Introduction

According to Ernst Mayr, there are three types of questions in biology: "What?", "How?" and "Why?" (Mayr 1997). The "What?" questions are aimed at the description of facts. The answers to these questions depend on the methods and tools that are available in a given time (Kováč 2007a, b). The sophistication of these methods has grown dramatically during the last few decades, enabling us to observe fine details of biological objects by direct or indirect means. The "How?" questions deal with understanding mechanisms of action of a particular biological system (molecule, cell, tissue, organ, organism, ecosystem). The answers to these questions are essential for understanding underlying mechanisms, and, in turn, for therapeutic interventions with their pathological forms in case of diseases. Finally, the "Why?" questions are aimed at historical and evolutionary actors that account for all aspects of living organisms. Addressing all three types of Mayr's questions is essential for obtaining a complex view of a biological phenomenon.

The legitimate answers to the "Why?" questions started to emerge after the publication of Darwin's The Origin of Species (Darwin 1859). The concept of descendance with modification(s) based on heritable variability and driven by natural selection seemed to provide a straightforward answer. To put it in an oversimplified form, disregarding chance events such as genetic drift or migration, a population of reproducing biological entities (cells, organisms) exhibit variation, part of which is heritable. Those entities producing more offspring have higher reproductive fitness, and their relative proportion increases from generation to generation until a new, better adapted, and more efficiently reproducing variant takes over. This simple scenario becomes more complicated when we consider the "entity" in a broader sense. Take for example the leafcutter ants (Fig. 1a) (Hölldobler and Wilson 2011). Their colony is composed of numerous castes specialized in particular tasks. Explorers, workers, and gardeners work collectively to maintain a garden dedicated to the cultivation of a special fungus, Lepiotacea, whose spores are used as the main food source for the entire colony. The complexity of this system is increased by other members. There are parasitic fungi, such as Escovopsis, feeding on the Lepiotacea spores that could destroy the whole ant colony; yet, the ants' cuticle contains actinomycetous bacteria (e.g. Pseudonocardia) producing antibiotics killing the parasite (Futuyma and Kirkpatrick 2017). In addition, the garden contains a community of yeasts, some producing antifungal toxins fatal to Escovopsis, as well as Beauveria, a fungal pathogen of ants (Rodrigues et al. 2009). What a fascinatingly complex ecosystem of interdependent participants! A change encountered by any member that will affect its fitness might have an impact on the fitness of other members of the consortium. This illustrates the central question of our review: how do these ecosystems co-evolve?

Fig. 1
figure 1

Examples of biological systems with interdependent members. In both examples the individual members of the consortium must co-evolve to keep the whole system functional. a Colony of leafcutter ants. In addition to leafcutters and other castes the ant colony includes other members exhibiting both positive (green) and negative (red) interactions. Photograph of Atta cephalotes, Wilhelma Zoo, Stuttgart, Germany; https://commons.wikimedia.org/wiki/File:Atta_cephalotes-pjt.jpg). b Typical heterotrophic eukaryotic cell. The cell contains two interdependent genetic systems, the nucleus and the mitochondria, expressing proteins involved in their communication. Mt, mitochondrial, CI, CIII, CIV and CV, complexes I, III, IV, and V of the respiratory chain and ATP synthase. Black rectangles represent subunits that are products of mitochondrial genes, white rectangles are subunits encoded by nuclear genes and imported from the cytosol

Full size image

The question of co-evolution is by no means new and it has broad implications including speciation (Lawlor and Maynard Smith 1976; Thompson 1989, 1994, 2005, 2014, 2017). There are a number of studies aimed at understanding the evolutionary dynamics of various model systems including predator–prey, plant–herbivores, host–pathogen, plant–pollinator (Van Valen 1973; Futuyma and Slatkin 1983; Anderson and May 1982; Ridley 2004; Yuan et al. 2013; Futuyma and Kirkpatrick 2017; Berezovskaya et al. 2018). For the latter, probably the most famous example is the Madagascar star orchid Angraecum sesquipedale, with a nectary up to 30 cm long, and, at the time of its discovery, an inscrutable pollination mechanism. Both Darwin and Wallace independently predicted the pollinator would be a moth that would have a foot-long tongue. Indeed, in 1903, 20 years after Darwin’s death, a Madagascan hawk moth Xanthopan morganii praedicta with a 25–30-cm-long proboscis was discovered (Nilsson et al. 1985; Wasserthal 1997; Whittall and Hodges 2007). Another textbook example of a co-evolving system is mimicry, mostly studied in butterflies from the genus Papillio, in which one (edible) species mimics its toxic relative to avoid predators. This is a system of three players (toxic butterfly, its edible mimic and a predator) that undergo co-evolutionary race (Mallet and Joron 1999).

Many plants and animals (including humans) are filled with both prokaryotic and/or eukaryotic symbionts required not only for their hosts’ survival, but which are also interdependent on each other (Wernegreen 2012; Archibald 2015). A classic example of the essential links between multiple symbiotic partners are lichens, some containing one algal and three fungal, cohabitating species (Tuovinen et al. 2019). Champions among numerous examples of symbiosis in insects are termites hosting in their intestines at least ten endosymbionts, each required for both the insect and for most of their other endosymbiotic partners (Peterson and Scharf 2016). Closely related lineages of the parasitoid wasp Nasonia cannot form viable hybrids due to the incompatibility of their gut microbiota (Brucker and Bordenstein 2013). Tilapias, tropical fish able to adapt to a wide range of temperatures, do so in concert with their gut microbes (Kokou et al. 2018). Corals form consortia with dinoflagellates, viruses, fungi, archaea, and bacteria involved in various aspects of their life style (van Oppen and Blackall 2019). Mammals (Integrative HMP (iHMP) Research Network Consortium 2019) and plants (Uroz et al. 2019) are also associated with a great diversity of microorganisms, such as the ‘microbial terroir’ of the common grape vine affecting the quality of produced wine (Vitulo et al. 2019). All these consortia are built on a jungle of associations. According to one theory, the hosts and their resident microbes function as an evolutionary unit, dubbed a holobiont, carrying a “hologenome”, the genetic information encoded by all partners (Richardson 2017; Morris 2018; Rosenberg and Zilber-Rosenberg 2018). Although the concept of holobionts has its opponents (Doolittle and Inkpen 2016; Skillings 2016), it also highlights challenges associated with understanding the co-evolution of the partners involved in the intricate partnerships such as those described above.

