Faraway, so close. The comparative method and the potential of non-model animals in mitochondrial research

Abstract

Inference from model organisms has been the engine for many discoveries in life science, but indiscriminate generalization leads to oversimplifications and misconceptions. Model organisms and inductive reasoning are irreplaceable: there is no other way to tackle the complexity of living systems. At the same time, it is not advisable to infer general patterns from a restricted number of species, which are very far from being representative of the diversity of life. Not all models are equal. Some organisms are suitable to find similarities across species, other highly specialized organisms can be used to focus on differences. In this opinion piece, we discuss the dominance of the mechanistic/reductionist approach in life sciences and make a case for an enhanced application of the comparative approach to study processes in all their various forms across different organisms. We also enlist some rising animal models in mitochondrial research, to exemplify how non-model organisms can be chosen in a comparative framework. These taxa often do not possess implemented tools and dedicated methods/resources. However, because of specific features, they have the potential to address still unanswered biological questions. Finally, we discuss future perspectives and caveats of the comparative method in the age of ‘big data’.

This article is part of the theme issue ‘Linking the mitochondrial genotype to phenotype: a complex endeavour’.

1. Model organisms

For a large number of problems there will be some animal of choice or a few such animals on which it can be most conveniently studied

—August Krogh, 1929 [1, p. 247]

In August 1929, at the Thirteenth International Physiological Congress in Boston, the Danish scientist Schack August Steenberg Krogh delivered a lecture entitled ‘The Progress of Physiology’ [1]. Krogh touched on many significant points—still relevant to this day—but the focus of his talk was the need for physiology to undertake a comparative approach to study processes in all their various forms across different organisms. In this context, Krogh emphasized the importance of choosing the right model for a given biological problem. As an example, he cited his former teacher, Christian Bohr (father of the famous physicist Niels Bohr), that used a tortoise to study the respiratory mechanism of the lung because its anatomical features facilitated the experiments (see [2] for a review). A postdoctoral assistant attended Krogh's lecture, and was deeply inspired by it. His name was Hans Adolf Krebs, who discovered the citric acid cycle (also known as Krebs cycle or tricarboxylic acid cycle). Krebs was so inspired that 46 years later wrote a paper entitled ‘The August Krogh principle: for many problems, there is an animal in which it can be most conveniently studied’ [3]. At the time of Krogh's lecture, Krebs was working in the laboratory of Otto Heinrich Warburg who had great success by choosing the right models, such as the sea urchin to study the changes in energy metabolism upon fertilization, the unicellular alga Chlorella vulgaris to study photosynthesis, and rats to study the metabolism of cancer [3]. In particular, the latter studies led to the discovery of the Warburg effect (cancer cells tend to favour glycolysis over oxidative phosphorylation (OXPHOS), even in aerobic conditions). Krebs himself applied the Krogh principle by choosing the pigeon breast muscle to study the pathways of oxidative metabolism. A causal correlation between great accomplishments in life science and well-chosen model organisms is undeniable, and has been acknowledged for a long time. However, given the complexity of biological systems and the astonishing diversity of life, it is important to use models carefully, to fully exploit the advantages of the reductionist approach without falling into over-extrapolation.

What is a model organism? According to the definition by Levy & Currie [4, p. 328], a model is ‘a cognitive stand-in: instead of investigating the phenomenon directly, one studies an easier to handle alternative’. In biology, the term is used in the context of mathematical modelling and computational simulations (theoretical models), and referred to particular organisms (model organisms). First, in its most familiar form, a model is a modified theoretical analogue that retains some features of a phenomenon under study, while simplifying others; analysing a simplified surrogate based on a mathematical construct it is possible to draw conclusions about a real-world target system. In the second acceptation, a model organism is a representative specimen of another organism or of a broader biological group, and inferences from such models are empirical extrapolations [4]. Jessica Bolker [5] classified model organisms in two types: surrogate and exemplary models. This distinction is important because it entails quite different criteria regarding the choice of the model, and the different kinds of conclusions that can be drawn from studying it. On the one hand, surrogate models are mostly used in biomedical research to understand the mechanism of human diseases and to develop treatments, but the same approach is also applied to commercially relevant species and for the conservation of endangered species (e.g. assisted reproduction). A fundamental assumption that is made when using such type of models is that the surrogate (e.g. the mouse) will respond to manipulation in the same way as the target (e.g. humans), in other words, we assume that the biological process we are investigating in the surrogate model is shared with the target, and we translate the findings from surrogate to target by direct substitution. Accordingly, the surrogate does not need to be phylogenetically close to the target, the only important thing is that the surrogate matches the target in regard to the investigated biological process [5]. On the other hand, exemplary models are used in basic science as representatives of broader groups with the purpose of understanding fundamental biological mechanisms and evolutionary processes. In this case, the results obtained are used to draw general conclusions based on inductive reasoning—that is making broad generalizations from specific observations—and the phylogenetic context is important, because it allows us to evaluate the representativeness of the model in a wider context with respect to particular traits [5].

