matrix and D is the degree matrix. Values closer to 1.0 indicate higher similarity.

The Wasserstein (Earth Mover’s) Distance between the Laplacian eigenvalue distributions of two networks was computed, and converted to a bounded similarity score as:

$$\mathrm{S}\mathrm{p}\mathrm{e}\mathrm{c}\mathrm{t}\mathrm{r}\mathrm{a}\mathrm{l}\mathrm{s}\mathrm{i}\mathrm{m}\mathrm{i}\mathrm{l}\mathrm{a}\mathrm{r}\mathrm{i}\mathrm{t}\mathrm{y}=\frac{1}{1+W_d}$$

where W_d is the 1D Wasserstein Distance.

Graph Structural Similarity - captures the local topological patterns beyond the global structure. Graphlets (subgraphs of 2 to 5 nodes) quantify motifs. For each network the frequency of graphlets was calculated and the pairwise similarity was estimated using the graphlet degree vector (GDV) correlation. Values close to 1.0 indicate higher similarity.

McDonald Kreitman (MK) Test and Neutrality Index

To identify potential selection on protein-coding genes during the different ages, we applied the McDonald-Kreitman (MK) test, comparing nonsynonymous and synonymous substitutions within archaic humans (polymorphisms) and between archaic humans and chimpanzees (fixed differences or substitutions). Coding sequences were aligned using reference sequences, and substitutions were classified as nonsynonymous or synonymous. For each gene, a 2 × 2 contingency table of polymorphic and fixed sites was constructed, and odds ratios, a.k.a., the Neutrality Index (NI) were calculated. Statistical significance was assessed using Fisher’s exact test. Genes with an excess of fixed nonsynonymous substitutions were considered candidates for positive selection, while those with a relative excess of nonsynonymous polymorphisms were interpreted as undergoing purifying selection. By comparing the relative rates of nonsynonymous and synonymous substitutions within and between species, the test can distinguish neutral evolution (NI ~ 1) from adaptive (NI > 1) or purifying selection (NI < 1) (McDonald and Kreitman 1991; Stoletzki and Eyre-Walker 2011).

Simulated mtDNA Analysis

• iTo generate simulated ancient mitochondrial DNA (mtDNA) tracking distinct chronological periods, we implemented a simulation workflow utilizing gargammel (Renaud et al. 2017). Modern European mtDNA sequences served as the endogenous templates (endo/). To evaluate the effects of temporal degradation, we simulated six distinct chronological groups ranging from the Mesolithic to Middle.

Post-mortem DNA damage (PMD) was modeled using the four-parameter (Briggs et al. 2007) framework, governing nick frequency (ν), geometric distribution of single-stranded overhangs (λ), double-stranded internal deamination (δ_D), and single-stranded overhang deamination (δ_S) using mapdamage tool (Jónsson et al. 2013). Fragment size distributions were simultaneously adjusted using log-normal distribution parameters (--loc and --scale) to capture the progressive fragmentation characteristic of deep-time specimens.

• iiPrior to simulation, a multiple sequence alignment consisting of 70 modern European mtDNA sequences was parsed. Alignment gaps (−) were removed, and sequences were separated into individual FASTA format templates to serve as independent endogenous targets.

For each sequence within a chronological group, a shell loop was executed to isolate the singular target within the endo/directory while maintaining empty contamination (cont/) and microbial (bact/) directories. This architecture enforced a strict 100% endogenous composition via the parameter flag --comp 0,0,1. Execution of gargammel.pl generated paired-end Illumina reads (forward _s1.fq.gz and reverse _s2.fq.gz) at a simulated target depth of 20x coverage (-c 20), incorporating sequencing errors and platform adapters.

• iiiThe raw, simulated paired-end FASTQ datasets were processed through a standard ancient DNA bioinformatic pipeline.
• ivThe final consensus sequence for each template across all six groups was generated using bcftools consensus, applying the identified variants back onto the rCRS reference scaffold to yield the high-confidence simulated ancient mtDNA FASTA sequence.

Results

Variable Site Determination

We obtained 812 ancient mtDNA sequences sampled from six different ages, ranging from the Mesolithic Age to the Middle Age that we were able to align to rCRS along their full lengths. The numbers of variable sites for each age are given in Table 2. The Mesolithic Age exhibited the highest proportion of ancestral SNPs (6.8% of 219 variable sites), whereas the Middle Age showed the lowest (1.0% of 580 sites).