Of course, there are instances of systems composed of components not exhibiting signs of co-evolution. For example, it was argued that that lichen photobionts, compared to mycobionts, have very limited capacity to evolve adaptations to lichenization, so that the symbionts in lichens do not co-evolve (DePriest 2004). It was shown that compared to mycobionts, the contribution of beneficial mutations acquired by photobionts seems to be insignificant for lichenization. This may be due to the fact that there is no sequential selection of photobiont cells from one lichen into another needed for natural selection and there is no photobiont sexual reproduction in the thallus (Hill 2009). The additional example is based on the studies of mating preferences of females in two sister species of swordtail fish, Xiphophorus multilineatus and X. nigrensis. Whereas X. multilineatus females were similarly attracted to conspecific males with vertical bars as to X. nigrensis males without bars, females of X. nigrensis were more attracted to X. multilineatus males with vertical bars than to the males of their own species (Morris and Ryan 1996). According to these results, it was inferred that in swordfish the male traits and female preferences do not co-evolve, but instead traits evolved to exploit pre-existing preferences (Ryan 1998, 2005). Similar results were obtained in studies of preference of chick calls by females of two species of Physalaemus (P. pustulosus and P. coloradorum) demonstrating the preference for complex calls exists in species lacking such calls (Ryan and Rand 1993; Ryan 2005). On a molecular level, an apparent absence of co-evolution is exemplified by an extremely conserved amino acid sequence of histones compared with a highly variable nucleotide sequence composition (e.g., G+C content) of nuclear DNA (Bakkaiová et al. 2016). The binding of histones to DNA exhibits a high degree of robustness enabling to evolutionary diversification of nuclear DNA. Similarly, the genes for antifungal peptides (drosomycins) in Drosophila do not seem to co-evolve with the pathogens possibly due to a lack of specific fungal and bacterial parasites in Drosophila populations (Jiggins and Kim 2005).

The examples provided in this paper (Tables 1, 2, 3), although far from being complete, illustrate that there is a wide repertoire of possible evolutionary relationships (from no to strong co-evolution) between the partners of a given system. The associations between the partners in biological consortia can be of different strengths, whose measure is a resistance to errors. As we will illustrate, many co-evolving biological systems exhibit rather high tolerance to such errors. In addition, we want to argue that similar to natural ecosystems, the intracellular environment represents a network of co-evolving entities. Hence, the same principles may operate in the co-evolution of leafcutter ant societies as well as cell compartments, components of biomolecular complexes and chains of nucleic acids and proteins representing “ecosystems” of nucleotides and amino acids, respectively (Galtier and Dutheil 2007; Lanier et al. 2017) (Tables 2 and 3). We will start with the co-evolution of nuclear and mitochondrial genetic systems and then move on to the chromosomal ends as they provide several lessons relevant for addressing the central theme of this study.

Table 1 Examples of co-evolution between and within species [for general introduction to co-evolution and other examples see (Futuyma and Slatkin 1983; Ridley 2004; Futuyma and Kirkpatrick 2017; Thompson 1989, 1994, 2005, 2014, 2017)]
Full size table
Table 2 Examples of co-evolution between nuclear and organellar genomes
Full size table
Table 3 Examples of inter- and intramolecular co-evolution of nucleic acids and proteins
Full size table

Nucleo-mitochondrial Compatibility: Limits and Consequences

The nucleus and the mitochondria are two compartments of a typical eukaryotic cell (Fig. 1b). Eukaryotes diversified from a common ancestor, which was accompanied by a concerted co-evolution in different phylogenetic lineages (Rand et al. 2004; Embley and Martin 2006; Greiner and Bock 2013; Archibald 2015; Sloan et al. 2018) (Table 2 provides examples of co-evolution of nucleus with mitochondria as well as with plastids, another organelle originated by intracellular endosymbiosis). Since subunits of respiratory chain complexes are encoded by both the mitochondrial and nuclear genomes, and must functionally interact, co-variation of the respective genes represents a compensatory mechanism affecting their fitness (Levin et al. 2014; Pfanner et al. 2019). Therefore, mixing the nucleus of one species with mitochondria from another is often unsuccessful. Indeed, nucleo-mitochondrial incompatibility may represent one of the mechanisms of reproductive isolation. This was shown by studies employing techniques of mitochondrial replacement between various strains of the yeast Saccharomyces cerevisiae (Sulo et al. 1989). Subsequently, it was demonstrated that it is possible to exchange mitochondria between closely related species of the genus Saccharomyces, yet, when their evolutionary distance is too large, the cybrid cell is unable to grow on respiratory substrates (Osuský et al. 1997). Similarly, transfer of mitochondria from several large primates (such as chimpanzee or gorilla) into human cells lacking mitochondrial DNA (mtDNA) (dubbed rho0) restored their oxidative phosphorylation (Kenyon and Moraes 1997), although all these cybrids exhibited a respiratory complex I deficiency (Barrientos et al. 1998). Yet again, there is a limit beyond which the functional transplantation is impossible: mitochondria from orangutan, monkeys and lemurs did not complement the respiratory deficiency of the human rho0 cells. In contrast to chimpanzee and gorilla, the human and orangutan nuclear and mitochondrial genomes are too different such that their gene products cannot make functional respiratory complexes (Kenyon and Moraes 1997). In yeasts, the differences between the nucleus and mitochondria can be very subtle, e.g. the presence or absence of an intron in mtDNA, or a difference in a single translation factor that make the two genetic systems incompatible (Špírek et al. 2000, 2014; Lee et al. 2008; Chou et al. 2010). Similar results were obtained in Drosophila indicating that the nuclear genetic background affects the compatibility of a corresponding mtDNA haplotype (Mossman et al. 2019b). Golden- and blue-winged warblers share 99.9% of their nuclear genome, but they have unique mitochondrial genomes resulting in nucleo-mitochondrial incompatibility and reproductive isolation (Hill 2017). These observations resulted in the “mitochondrial species concept” where a species is defined by a set of coadapted mitonuclear genotypes (Hill 2017). These studies were important not only from an evolutionary point of view but also for medicine, where the attempts to correct defects caused by a single mutation in mtDNA are based on the replacement of mitochondria with that of a healthy donor (Saxena et al. 2018). These replacements may fail because the recipient nuclear genome and the donor mtDNA are not compatible, as exemplified by numerous diseases caused by a single-nucleotide substitution in mtDNA (Wallace 2018; Mottis et al. 2019). The importance of maintaining the compatibility between nuclear and mitochondrial genomes has been demonstrated by a recent large-scale survey providing evidence for selection of mtDNA variants in female germline over a single generation (Wei et al. 2019). Like in the leafcutter ant colonies, the two genetic systems co-evolve and, when a change in one system is not tolerated by the other, the whole consortium fails.