The two approaches have been extremely successful, but it is important to issue caveats and limitations associated with both of them. As evolutionary biologists—and given the evolutionary framework of this Issue—we are focusing on exemplary model organisms. In this context, the most important concept that we need to keep in mind is that inductive reasoning is a powerful method, but can lead to false conclusions even if all the premises are true [6]. We discuss the caveats associated with generalization in the following section.

2. The problem with overgeneralization

Generalization is a double-edged sword. Inference from model organisms has been the engine for many discoveries in life science, but indiscriminate generalization leads to oversimplifications and misconceptions. It is quite common—both in specialized publications and even more so in the mass-media—to find titles of articles that omits the model organism from which the results were obtained. In this way the pairing between the model organism and the taxon that the model is representing becomes invisible [5]. The scientific community is well-aware of such misbehaviour, as proven by the popularity of the Twitter handle @justsaysinmice run by James Heathers, a data scientist at Northeastern University in Boston, Massachusetts. All that the account does is to retweet a news story or press release containing a catchpenny headline or bold statements, and adding the words ‘IN MICE’ to it. The account was created in April 2019, and now (just three months later) it counts nearly 66 000 followers. What initially started as a joke is now generating debates about the misuse of generalization and inductive reasoning. To omit the subject of the study, especially (but not only) in biomedical research, generates the misconception that the finding is valid for every/most organisms, humans included. The bad practice of inflating research findings by overgeneralization can originate either at the source—owing to the increasing pressure for publishing ‘striking and breakthrough’ research in order to get funding or make a career—or during the process of promulgation to the public. The dissemination of wrong or distorted concepts is particularly bothersome because it undermines the trust of people towards science, or, even worse, creates anti-scientific mediatic monsters. There is an ongoing crisis of mistrust in science [7], and misinformation is playing a key role—albeit not being the only cause—in the decline of public confidence in scientists and experts in general. We think that the scientific community should take action to reverse the trend that has been pushing science in the direction of sensationalism and ‘breaking news’.

During his life, Thomas Hunt Morgan—one of the fathers of genetics and responsible for making Drosophila a model organism—worked on a considerable number of different species [8], actually following Krogh's principle, nevertheless, he considered the results of research on fruit flies as applicable to all sexually reproducing organisms. Such perception regarding Drosophila and other model organisms is quite common among biologists and almost ubiquitous in modern textbooks [4]. Francis Crick called his theory about the information flow from DNA to proteins ‘Central Dogma’ [9]. A dogma is a principle laid down by an authority as incontrovertibly true, which is a concept foreign to science. According to Sydney Brenner—who established Caenorhabditis elegans as a model organism—Crick regretted using the word ‘dogma’, even if it was a joke, because Crick was not religious and did not believe in dogmas [10]. Also Jacques Monod, in 1954, stated that ‘Anything found to be true of E. coli must also be true of elephants'. Actually, the phrase was coined by the microbiologist Albert Jan Kluyver in 1926 [11], in any case, this is another famous example of an overstatement, it is only partially true.