Table 2.

Structural properties of co-mutation networks

Age	Variable sites	N LCC	ELCC	< k>	< cc>	Modularity
Mesolithic	219	185	1087	11	0.74	0.67
Neolithic	530	511	3400	13	0.67	0.71
Copper	491	461	3005	13	0.76	0.82
Bronze	722	712	4003	11	0.26	0.59
Iron	375	335	1723	10	0.80	0.83
Middle	580	560	6338	22	0.84	0.79

Variable sites are number of nodes participating in network construction at Pij < 0.05, N_LCC is the number of nodes in the largest connected component, E_LCC is the number of connections, < k> is the average degree and <cc> depicts the average clustering coefficient of the largest connected component

Co-mutation Networks

We constructed co-mutation networks for each age based on variable sites identified in the mtDNA. All networks were initially disconnected, indicating the absence of co-mutation among certain site pairs. For downstream analyses, we focused on the largest connected component (LCC) in each network. The global structural properties of the co-mutation networks are summarized in Table 2.

Not all significantly co-mutating variable sites were integrated into the networks, and therefore we quantified gene-wise contributions by comparing the number of variable sites in each gene to those forming part of the network (Fig. 3). We found that the proportion of variable sites engaged in co-mutation increased from the Mesolithic to the Middle Age. Interestingly, the participation of variable sites was not determined by gene size: the smallest gene (ATP8) and the largest gene (ND5) were both present across multiple ages, while ND4L, despite being small, showed a marked shift in participation—from nearly 50% non-co-mutating sites in the Mesolithic Age to fewer than 10% in the Middle Age. The networks were sparse, reflecting specific interactions among variants. A negative correlation between node degree and clustering coefficient was observed in all ages except the Copper and Middle Ages (Fig. 4). Furthermore, high modularity across all ages indicates that networks were organized into tightly connected clusters, suggesting that evolutionary or functional constraints structured mitochondrial variation independently of temporal context.

Fig. 3.

The ratio of total variable sites to the variable sites taking part in the network construction

Fig. 4.

The relation between the node (variable sites) degree (k) and clustering coefficient (cc) is plotted here for each era. The plots show an inverse relationship except for Copper and Middle Age

Temporal Dynamics of mtDNA Haplogroups

Over 80% of Mesolithic samples belonged to haplogroup U, with minor representation from haplogroups K, H, and J (Fig. 1). Haplogroup H was present at low frequency (4%) during this period. In the Neolithic Age, haplogroup U remained the most frequent (27%), accompanied by 10 additional haplogroups at varying frequencies, reflecting an increase in mtDNA diversity. Diversification continued into the Middle Age, where 14 major haplogroups were detected. Across temporal groups, haplogroup U dominated during the Mesolithic, Neolithic, and Bronze Ages, whereas haplogroup H became dominant in the Copper, Iron, and Middle Ages. Overall, the observed haplogroup patterns parallel shifts in subsistence strategies, from foraging in the Mesolithic to agriculture and sedentism in later periods.

Hub/Global Nodes and Changes in Gene Interactions Over Time

Eight hub (high degree) nodes were identified across the geological ages (three in the Copper Age and one in each of the other ages). We report which genes they occur in as well as the nature of the point mutation in Table 3. In Mesolithic Age, the G9477A variant in CO3 is a non-synonymous change with a high frequency in the population. This subunit is a part of respiratory complex IV. This polymorphism has been reported as naturally occurring in the population and is not detrimental though it has been associated with oncogenesis (Kozakiewicz et al. 2021) and immune cell activation such as monocytes (Khotina 2022). In the remaining ages, all the hub nodes are synonymous changes in complex III and IV which plays important role in energy metabolism.

Table 3.

Hub nodes and their properties

Age	Position	Gene	Type of mutation	Amino acids
Mesolithic	9477	CO3	Non-synonymous	Valine -> Isoleucine
Neolithic	9428	CO3	Synonymous	Proline
Copper	6116	CO1	Synonymous	Methionine
	6047	CO1	Synonymous	Leucine
	15,511	CYB	Synonymous	Arginine
Bronze	930	12 s rRNA	NA	NA
Iron	12,705	ND5	Synonymous	Isoleucine
Middle	15,454	CYB	Synonymous	Leucine

We further identified 25 global nodes (nodes present in networks across all ages). Here we are interested in finding the pattern of the global nodes and their interactions as global nodes are present in all the ages, but the interesting part is their dynamics with other nodes.