A Curious Case of Mitochondrial Genomes

From the above examples, it would seem that the relationship between the nucleus and mitochondria tolerates only minor and incremental changes. This is not always the case, however, as illustrated by studies on the evolution of mitochondrial genome topology. Although usually depicted as a small circle, as it occurs in human mitochondria, plant and fungal mitochondrial genomes are usually present as a population of polydisperse DNA molecules of various lengths engaged in frequent recombination events (Bendich 1993; Gerhold et al. 2010). The majority of these molecules are linear and longer than the size of a single genome, but their physical mapping (such as restriction enzyme analysis or DNA sequencing) produces circular maps. However, as early as in 1968 it was shown that the mitochondrial genome of Tetrahymena is represented by linear molecules of defined lengths (Suyama and Miura 1968). Today, it is evident that the mitochondrial genetic system is subject to intense evolutionary experimentation (Burger et al. 2003), and that the topology of mitochondrial genomes can vary greatly in closely related organisms or even in different strains of the same species (Kosa et al. 2006; Valach et al. 2011, 2012). Early studies demonstrating the linearity of mitochondrial genomes in protists and yeasts (Suyama and Miura 1968; Goddard and Cummings 1975; Wésolowski and Fukuhara 1981; Kováč et al. 1984; Morin and Cech 1986; Dinouël et al. 1993; Fukuhara et al. 1993; Vahrenholz et al. 1993; Nosek et al. 1995) indicated that they face the same challenges as their nuclear counterparts, i.e. the end-replication and end-protection problems associated with the chromosomal termini (Blackburn 2010; de Lange 2018).

Organisms with linear mtDNA have dealt with these problems in intriguingly diverse ways (Fig. 2a) (Nosek and Tomáška 2003; Smith and Keeling 2015; Burger and Valach 2018). Indeed, there are many possible answers to the “How?” questions related to linear mitochondrial genomes, such as how their ends, termed mitochondrial telomeres (Morin and Cech 1986, 1988), are maintained. The wide repertoire of the solutions to the end-protection and end-replication problems is fascinating not only from a mechanistic, but also from an evolutionary point of view. The linearization of the originally circular genome appears to be very rapid; so how was the cell prepared to deal with all the challenges posed by the ends of the linear chromosomes? This leads to a series of “Why?” questions. Why do mitochondrial telomeres look the way they look? Why is their presence, which we now know occurs in phylogenetically independent branches, so frequent? Where do telomeric repeats with no known homologs come from? One possibility is that the reshaping of mitochondrial genomes is mediated by the recurrent invasion of mobile genetic elements, such as transposons or linear plasmids that play a sort of game for the mitochondria. In these games they perform a dual role: they cause a problem and at the same time they provide the solution by hijacking and employing the host machinery (Nosek et al. 2006; Valach et al. 2011). For example, terminal sequences of linear mitochondrial genome of Candida parapsilosis are formed by arrays of long (738 bp) tandem repeats (TRs) (Nosek et al. 1995). Employing the machinery for homologous recombination (HR) TRs recombine out of the genome forming extrachromosomal circular molecules (telomeric circles; t-circles). T-circles replicate via a rolling-circle (RC) mechanism resulting in an expansion of TRs and their re-integration into the main genome thus preventing its shortening (Tomáška et al. 2000, 2009; Nosek et al. 2005). All molecular components needed for HR and RC replication are present in mitochondria of various yeast species regardless of the topology of their mtDNA (Maleszka et al. 1991; Gerhold et al. 2010, 2014; Guaragnella et al. 2018), so the only requirement for the selfish element to act as a mitochondrial telomere is to be recognized as a substrate for HR and RC machineries.