The problem with generalization was well known to both Krogh and Krebs. Krogh considered ‘a general physiology which can describe the essential characteristics of matter in the living state’ as an ideal goal that might need a long time to be achieved, and only by studying the same processes/structures in ‘myriads of organisms’ [1, pp. 246–247; 12]. In 1980, Hans Krebs, together with John Krebs, wrote a paper warning about the ‘uncritical application’ of the Krogh principle [13]. As they put it: ‘This article is a postscript to the earlier essay. It makes the point that an uncritical application of the principle may lead to fallacious generalizations, because extrapolating findings from one species to another is not invariably valid [13, p. 379].’ They also pointed out that while a ‘unity of life’ certainly exists, there is also an ‘infinite diversity of life’, based on different combinations of a small number of basic units. Some biologists are devoted to finding common features, others are devoted to finding diversity, and often the choice between being, respectively, a ‘lumper’ or a ‘splitter’ depends on the personal attitude towards the same scientific issues. Needless to say, both approaches are equally important, and need to be carried out together [14]. Biology is extremely complex, but while simplification is fundamental to understand complex mechanisms, we cannot underestimate diversity [5,12,14–18]. Diversity is both the motor and the result of the evolutionary process. Evolution ‘overcomes’ problems using whatever is available at the moment—a process described by François Jacob as ‘tinkering’ [19]—thus finding multiple solutions for a given problem. An astronomically high number of genotypes can produce the same phenotype, and all these genotypes are organized in networks connected by point mutations [20]. At the same time, the chances of overcoming a problem depend on what is available, namely from variability: contingency is an intrinsic property of evolution, so is diversity. For these reasons, it would be wise to avoid the terms ‘dogma’ or ‘law’ in biology.

Model organisms and inductive reasoning are irreplaceable: there is no other way to tackle the complexity of living systems. At the same time, it is not advisable to infer general patterns from a very limited number of species, which are very far from being representative of the diversity of life. How to solve this conundrum? August Krogh in 1929 pointed the way by endorsing the comparative method: ‘study all sorts of organisms'. Some are better than others, depending on the problem under study. We must choose an organism because it has properties suitable for investigating a particular phenomenon, not as a representative for a wide taxonomic group or life in general. We must have ‘myriads’ of models.

3. ‘Rosetta Stones’ or aberrant creatures? The model-organism-based approach versus the comparative method

Model organisms are often chosen because of practical advantages. They are usually easy to find and breed, convenient to maintain in the laboratory, have a short generation time and rapid development, and respond well to experimental techniques and manipulations [5,17]. We will incorporate such and related features under the term ‘accessibility’, which is what makes a model system more tractable and allows the researchers to focus on specific biological questions, disentangling them from a complex background. Other than accessibility, when choosing a model system it is fundamental to assess its ‘representativeness’—that is, the ability to serve as a basis for inference [5]—with respect to the biological feature that is investigated. At the dawn of molecular biology, bacteria and bacteriophages were used as model organisms to study DNA replication, transcription and protein synthesis. Because such processes and the molecules involved are shared by all organisms on this planet, the representativeness of these models is very high, and the choice turned out to be quite successful. In that context, bacteria and bacteriophages were ‘Rosetta Stones’—namely, something that acts as a key to some previously unattainable understanding [14]. Moving on to studying more complex biological processes, the researchers needed to choose other model organisms, like C. elegans, Drosophila melanogaster, Mus musculus, Rattus norvegicus and others (see [21] for a detailed analysis). Of course, over time, as more knowledge accumulates, the life science community is going to study increasingly more complex traits, and this is where biodiversity kicks in, undermining the representativeness of the ‘traditional’ model organisms. As pointed out by Krogh, the solution is to find multiple models, that is, to explore diversity to have a complete picture of the subject of study. From a conceptual point of view, on the opposite side of ‘Rosetta Stones’, there are ‘aberrant creatures’ [22] or ‘extreme organisms’, that is, highly specialized organisms that show very peculiar features. Frequently, the traits that make an organism an attractive model are the result of adaptations, distinctive features that can be experimentally convenient and increase its accessibility. One good example is the work of Krebs on OXPHOS that was carried out on pigeon breast muscles because of the robustness of the mitochondria in this kind of tissue. The adaptation to flight made breast muscle mitochondria in birds particularly resistant to stress, thus to experimental protocols [18]. However, adaptations reduce the representativeness of a model, so there is often a trade-off between accessibility and representativeness [5], which needs to be taken into consideration when extrapolating findings from a model species to more inclusive taxa. It is interesting to note that while ‘Rosetta Stones’ are suitable to find similarities across species—even if genuine ‘Rosetta Stones’ for living systems, in general, do not exist—extreme organisms can be used to focus on differences. Such kinds of negative models can be exploited to understand, for example, plasticity and limitations of a biological process (e.g. naked mole-rats for cancer research, and snakes for physiological adaptation to starvation, see [18] for a thorough discussion).