We mapped their first neighbors to corresponding genes and focused on five protein-coding genes—CO1, CO3, CYB, ND2, and ND4—which showed notable variability in their degrees across eras (Fig. 5).

In the Neolithic Age, degrees were widely distributed, reflecting heterogeneous interaction patterns. A marked reduction in degree was observed in the Copper Age and remained relatively stable through subsequent periods, except for CYB. The CYB gene exhibited a distinct pattern: its degree increased sharply after the Copper Age, indicating a potentially unique role in post-Copper Age mitochondrial evolution.

Genetic Interaction Networks and Phylogeny-Based Gene Networks

The phylogeny based, co-mutation-enhanced and co-mutation reduced networks are shown in Fig. 6. Across all periods, more interactions were present in the genetic-reduced networks than genetic-enhanced ones. To complement the co-mutation network analysis, we constructed thirteen phylogenetic trees of protein-coding genes for each age. We found that the co-mutation reduced-genetic networks showed significantly more similarity with the phylogenetic tree-based networks (Fig. 7). The co-mutation enhanced networks were age specific. In the Mesolithic Age, strong associations were observed between ND5–ND4 and CYB–ND4. In the Neolithic Age, ND1 interacted prominently with CO3 and CYB, while ND5 connected with CO1, CYB, ND1, and ND6. During the Copper Age, key interactions included CO1–ND6 and ND2–ND5/CO1. In the Bronze Age, ND4–ND1, ND6–ND5, and CO1 with multiple partners were dominant. Finally, in the Iron Age, ND6–ND3, CYB–ND5, and a dense set of CYB, ND4, ND1, and ND2 interactions were enriched. Notably, ND genes (complex I) consistently featured as central hubs across ages, often in combination with CO genes (complex IV) during the Neolithic and Copper Ages. Interestingly, in the Middle Ages, no strong co-mutation enhanced nor reduced associations were observed for any combinations of genes thus mutations were largely happening independently of each other. We have collected the non-synonymous variable sites, their amino acid substitutions and structural consequences in Supplementary information.

Fig. 6.

The tree-based gene networks and the genetic interaction networks are compared here. The co-mutation-reduced are the genetic interaction networks that show co-mutation less than the corresponding random networks while the co-mutation-enhanced show higher co-mutation than the corresponding random ones. The edge weight describes the occurrence of co-mutation between genes

Fig. 7.

The network similarity scores are calculated between the reduced and enhanced co-mutation networks with tree-based genetic networks. All metrics indicate that the reduced co-mutation networks are more similar to the tree-based networks. Note: the smaller the values of Wasserstein score, the higher the similarity, and vice versa for all other metrics

We also calculated the correlation (Jaccard) distance between the Ages based on the genes from the co-mutation networks and plotted them as a dendrogram (Fig. 8).

Fig. 8.

Jaccard’s distances between the ages were calculated based on presence of genes in the co-mutation networks and plotted as clusters using the correlation metrics

Temporal Dynamical Selection and Neutrality Index

The results of the MK tests and neutrality indices (odds ratio) for the corresponding genes that contain these variants are given in Fig. 9. It is observed that the Middle Ages stand in stark contrast to all preceding Ages in that it is the only time-period where we detect significant positive selection, on ATP8, CYB, ND4L and ND6, and no significant purifying selection on any of the protein coding genes. Prior to the Middle Ages, ATP8, ND4L and ND6 were neutral whereas, CYB was mostly under purifying selection. For the remaining genes prior to the Middle Ages, ATP6, CO1, ND2 and ND5 were mostly under purifying selection, CO3 and ND1 were mostly neutral with some purifying selection and CO2, ND3 and ND4 were always neutral.

Fig. 9.

MK test is performed to calculate the odds ratio between archaic human population and chimpanzee to get the neutrality index