Fig. 2
figure 2

Studies on the evolution of telomeres in both nuclear and mitochondrial DNA may help to elucidate the mechanisms of co-evolution at various levels. a Linear mtDNAs are frequently formed in various phylogenetic lineages of eukaryotes, likely as a result of invasion of selfish genetic elements (e.g. transposon, plasmid, virus). This is accompanied by the stabilization of linear DNA molecules via various types of terminal sequences/structures (i.e. telomeric array (t-arrays), telomeric loops (t-loops), telomeric hairpins (t-hairpins), or telomeric proteins (t-proteins)). The maintenance of mitochondrial telomeres depends on the same components of DNA replication and recombination machineries as those operating in organisms with a circular or polydisperse form of the mitochondrial genome. The arrow pointing back to the circular/polydisperse forms indicates that the linear genomes can undergo re-circularization and loss of telomeres. For further details see (Nosek et al. 2006; Kosa et al. 2006; Tomáška et al. 2009; Valach et al. 2011). b The diversification of nuclear telomeric sequences in yeasts is accompanied by frequent changes in the composition of proteins interacting with ds or ss regions of the telomeric DNA. In spite of a sequence divergence between telomeric repeats, at least some variants may retain the propensity to form secondary structures such as G-quadruplexes (Juríková et al., manuscript in preparation). Tay1 and Teb1 are homologous proteins and therefore they are depicted in the same color. c Co-evolution can be also studied at the level of individual biomacromolecules such as RNA (top), or proteins (bottom). For example, the sequences of the RNA subunit of telomerase (TER) in closely related yeast species rapidly diversified, yet they retain similar structural features (Tzfati et al. 2000; Zappulla and Cech 2004, 2006; Lin et al. 2004; Gunišová et al. 2009; Červenák et al. 2019). In proteins, some amino acid positions have a tendency to co-evolve (red letter), whereas others are either conserved (green letters), or are prone to frequent independent substitutions (black letters) (Halabi et al. 2009; McLaughlin et al. 2012). Shown are hypothetical examples. S1–S5, sequences in species 1–5

Full size image

And now perhaps more difficult questions related to our central topic arise. How is it possible that the two genomes, so dependent on each other, are able to accommodate such dramatic changes? The change in mtDNA topology may result from an alteration in the mtDNA itself, or it could be a direct consequence of a mutation in nuclear gene(s). However, how does a nuclear genome, encoding all proteins necessary for the replication and maintenance of mtDNA, react so rapidly to such dramatic change? If telomeres need special means to deal with all the problems associated with linear DNA molecules, where did all the components dedicated to dealing with these problems come from? How is it that the problems, so specifically and intricately solved in nuclei (see below) were dealt with so easily when encountered in the newly altered mitochondria? The answer, according to several studies, is that the components necessary for the stabilization and maintenance of mitochondrial telomeres did not have to evolve from scratch. In contrast, they were either provided by the invading mobile element (Fričová et al. 2010), or (and more often) they were already available and fulfilling other tasks important for mtDNA maintenance in general. The protein protecting the termini of mitochondrial telomeres is the same protein that serves as single-stranded (ss) DNA-binding protein during DNA transactions (Tomáška et al. 1997, 2001; Nosek et al. 1999); the machinery necessary for recombination-dependent maintenance of some types of mitochondrial telomeres is already present in mitochondria with circular-mapping genomes, and responsible for the replication and recombination of the molecules (Tomáška et al. 2000, 2009; Nosek et al. 2005; Miyakawa et al. 2009; Pevala et al. 2016).

But then the logical "Why?" question follows: why is a drastic alteration of the genomic architecture—such as its linearization—so well tolerated by the mitochondria and entire cell? This question is even more general: in addition to diverse topologies of mtDNA, the mitochondrial-genetic system exhibits other intriguing oddities such as extensive RNA editing (Simpson et al. 2000), a plethora of introns (Lang et al. 2007) and translational bypassing elements (byps) (Lang et al. 2014; Nosek et al. 2015), or alterations in the genetic code (Bernt et al. 2013). Here, the answer may lie in the polyploid nature of the mitochondrial genome. Because the intracellular population of mtDNAs consists of tens or hundreds of molecules, every now and then one may encounter a change that can be as dramatic as the invasion of a mobile element causing its linearization, or the insertion of an intron or byp into the coding region of one of its genes. In most cases, such an event is detrimental for the given molecule, its fitness will decrease, and it will eventually die out. But as a matter of probability and taking advantage of the robustness of the mitochondrial genetic system, such an accident may lead to a selectively neutral, or even fitter molecule, which, combined with mitochondrial segregation drift occurring during cell division, could lead to the successful propagation of such variant DNA and its (at least transient) fixation.

This hypothesis is supported by the fact that in addition to mtDNA, champions in experimenting with genomes (in many different respects) are viruses. After infection of the host cell, they also make numerous copies, and this population of molecules can be subject to similar "experimentation" as in the case of mtDNAs. This may be one of the reasons why we see such a wide repertoire not only of genome architectures, but also of other peculiar features associated with the life cycle of viruses, i.e. their mode of replication, transcription and translation events (Fields et al. 2013).

There are even more instances of the fast co-evolution of organellar genomes, where mitochondria were proven to be excellent playgrounds for evolutionary tinkering (Jacob 1977). This is exemplified by another class of molecules which one would intuitively predict to be rather conserved and resistant to evolutionary change. Namely, mtDNA is compacted into structures called mitochondrial nucleoids by DNA-binding proteins containing a high-mobility group (HMG)-box. As these proteins in mitochondria fulfill analogous functions as the most conserved eukaryotic proteins—histones—in the nucleus, it is quite surprising that they are among the fastest (if not the fastest) evolving group of mitochondrial proteins (Višacká et al. 2009). Their amino acid sequences as well as their affinity for diverse DNA structures are different (Bakkaiová et al. 2016), yet they all perform the same principal functions, that is to compact and protect mtDNA, as well as to mediate essential DNA transactions such as recombination and repair. They collectively indicate that the relationships between the nucleus and mitochondria are very liberal, and that the robustness of the mitochondrial genetic system provides a relatively large playground for evolutionary experimentation (Burger et al. 2003).