(a). How to infer general concepts and the role of model organisms in comparative biology

When making inferences in biology, we have to pay particular attention to the kind of questions we want to address and to the taxonomic range we want to investigate: are we looking for uniqueness or for more general mechanisms? A significant contribution to the understanding of this dualism is the concept of idiographics versus nomothetics. These terms were used by the Neo-Kantian philosopher Wilhelm Windelband to describe two distinct approaches to knowledge [23,24]. Several disciplines—historical sciences such as anthropology, palaeobiology, evolutionary biology and evo-devo—embody both principles. Idiographics is the description of unique and historically contingent features, while nomothetics is the discovery of regularities that can be described as general rules, and concerns the formulation of general concepts. This dualistic perspective helps us to understand the role of model organisms in biological disciplines [25]. When we study a specific model organism, we can only deal with the idiographic range. It is not correct to generalize when studying a single taxon, because the comparison between individuals of the same species can provide insights into the biology and evolution of the species itself, but not necessarily into the biology and evolution of its relatives or other distant taxa. Generalizations are easy to propose in biology but seldom valid across all organisms [26,27]. How can we translate unique features to regularities? Namely, how can idiographics be related to nomothetic conceptual categories? We can compare and generalize only if we have a rich idiographic foundation that documents the evolutionary change of the traits we are taking into consideration in our study. This is the reason why there is a need of a judicious choice of model organisms to mine for best idiographics [25]. What is the best choice? If we minimize character overlap between model organisms by maximizing phylogenetic diversity, we consequently maximize the amount of unique idiographic details captured by the models. Consequently, we minimize the ability to draw general conclusions from them. As we pointed out above, given that model organisms are usually chosen for specific traits, any general conclusion drawn from these species is rarely generalizable [28].

Even though analysing a broad phylogenetic range of taxa is fundamental to assess diversity, it may not be the best choice to generate nomothetic insights [25]. We can access nomothetic themes if we maximize character overlaps between model organisms, choosing them for specific shared traits, as part of a same class. A class is a group of biological entities (e.g.: organisms or parts of them, biochemical pathways, gene regulatory networks, etc.) that share features which can be meaningfully compared. Such a concept expands, or ‘completes’, the concept of homology, and allows for comparisons beyond monophyletic taxa [27]. A class can be a group of organisms that share common traits, even if independently evolved, thus distinguished from the taxonomic term ‘class’. From here ahead, when speaking about classes, we refer to the non-taxonomic entity. This ahistorical formulation of classes provides the necessary basis to propose nomothetic insights. Idiographics are most efficiently translated into nomothetic insights when model organisms are chosen based on sharing certain traits, that is, as members of the same class [25,27]. So, models should be chosen for possessing traits that provide an independent support for a particular concept rather than for their phylogenetic position per se; by doing this, the explanatory force—namely, the extension of predictive reliability—can be maximized. Of course, the availability of robust and extensive phylogenetic hypotheses is still fundamental to understand the direction of evolutionary change and the conditions under which certain traits can evolve.

To understand how we can approach generalization, we have to consider the specific jurisdiction in which the outcome of a specific experiment can be included. We have to consider the ‘experimental model-organism-based approach’ versus the comparative approach. There is no a priori reason to generalize the outcome of experiments based on a single model organism, as a single counterexample actually falsifies any generalization [29]. Furthermore, this kind of approach cannot give direct evolutionary inference, because the experimental changes that we can induce in the laboratory are just an analogy to evolutionary changes: it is not correct to compare a newly obtained phenotype in one species with naturally occurring phenotypes in that species or in others, even if similar, and deduce that a similar change has indeed happened in the course of evolution [29]. Similarly, it is not correct to compare the performance and fitness of manipulated organisms with ‘wild-type’ organisms, because the latter are functionally balanced and probably mutations disturb this functional balance [29]. For this reason, a reduced fitness may not be a support for the evolutionary stability of characters: the proximate causes of functional disadvantages of change cannot directly explain the ultimate causes of evolutionary change [29]. Only comparison allows generalization.