Discussion

We used a combination of co-mutation network analysis, gene tree analysis and McDonald Kreitman tests interpreted with a Neutrality index to interrogate human mitochondrional genome sequences sampled from the Mesolithic to the Middle Ages. The Mesolithic Age being dominated (> 80% of samples) by haplogroup U is consistent with its strong association with hunter–gatherer groups (Bramanti et al. 2009; Walsh et al. 2020) and reflects the reliance of Mesolithic populations on foraging and hunting. In the Neolithic and Copper Ages, haplogroups H and K emerged alongside U, marking the transition to farming and the establishment of agricultural practices in Europe (Haak et al. 2010; Villalba-Mouco et al. 2021). Haplogroup H is strongly linked with early farmers, while haplogroup K is also indicative of sedentary agricultural lifestyles (Haak et al. 2010). In subsequent periods, haplogroup H became dominant, particularly during the Copper, Iron, and Middle Ages, reflecting the consolidation of agrarian societies. Interestingly, the Bronze Age displayed an expansion and substitution of haplogroup U, likely driven by genetic drift, population bottlenecks, or migration events (Villalba-Mouco et al. 2021).Together, these patterns illustrate a clear shift in haplogroup composition from forager-associated lineages to farmer-associated lineages, in line with major cultural and demographic transitions in prehistoric Europe.

The enrichment of ancestral SNPs in Mesolithic populations suggests limited bottlenecks and admixture, consistent with retention of basal haplotypes. In contrast, the reduction of ancestral SNPs in later ages (the lowest number being in the Middle Ages) reflects stronger genetic drift followed by population expansion, likely driven by the transition from foraging to agriculture and subsequent demographic growth (Kuhner and Yamato 2015). Several stable isotope studies of human remain from Eurasia have further shown that humans made a large-scale transition from diets balanced with animal proteins and early grains to primarily processed-grain-based diets as they transitioned from the Iron to the agriculturally intensive Middle Ages.

We further found that the proportion of variable sites engaged in co-mutation increased (Fig. 2) from the Mesolithic to the Middle Age, indicating greater genetic homogenization with the rise of agrarian societies (Lott et al. 2013; Brandt et al. 2013; Bird et al. 2019). Such lifestyle transitions likely facilitated enhanced gene flow, reduced heterogeneity, and more coherent mutational patterns. Interestingly, the participation of variable sites was not determined by gene size: the smallest gene (ATP8) and the largest gene (ND5) were both present across multiple ages, while ND4L, despite being small, showed a marked shift in participation—from nearly 50% non-co-mutating sites in the Mesolithic Age to fewer than 10% in the Middle Age. The co-mutation networks were sparse, reflecting specific interactions among variants. A negative correlation between node degree and clustering coefficient was observed in all ages except the Copper and Middle Ages (Fig. 4). This correlation was strongest in the Bronze Age, consistent with hierarchical organization, possibly shaped by technological innovations, agricultural intensification, and demographic growth through inter-population contact (Gowdy and Krall 2014). Furthermore, high modularity across all ages indicates that networks were organized into tightly connected clusters, suggesting that evolutionary or functional constraints structured mitochondrial variation independently of temporal context.

To complement the co-mutation network analysis, we constructed thirteen phylogenetic trees of protein-coding genes for each age. The networks with reduced co-mutation followed relatively similar topology to the phylogenetic tree-based gene networks (Fig. 6). We compared the degrees of the genes of tree-based networks and that of the genetic networks. The tree-based networks showed a similar trend to that of the genetic-reduced networks. This suggests that the genetic networks with reduced co-mutation are informative of the shared ancestry while the genetic networks with a high co-mutation might be capturing phylogenetically independent functional aspects of the mitochondrial genome in terms of temporal and lifestyle-based evolution (Wilson and Vinogradov 2014; Översti and Palo 2022). We also calculated the correlation distance between the periods based on the genes from the genetic interaction networks (Fig. 8). The dendrogram of genes clearly denotes separate clusters for ages with a branching pattern that reflects temporal associations. The first branching clusters the ages into two groups, one with the Mesolithic and Neolithic Ages and the other with the remaining four ages. Iron and Middle Ages along with the Copper and Bronze Ages formed two separate clusters. The Middle and Iron Ages were farthest from Mesolithic and Neolithic Age which is in line with their temporal state. In the dendrogram of genes, the Copper and Bronze Ages were closest to each other. The dendrogram was constructed to check if the genes that are taking part in the co-mutation networks are also following a temporal pattern, which they are.