Co-evolutionary Tales of the Nuclear Telomeres

How prone to evolutionary experimentation is the nucleus? To address this question, we can use a similar “case study” as for the mitochondria. The ends of nuclear chromosomes face the same problems already mentioned for linear mitochondrial genomes: prevention of degradation, inappropriate DNA repair, and mediation of complete replication. Nuclear telomeres are usually formed by tandemly repeated sequences with a 3′ ss overhang at the very end of the molecule. The overhang is complementary to a short region of the telomerase RNA (TER) subunit that anchors it and also serves as a template for the catalytic subunit of the holoenzyme to extend the overhang and thus provide a substrate for the conventional DNA polymerase-based replication machinery (Shay and Wright 2019). The telomeric repeats interact with specific DNA-binding proteins (telomere-binding proteins, TBPs), that are principal components of a protective multisubunit complex often called shelterin or telosome (de Lange 2005; Xin et al. 2008); they also stabilize a structure called telomeric loop (t-loop) formed by invasion of the ss overhang into the duplex region of a telomere (Griffith et al. 1999). A platform for all these interactions is represented by the telomeric DNA sequence, which is specific for eukaryotic chromosomes. At least according to textbooks, telomeric DNAs are composed of an array of short G-rich repeats that are similar in organisms as diverse as protozoans, plants, fungi, and mammals (Blackburn 2010). With a few exceptions, such as Drosophila (Kordyukova et al. 2018), most eukaryotes have only minor (if any) variations of the sequence 5′-TTAGGG-3′. One would expect that the conserved nature of the telomeric sequence is reflected by a conservation of the shelterin complex protecting the ends of chromosomes and the telomerase involved in the synthesis of telomeric DNA repeats. The components of both shelterin and telomerase appear optimized for their functions, so much so that one does not intuitively see much room for evolutionary experimentation.

Yet, there are ascomycetous yeasts whose telomere repeats do not strictly follow the rule of being short and G-rich. They can be as short as 6 nucleotides, but their length can go up to 35 nucleotides. They can be G-rich, but the fraction of Gs in a repeat can go down to 25% (Gunišová et al. 2009; Červenák et al. 2017). The telomeric motifs can be different in closely related species and the telomeric tracts can even be made of heterogeneous repeats (Fig. 2b). In short, it seems that (almost) anything goes with telomeric repeats in ascomycetous yeasts! This is the answer to the "What?" question: we observe a great deal of variability of telomeric sequences among yeasts. Immediately, the "How?" questions arise: how is this interspecific variability generated and how are these noncanonical telomeres maintained? The answer to the first question seems to be simple: the sequence of telomeric repeats is dictated by a template region of TER, so any change in the template region will be reflected by a change in the telomeric repeat sequence. The latter, however, is the more difficult question: how do the TBPs adapted for a given telomeric motif keep up with the runaway evolution of telomeric repeats in yeasts?

Addressing these questions experimentally resulted in a few lessons. First, the evolution of TBPs does not maximize their specificity and/or affinity. For example, the Tay1 protein, a TBP from the yeast Yarrowia lipolytica (Kramara et al. 2010), has higher affinity to human telomeric repeats than to its native DNA substrate (Višacká et al. 2012). Conversely, the human telomeric proteins TRF1 and TRF2 have about a tenfold lower affinity for human telomeres compared with Tay1. Recently, it was demonstrated that an increase in the specificity of the Cdc13 protein of S. cerevisiae that binds to 3′ ss telomeric overhangs, compromises its function in vivo (Glustrom et al. 2018). This underlines one general principle: the kinetic properties of a DNA-binding protein are often tuned to inferior values, thus enabling its dynamic association with the target DNA. This is in line with the idea that there is an upper limit of the specificity of interaction between partners (e.g. enzyme–substrate, ligand-receptor, protein-DNA sequence), as too specific interactions would lack flexibility and a perfect recognition would be too rigid and possibly nonfunctional (Kováč 1987). This means that increasing the perfection of molecular recognition is not necessarily an optimal evolutionary strategy (Bar-Even et al. 2011). Thus, a decrease in binding specificity might be beneficial for the host, especially when the more promiscuous binding might be helpful to keep up with an increased rate of evolutionary diversification of the ligands (Tomáška et al. 2019b). The evolution of other DNA-binding proteins also underlines this point. For example, the increase in the binding affinity of herpes simplex virus processivity factor UL42 to DNA reduces the DNA replication fidelity (Jiang et al. 2009). Similarly, a decrease in specificity accompanied by an increase in flexibility of DNA-binding of transcription factors depends on an evolutionary context (Ramos and Barolo 2013; McKeown et al. 2014).

The flexibility of DNA-binding proteins may be due (at least in part) to regions known as "molecular recognition features" (MoRFs) that are present in many nucleotide-binding proteins in all domains of life (Mohan et al. 2006; Peng and Kurgan 2015; Yan et al. 2016). When in the free state (without a ligand), MoRFs are intrinsically disordered, extremely flexible and often lack well-defined tertiary structure. Upon recognition of their ligand (such as DNA or RNA sequence), MoRFs adopt a stable fold facilitating binding to their target. This property of MoRFs results in relatively wide repertoire of recognized sequences that are bound with relatively low affinity.

Even though such flexibility of DNA-binding proteins may act as a buffer helping to tolerate frequent changes in the target nucleotide sequences, at certain point it may reach its limits. In the case of telomeres, rapid diversification of telomeric sequences in ascomycetous yeasts is accompanied by frequent replacement of specific telomere-binding factors by proteins exhibiting a more flexible binding to the DNA sequence (Steinberg-Neifach and Lue 2015; Sepšiová et al. 2016; Červenák et al. 2017). There seem to be at least four different proteins—Rap1, Taz1, Tay1 and Tbf1—that are able to recognize and protect the telomeric tracts in distinct phylogenetic lineages (Červenák et al. 2017). This come-and-go strategy based on replacement of one TBP by another DNA-binding protein remains a mystery: not so much the ability of a DNA-binding protein to bind a particular telomeric sequence (they may have originally served as transcription factors whose cognate sequences are reminiscent of the telomere in a given species), but rather their ability to make a functional telosome/shelterin complex with other of its required components. The co-evolution of these components must be rather fast to retain functionality of the complex, yet its mechanisms are unknown.