Most of the research performed with model species has been justified by their potential power for understanding human biology. To trace general themes and eventually understand life, we have to overcome such limited view and expand research to ‘new’ model organisms from a wider taxonomic range and in a wider set of environmental conditions. The term model organism was initially used to describe ‘an organism that is inherently convenient to study a particular area of biology’. When saying ‘model organism’, we often mean ‘an organism for which a wealth of tools and resources exist’ [30]. To date, several organisms are used as models particularly suited to address some research areas—thus being model organisms in the original sense, that is convenient for the study of a biological process—but these often do not possess implemented tools, and dedicated methods and resources. Russell et al. [30] refer to these organisms as ‘non-model model organisms' (NMMO).

(b). An example of the use of classes: the link between mitochondria and germline formation

A clear example of the use of classes in building reliable generalizations regards germline development [31] and its link with the mitochondrial inheritance process. Looking at the majority of model species studied so far, the segregation of germline from somatic lineages appears to happen early. The so-called preformation mode—that is, early development controlled by maternal factors stored in the egg—seemed to be the predominant mode of germline specification. Preformation was described for C. elegans, Drosophila, Gallus gallus and plenty of other models. However, as data from additional animal taxa are collected, it is becoming clear that the most common and ancestral mode is epigenesis, characterized by a later segregation of germ cells owing to inductive signals in the developing embryo [31]. The first idea was the result of the biased choice of models which have in common a faster development. Actually, developmental speed is deeply connected to the use of maternal material in early development. If transcripts newly produced by the embryo and their products are necessary for development to proceed, of course, the process is slowed down. If we chose models with slow development, the chance of choosing animals developing through inductive signals would probably increase. If we consider the class of taxa sharing the preformation mode of germline specification, we may be able to identify the general features that can lead from the ancestral epigenesis to an early way of germline specification and understand how the segregation of material delivered to the progeny has evolved. This kind of approach can be applied to any kind of meaningful trait.

The relevance of germline formation goes well beyond the field of developmental biology. Investigating the times and modes of mitochondrial segregation into the germline is of basic importance: the mechanisms leading to the segregation of mitochondria into germ cells can vary depending on the type of germline specification, early or late [32]. By sampling animals with one or the other mode, in distantly related taxa, we have the possibility to address the general principles that lead to one or the other way of specification and at the same time to find, for the two modes, the linked mechanism of mitochondrial selection [32].

Given the increasing knowledge about the role of mitochondria in germline formation, these two topics have become tightly intermingled. The presence in many distantly related taxa of mitochondria associated with the germ plasm—material found in germ cells and containing germline determinants—suggested a general mechanism: mitochondria release material for germline specification/formation. Many studies on several animals documented the emission of mitochondrial material merging with germ plasm; this was observed in model organisms such as D. melanogaster, Hemicentrotus pulcherrimus, Xenopus laevis and M. musculus (see [33,34], and references therein), as well as in non-model organisms (e.g.: [35]). Moreover, a mitochondrial ribosome-dependent translation in germline formation is required at least in Drosophila embryos [36]. However, despite this evident link between mitochondria and germ plasm, the mechanism of such interplay is still unknown.

4. Rising non-model model animals in mitochondrial research

The growing understanding of the diversity of life clearly advises against the routine of formulating hypotheses and inferring general patterns based on a small number of species. Historically, general conclusions inferred from studies based on a skewed and limited taxon sampling led to a restricted and incomplete knowledge of mitochondrial biology. This uneven sampling is reported in table 1 which shows the number of complete mitochondrial genomes available in the NCBI Nucleotide database, subdivided across animal phyla. The poor (or non-existent) sampling in some groups is evident, as it is the skewed sampling in other groups. For example, human mitochondrial DNA (mtDNA) represent the 76.7% of all the chordate sequences deposited in the database. More recent evolutionary research (see for example: [37–40]) produced clear evidence that some of the ‘textbook notions’ about mitochondria are to be deemed as inaccurate, or plainly wrong. A major consequence of this is an increased awareness of the central role of mitochondria in eukaryotic life that goes beyond the concept of ‘powerhouse’ [41–45]. Accordingly, Geoffrey Hill recently coined the term ‘mitonuclear ecology’ to indicate the emerging interdisciplinary field that studies fundamental concepts of evolutionary ecology such as sexual reproduction, the origin of two sexes, sexual selection, adaptation and speciation in the light of mitonuclear interactions [41,46]. Exploratory research on non-model organisms broadened the variability of mitochondrial biological features well beyond the ‘textbook notions’ [47]. For this reason, it is necessary to exploit the new technologies to extend the number of species under investigation, including new models and understudied groups. Such ‘uncommon’ biological systems can help unveiling unknown elements of mitochondrial biology and evolution. In this section, we discuss the potential of non-model animal systems in mitochondrial research and highlight some promising candidates. Moreover, this Issue includes research papers focused on non-model animals such as bivalve molluscs [48], electric fishes [49] and mayflies [50].