Genes that were under purifying selection for three or more ages include ATP6, CO1, ND2 and CYB (Fig. 9). ATP6 was also a highly connected node (Fig. 6) in the tree-based Neolithic and Copper Age networks and was well represented across all periods for the co-mutation reduced networks (although the connections seem to decrease as the Middle ages were approached). ATP is a highly variable gene among humans and at the same time, it is the most conserved one across the latitude. CYB and CO1 comes next in the variability based on the Ka/Ks ratio while moving from temperate to tropical regions (Ingman et al. 2000). CO1 was a highly connected tree-based node in the Iron Age, a highly connected co-mutation enhanced node in the Copper age and a highly connected co-mutation reduced node in the Bronze Age. ND2 was a highly connected tree-based node in the Copper, Bronze and Iron Ages and only showed up in the co-mutation networks as a reduced co-mutation node in the Iron Age. The CYB gene is interesting in that it is one of four genes where positive selection was detected, but only in the Middle Age. In the Mesolithic, the CYB gene was a highly connected tree-based and co-mutation reduced node. This pattern continued until the beginning of the Iron Age, where it switched to being a highly connected co-mutation enhanced node which continued up to the Middle Age. Thus, the switch to positive selection seems to be associated with a switch to enhanced co-mutation. Collectively, these genes are components of complex I, III, IV and V, which are all the critical part of the respiratory chain which shows that with time all the complexes have gone through certain degree of purifying selection to maintain the optimal functionality of the mitochondria. The purifying selection on these genes suggests that the functional integrity of these genes was strongly constrained. However, we are not sure about the types of constrains but they are particularly relevant to changing environment and external conditions such as dietary requirements, physical activity, and population-level stresses.

Other genes were mostly under one or two ages of purifying selection or neutrally evolving, these being ATP8, CO2, CO3, ND1, ND3, ND4, ND4L, ND5, and ND6, except that in the Middle Age ATP8, ND4L and ND6 were detected as being under positive selection. ATP 8 is also a highly connected tree-based network node in the Mesolithic, Copper and Bronze Ages (and still connected, but less so for the other ages), but does not appear prominently in the co-mutation enhanced or reduced networks for any age. This seems to be a gene that drifted to high frequency (as evidenced by its tree-based network connections) and then took on some role that underwent positive selection in the Middle Age. ND4L only appears as a highly connected node in the tree-based network as well as the co-mutation enhanced network of the Middle Age, thus as with the CYB gene, we can see an association with positive selection and enhanced co-mutation. Similarly, ND6 only appears prominently in the genetic enhanced networks of the Copper through to the Iron Ages, suggesting that the co-mutation enhanced genes have more chance of being under positive selection. ND6 genes codes for the subunit of complex I, climate change, high-altitude and flight response of the parasympathetic nervous system can induce for the positive selection in ND6 gene.

We next discuss protein coding genes in relation to contribution of these variable sites in each temporal network along with their evolutionary significance and neutrality index for the corresponding genes (Fig. 9). In Mesolithic Age, the position 9477 of the CO3 gene is a hub node which is a defining variant for the U5 haplogroup and is a non-synonymous variant causing a change from valine to isoleucine at position 91 in CO3 protein subunit. This variant is responsible for high sperm mobility and quality in human males (Silva et al. 2022). Despite the Neutrality index for the CO3 gene of 3 (Fig. 9) suggesting functional constraints on this gene in the Mesolithic Age, the MK test did not detect a significant departure from neutrality, which is consistent with valine and isoleucine being interchangeable as non-polar amino acids in the transmembrane portion of the protein.

Neolithic Age:

The transition from the Mesolithic to Neolithic Age was marked by a Eurasian climate shifting from a cold post-ice age period to warm, stable weather that facilitated early farming (Milisauskas 2013). In our analyses we found significant changes in this age associated with the CO3 gene, coding for a subunit of cytochrome c oxidase (complex IV). Complex IV is the point where oxygen is reduced and CO3 variants influence cellular respiration and energy metabolism. The CO3 gene is primarily governed by purifying selection and elimination of deleterious mutations (Malyarchuk and Derenko 2023). We found position 9428 of the CO3 gene is a hub node which is a synonymous variant implying its influence may lie in mRNA stability, codon bias, or nascent protein folding (Milisauskas 2013). The Neutrality index for CO3 of 2 also implies possible purifying selection on CO3 in the Neolithic, however, it was not a significant departure from neutrality.