There is another intriguing "Why?" question related to the variability of telomeric repeats in yeasts: if they are so diverse and do not comply with the textbook requirements for a "proper" telomeric repeat (i.e. short, G-rich and homogeneous), how is it that they can still perform telomeric functions, which are so conserved? Is there any structural feature that makes these particular sequences suitable (Fig. 2b)? Some, but not all can form G-quadruplexes in vitro (Tran et al. 2011), so this noncanonical DNA structure does not seem to be a conserved feature of yeast telomeric sequences. Thus, the sequence and/or structural constraints placed on yeast telomeric repeats are yet to be identified.

So far we have learned that co-evolution, which is mostly studied from the point of view of ecosystems of different species, is very central for the interdependent genetic systems within a cell and between components of biological complexes (either binary such as DNA–protein, or multicomponent complexes such as shelterin), but its principles also apply to individual molecules. For example, in the case of TER, it is not only the template domain that undergoes nucleotide substitution, but the entire molecule evolves extremely rapidly: so fast that identification of TER homologs solely by means of bioinformatics is almost impossible even in closely related species (Waldl et al. 2018). And yet the molecules tend to retain their secondary structural motifs that are important for other functions of TER, such as interaction with the catalytic subunit, factors recruiting the enzyme to telomeres, or factors required for its biogenesis (Fig. 2c) (Gunišová et al. 2009; Červenák et al. 2019). The substitutions in the molecule co-evolve so rapidly that it is rather difficult to reconstruct the evolutionary pathways leading to TERs in individual species. Some authors even argue that TER genes are opportunistically recruited from a large pool of long noncoding RNAs that by chance fit the other components of the telomerase holoenzyme (Beilstein et al. 2012). Although recent studies shed some doubt on this scenario (Song et al. 2019; Fajkus et al. 2019), such evolutionary strategy would take advantage of the promiscuity of the catalytic subunit and its ability to interact with a variety of RNAs and perform various moonlighting functions in the cell unrelated to telomere-maintenance (Sharma et al. 2012; Maida et al. 2016).

One may argue that a rapid co-variation of positions in sequences of noncoding RNA is not surprising, because all that counts is the secondary structure, not the sequence per se. This is in part a valid argument, but, interestingly, the question of co-evolution seems to be very important also for evolutionary changes within individual proteins (Shah et al. 2015). The possibility of reconstruction of protein evolution by means of “molecular (chemical) paleogenetics” was articulated by Pauling and Zuckerkandl in their seminal paper almost six decades ago (Pauling and Zuckerkandl 1963). Thanks to the rich protein databases it is now possible not only to “restore” amino acid sequences of ancestral polypeptides, but also to reconstitute most likely step-by-step evolutionary paths leading to contemporary polypeptides, i.e. exploring networks of protein sequence spaces (Ferrada and Wagner 2010). This is one of the main goals of a field of evolutionary biochemistry that has an ambition to dissect the physical mechanisms and evolutionary processes by which proteins diversified and to reveal how their physical architecture facilitates and constrains their evolution (Harms and Thornton 2013). Likelihood-based reconstructions of protein phylogenies rely on a stochastic model of the substitution process. Commonly used models (e.g., models in the general time-reversible) are overly simple (e.g. site-independence, constant rate matrix, no heterotachy), and there is evidence that the primary factor in amino acid replacement is epistasis in the genetic basis of protein structure, thermodynamic stability, substrate specificity, allostery, folding and function (Breen et al. 2012; Harms and Thornton 2013). This means that evolutionary pathways to the same location in protein sequence space may be very diverse such as a mutation that is beneficial in one genetic context may result in inactivation of the protein if present in a different context.

Importantly, even if there is selection only for a change in the substrate specificity of an enzyme, the critical amino acid substitutions are not limited to the active center, but also to other regions of the polypeptides, including those located at the surface on the opposite side of the protein. This indicates the existence of sparse networks of co-evolving amino acid positions (termed protein sectors) that comprise the essence of three-dimensional structure and function (Halabi et al. 2009; McLaughlin et al. 2012). Analysis of the co-evolutionary signal among interacting residues in present-day proteins is instrumental not only for understanding rules of protein evolution, but also for the design of novel proteins with new functional specificities.

Earlier obstacles in reconstitution of evolution of protein sequences (Galtier and Dutheil 2007) are currently alleviated by novel approaches to formally describe and model the constraints governing protein evolution (Goldstein and Pollock 2016; Szurmant and Weigt 2018). These models take into account the possibility that amino acid substitutions that are neutral represent a preadaptation that can allow a phenotypic manifestation of subsequent mutations. Such "permissive" mutations have no effect when introduced singly but are required for one or more other mutations to change the function of a protein (Ortlund et al. 2007). This process has been called genetic accommodation (West-Eberhard 2003), stabilizing selection (Schmalhausen 1986), and genetic assimilation (Waddington 1942, 1953). An important discovery stemming from the work on protein evolution is the "entrenchment" of resident amino acids and the "evolutionary Stokes shift" facilitating neutral amino acid replacements (Pollock et al. 2012; Shah et al. 2015). The "Stokes shift" concept is also broadly applicable to scenarios in which several distinct components of a system must co-evolve (e.g., niche spaces of species in an ecosystem) thus alluding to Waddington’s concept of genetic assimilation.

Multiple simultaneous mutations falling within a single open-reading frame and enabling such concerted co-evolution can occur by various mechanisms including mutagenic break repair within loci whose transcription is induced by stress (Fig. 2c) (Fitzgerald and Rosenberg 2019). One example illustrating this issue is represented by the evolution of enzymes in psychrophilic organisms whose maximal reaction rate (Vmax) for a given reaction is the same as it is for their homologues in mesophilic organisms. It is puzzling how the rate of the reaction can be the same at temperatures as different as 10 °C and 37 °C. In many cases, it was observed that the substitutions in the active center of the enzymes from psychrophiles were accompanied by substitutions towards glycines located at the surface (Saavedra et al. 2018).

This and other examples show that proteins, similar to ant colonies, represent “ecosystems”, whose components must effectively co-evolve to keep the whole system functional. In the case of TBPs, it would be interesting to see which amino acid substitutions must accompany those in DNA-binding sites to retain an optimal balance between affinity, specificity, and satisfy other general constraints including rapid and correct folding (Worth et al. 2009), thermodynamic stability (Zeldovich et al. 2007), solubility (Drummond et al. 2005), and the ability to functionally interact with other components of the telosome/shelterin complex.

Speculations on the Central Question

Here, we restate the central question of this paper: how do individual components of a biological system (e.g. ant colonies, cellular compartments, components of biological complexes, or amino acids and nucleotides in biomacromolecules) co-evolve in a jungle composed of myriad players interdependent on each other? Although some questions were addressed, in many instances we have only vague answers. Below we provide a few speculations and discuss their possible experimental assessment and relevance to the main topic of the paper, i.e. evolution of telomeres (Fig. 3).

Fig. 3
figure 3

Co-evolutionary scenarios of a system components such as those involved in the maintenance of telomeres or genome topology. a Mutation occurring in one partner (A) are neutral in terms of its interaction of the second partner (B). For example, the binding of protein to a ligand can be achieved via three different sites, so three different variants of a ligand have similar affinities to the protein binder. b Constructive neutral evolution (CNE) of biochemical complexity (Gray et al. 2010; Lukeš et al. 2011). An enzymatic reaction (r1) is carried out by an enzyme A that fortuitously interacts with a protein B (not having an effect on its activity). Inactivating mutation in A (red dot) is suppressed by interaction with B resulting in a functional partnership undergoing co-evolution that may lead to a gain-of-function mutation (green dots) resulting in a novel type of reaction (r2) (for details see Lukeš et al. (2011)). c The ancestral genomes may have accumulated different types of TBPs by gene or DNA-binding domain duplication followed by sub- or neo-functionalization (rectangles), and functional specialization. The scenario of evolution of yeast TBPs by means of constructive neutral evolution is detailed in Višacká et al. (2012). d “Permisive” neutral mutations (red and green) predispose the protein A to interact with B resulting in an ability of the complex to catalyze a novel reaction (r). e The decreased fitness caused by a primary mutation resulting in a loss of interaction between A and B can be alleviated by a secondary compensatory mutation in either one of the partners or another component (C) restoring the functional complex. f Group selection as means of evolution of genome topology. Population composed of a mixture of wild-type and mutated components can have a higher fitness than the homogeneous population. Circular and polydisperse (i) or linear mtDNA molecules carrying telomeres (ii) can serve for different function (e.g., replication, segregation, gene expression) and this division of labor can be advantageous for the group. (iii) A mixture of both wild-type (black circles) and mutant (red circles) mtDNA molecules can be advantageous under some environmental conditions [e.g., during wine fermentation in S. cerevisiae (Taylor et al. 2002; Jasmin and Zeyl 2014)]. A genetic drift (gray arrows) can subsequently result in homogenization of the population. See text for details

Full size image

As articulated by Kimura (1968), many mutations are neutral and they do not have any effect on fitness (Fig. 3a). Hence, in many cases, we do not need to look for just-so-stories to explain their adaptive value. Such changes might have been forced by selfish genetic elements and, benefitting from the robustness of the system, they represent successful evolutionary experiments without leading to a selective advantage for the host. For example, spreading tandem repeats at the ends of a linearized chromosome results in a substrate for the homologous recombination machinery that can help stabilize them by mediating the formation of t-loops (Tomáška et al. 2009, 2019a; de Lange 2015; Červenák et al. 2019). Transcription of the repeats may enhance the frequency of the t-loop formation, thus solving both end-protection and end-replication problems (Kar et al. 2016; Tomáška et al. 2019a). According to this scenario, the host cell had been equipped with all necessary tools to maintain the chromosomal ends at the time of its linearization and the subsequent formation of telomeric repeats may have been a neutral adaptation in terms of the fitness of the host cell. The role of neutral mutations leading to complexity was articulated in the concept of “constructive neutral evolution (CNE)” (Stoltzfus 1999, 2012) and later “neutral evolutionary ratchet” (Gray et al. 2010; Lukeš et al. 2011). According to this hypothesis “if an autonomously functioning cellular component acquires mutations that make it dependent for function on another, pre-existing component or process, and if there are multiple ways in which such dependence may arise, then dependence inevitably will arise and reversal to independence is unlikely. Thus, CNE is a unidirectional evolutionary ratchet leading to complexity” (Lukeš et al. 2011) (Fig. 3b). The CNE could have played a role in evolution of yeast TBPs. According to such scenario, the ancestral genomes accumulated precursors of all three types of proteins possibly via the CNE and involving gene and domain duplications (Ohno et al. 1968; Ohno 1970). This resulted in a complex set of proteins that could have adapted to either general functions related to regulation of gene expression or more specialized functions at telomeres (Fig. 3c). In some cases, the proteins started to play both telomere- and non- telomere-associated roles (Rap1p in S. cerevisiae, Tay1p in Y. lipolytica) making them essential components of the cell. In other cases, the proteins were adopted for a specialized function(s), leading to loss of DNA-binding activity (Rap1p in fission yeasts or mammalian cells) or a complete loss of the gene (Višacká et al. 2012).