Copper Age:

The Copper Age was marked by the continued development of agriculture as Eurasian societies consumed increased dairy products and began to produce metal tools (Milisauskas 2013). Our analyses identified significant changes in CO1, important for oxygen metabolism and CYB, influencing ATP utilization. We found diversification of hub nodes with variants 6116 and 6047 in the CO1 gene and 15,511 in the CYB gene as critical hubs. Among the three hub nodes, the 6047 in the CO1 gene is a marker of the U4 haplogroup. The presence of multiple hub nodes may indicate complex selective pressures during this period, possibly reflecting a period of migration and genetic admixture (Montiel-Sosa et al. 2006). The association of haplogroup markers with hub nodes underscores the dual role of these sites in both defining genetic lineages and contributing to mitochondrial function. Interestingly, both CO1 and CYB genes are significant for the MK test and the respective Neutrality indices of 10 and 9, indicating strong purifying selection on both genes in this period. Looking back in time, the CO1 and CYB genes were also under strong purifying selection in the Neolithic Age. Looking forward, the CO1 gene is under constant purifying selection in the Bronze and Iron Ages whereas the CYB gene appears to be neutral in the Bronze Age (but with a Neutrality index ~2), but again under purifying selection in the Copper Age. Both genes for the most part appear to have been under purifying selection across the periods examined here (Malyarchuk and Derenko 2023).

Bronze and Iron Ages:

During the Bronze and Iron Ages the Eurasian climate cooled slightly and diets continued to be dominated by farmed grains and livestock products including meat and dairy (Milisauskas 2013). We identified significant changes in 12s-rRNA, part of the ribosomal machinery influencing overall translational regulation of mitochondrial genes. In the Bronze Age position 930 in the 12 s rRNA gene is a hub node. 12s-rRNA is part of the ribosomal machinery suggesting the influence on the translational regulation of mitochondrial genes. In the Iron Age, variant 12,705 of the ND5 gene is a hub node and a synonymous variant. The MK test suggests it is evolving neutrally in the Iron Age but under purifying selection in the Bronze Age (as well as in the Mesolithic and Neolithic Ages). The ND5 protein is a vital subunit of complex I which has undergone positive as well as negative selection (Malyarchuk and Derenko 2023).

Middle Age:

We identified significant changes in position 15,454 of CYB gene is a hub node which is a synonymous variant and a marker of the U3 haplogroup.

Our analyses found that significant changes at hub nodes were limited to synonymous variants suggesting selection for constancy in metabolic homeostasis. CYB, as mentioned above, consistently turns out significant for the MK test and the Neutrality index > 1 suggests long-term, strong purifying selection throughout the ages.

Overall, the genes containing hub nodes were enriched for significant and strong purifying selection where the non-Hub containing genes are enriched for neutrality. The ATP6 gene is an exception in that it does not contain any hub node variants, but the MK tests for this gene were significant in the Mesolithic, Copper, Bronze and Iron Ages and the minimum Neutrality Index was 4 indicating consistent, strong purifying selection (Fig. 9). The ATP6 gene is involved in ATP production and is the last complex of the electron transport chain. Previous studies have shown that ATP6 has undergone purifying selection and constrained the amino acid changes to maintain the functional integrity of this gene (Ruiz-Pesini et al. 2004; Kivisild et al. 2006). A limitation of the present study is that the analyses were performed on publicly available mitochondrial genome consensus sequences rather than uniformly reprocessed raw sequencing reads. Therefore, potential contributions from residual NUMT contamination could not be directly assessed through masking or read-level filtering.

Conclusion

In summary, our results demonstrate that co-mutation networks effectively capture the temporal dynamics of mitochondrial genome variation across human evolutionary periods. Statistical interrogation of these networks revealed age-specific genetic interactions, with CO1 and CO3 consistently emerging as high-degree nodes, underscoring a central role of the cytochrome oxidase complex in shaping diet- and lifestyle-associated evolutionary trajectories. Notably, the progressive increase in co-mutating variable sites from the Mesolithic to the Middle Ages suggests a shift toward greater genetic homogeneity, plausibly reflecting societal transitions toward agrarian lifestyles. By complementing network-based analyses with phylogenetic approaches, we observed that reduced co-mutation networks are consistent with lineage-specific evolutionary constraints, whereas enhanced networks may capture functionally relevant couplings that have been observed in our previous studies (Shinde et al. 2021; Verma et al. 2021, 2022). The additional simulation-based control analysis indicates that these network patterns are not readily reproduced by ancient DNA-like damage alone, although the present study does not by itself establish adaptive significance. Together, these findings establish co-mutation networks as a powerful framework that, when integrated with phylogenetic approaches, provides nuanced insights into the interplay of functional, epistatic and evolutionary forces shaping mitochondrial genome dynamics.

Supplementary Material

The online version contains supplementary material available at https://doi.org/10.1007/s00239-026-10347-y.

Data Availability

No datasets were generated or analysed during the current study.