In some cases, co-evolution may be driven by neutral mutations, but by other means. It was shown in several instances that several neutral mutations (sometimes occurring in a particular order) are required for a positive effect of the one genetic change that leads to increased fitness. Without these accompanying neutral mutations, the particular genetic change might have no, or even a detrimental effect on the biological system. This interplay between neutral (“permissive”) and positive mutations may drive the co-evolution of biological systems (Fig. 3d). This concept gained support from long-term evolutionary experiments by the Lenski group, who cultivates parallel Escherichia coli cultures under glucose limitations for thousands of generations (Fox and Lenski 2015; Lenski 2017). They proposed a three-step model of evolutionary innovations in a complex organism: (1) potentiation—genetic background evolves in which a trait is mutationally accessible, making the trait's evolution possible; (2) actualization—a mutation occurs that produces the trait, making it manifest, albeit likely in a weak form; and (3) refinement—once the trait exists, if it provides selective benefit, mutations will accumulate that improve the trait, making it more efficient. This phase is open-ended, and will continue so long as refining mutations arise and the trait remains beneficial (Blount et al. 2012). This may also be relevant for the precancerous state of a cell where a particular set of mutations sensitizes it to the effect of a mutation that turns it cancerous. Without the pre-existing neutral mutation(s) such a change would not have a phenotypic effect (Hanahan and Weinberg 2011). This scenario is consistent with the models of protein evolution identifying some amino acid substitutions as a preadaptation required for a phenotypic manifestation of subsequent mutations (Ortlund et al. 2007). What changes were required to replace a TTAGGG-repeat-binding protein (such as Tay1) by a more flexible TBP (e.g., Taz1)? Evolutionary experiments addressing this question could be based on placing Myb domain of Tay1p under selection pressure to change its DNA-binding properties and identification of the accompanying amino acid substitutions. Placing the affected residues on a 3D models of the evolved Myb domain and assessment its DNA-binding properties would lead to identification of both permissive and functional (positive) mutations.

In other cases, the genetic alterations causing a particular phenotype may put a selection pressure for compensatory mutations somewhere else in the genome (Fig. 3e). For example, it was shown that deletions of nonessential genes in the baker’s yeast genome are often followed by secondary mutations that compensate the phenotypic defect(s) caused by the primary genetic change (Teng et al. 2013). Intriguingly, for a given deletion, the repertoire of the selected secondary mutations is very limited and, to some extent, predictable. Similarly, in bacterial strains that are resistant to an antibiotic, elimination of the drug from the host or cultivation media leads to either rapid reversion of the mutation that represents a burden for cell fitness, or the appearance of secondary compensatory mutations that alleviate this burden in the corresponding cell (Andersson and Hughes 2010; Durão et al. 2018). Such clones appear very rapidly and often predictably as a result of fast co-evolution at the genomic level. In this context, it would be interesting to see what changes in the genomic sequence accompany evolution of cells with non-functional telomerase. This question can be addressed in yeasts by propagating independent clones of telomerase-deficient mutants and identification of mutations that would be common for the clones. Alternatively, controlled reprogramming telomeric repeats using telomerase RNA with a modified template region can be instrumental in identification of secondary genomic mutations that are selected to retain (or increase) the fitness of the cells. For example, studies on “humanized” telomeres in S. cerevisiae already shed some light on the components involved in evolutionary adaptation of the yeast cells to such dramatic changes in the sequences of chromosomal ends (Henning et al. 1998; Brevet et al. 2003; di Domenico et al. 2009).

Finally, in some cases, group selection explains some types of co-evolution in ecosystems (Okasha 2008; Nowak et al. 2010; Sober and Wilson 2011) (Fig. 3f). It is possible that in the case of telomere evolution, mixed strategies provide an advantage to the group during their evolutionary transitions. A cell containing a heterogeneous population of molecules of a given kind (e.g., mitochondrial genomes, heterogeneous types of telomeric repeats) may have higher fitness compared to cells with a homogeneous population. For example, during the transition state, circular molecules can serve as substrates for replication, whereas linear molecules would enable efficient segregation into daughter cells (Ling and Shibata 2002; Kosa et al. 2006; Valach et al. 2011). Studies on mixed populations of baker’s yeast that contain cells with both wild-type (rho+) and mutant (rho−/rho0) mitochondria provide insight into multiple levels of selection in the evolution of cooperating groups (Taylor et al. 2002; Jasmin and Zeyl 2014). In case of rapid evolution of nuclear telomeres in yeasts, a temporary recognition of telomeric repeats by two different TBPs may have been instrumental in a transition from mammalian type of nuclear telomeric repeats to more complex repeats. The models of mixed physiological strategies in microbial populations, or behavioral strategies in animals indicate that group selection may at least in part explain the co-evolution of intricate biological systems.

Conclusions

The main objective of this review was to highlight how investigation of evolution of mitochondrial and nuclear telomeres may be instrumental in addressing some questions related to co-evolution. Instead of providing ultimate answers to these questions, the reader may allow us to conclude with a metaphor. Co-evolution could be compared to a symphony, where different musicians play from notes for the given instrument, but the whole piece is written in a musical score where all the lines fit together in nice harmony. During the “evolution” of the symphony, the composer may take advantage of the robustness of the whole piece and experiment with the notes for a particular instrument and make adjustments in the other scores to avoid false tunes. The process of co-evolving scores in the symphony is certainly of a different kind than that of co-evolving biological systems and we may know very little about both processes. However, the result, to use a classic quote (Darwin 1859) is similar: "endless forms [biological or musical] most beautiful and most wonderful have been, and are being, evolved."