Table 1.

Availability of complete mitochondrial genomes on the NCBI Nucleotide Database. (The table shows the number of living species for each animal taxon, as reported by the 2019 Annual Checklist of the ‘Catalogue of Life’ (CoL) project (https://www.catalogueoflife.org/col/info/totals), and the number of complete mitochondrial genomes available on the NCBI Nucleotide database (data collected on 7 July 2019). In the ‘notes’ column, we indicated strong biases towards a restricted number of species. Between brackets the percentage of mitochondrial genomes of the cited species, with respect to the total number of available genomes.

taxon	current number of living species in CoL	no. of complete mt genomes	notes
Acanthocephala	1330	7
Annelida	14 399	100
Arthropoda	1 082 297	4427	Anopheles spp. 393; Drosophila spp. 206; Magicicada spp. 125; Daphnia spp. 90; Apis mellifera 67; Bombyx morit 61. Total = 942 (21.3%)
Brachiopoda	396	5
Bryozoa	5434	10
Cephalorhyncha	237	3
Chaetognatha	132	11
Chordata	69 913	63 801	Homo sapiens 48 026; Pan spp. 239; Menidia menidia 190; Camelus spp. 179; Mus spp. 162; Rattus norvegicus 118. Total = 48 914 (76.7%). Homo sapiens = 75.3%
Cnidaria	11 151	237	Acropora spp. 72 (30.4%)
Ctenophora	200	5
Cycliophora	2	0
Dicyemida	122	0
Echinodermata	6828	96
Entoprocta	171	2
Gastrotricha	852	1
Gnathostomulida	100	2
Hemichordata	139	6
Micrognathozoa	1	0
Mollusca	65 442	791	Architeuthis spp. 38; Mytilus spp. 38; Potamogyrus spp. 23; Octopus spp. 21; Unio spp. 20; Anodonta spp. 12. Total = 152 (19.2%)
Myxozoa	245	4
Nematoda	3455	258	Anisakis spp. 17; Caenorhabditis spp. 17; Ascaris spp. 12; Angiostrongylus spp. 11. Total = 57 (22%)
Nematomorpha	361	0
Nemertea	1371	19
Onychophora	167	5
Orthonectida	25	0
Phoronida	19	1
Placozoa	2	6	Polyplacotoma mediterranea 1; Trichoplax aderens 1; ‘Placozoan sp.’ 4.
Platyhelminthes	18 616	436	Schistosoma spp. 149; Echinococcus spp. 113; Taenia spp. 21. Total = 283 (64.9%)
Porifera	9092	66
Rotifera	2014	3
Sipuncula	205	7
Tardigrada	1018	4
Xenacoelomorpha	456	12

There are notable examples of NMMOs systems in mitochondrial biology. We enlist some representatives in table 2, including fishes, mammals, birds, crustaceans and bivalve molluscs. This is not intended to be a comprehensive review of the whole range of NMMO animal species under study to date, but it is just an example of how new model organisms can be chosen in a comparative biology framework. The NMMOs highlighted here have special, often unique, traits to address questions of general importance. These organisms are studied not only because they are a kind of ‘aberrant creatures’ useful to gather novel idiographic records, but they can also be fundamental system to address central questions about biology that have remained unanswered to this day [30].

Table 2.

Non-model model animals in mitochondrial studies.

taxon	special features	investigated traits	references
class Actinopterygii
Fundulus heteroclitus (Atlantic killifish) Nothobranchius furzeri (African turquoise killifish)	inhabits a steep latitudinal thermal gradient, with variation in mitochondrial properties currently, the shortest-lived vertebrate bred in captivity it recapitulates typical age-dependent phenotypes and pathologies	role of mitochondria and oxidative phosphorylation in variation in the pace of life mitochondrial dysfunction ageing	Baris et al. [51] Chung et al. [52] Chung et al. [53] Chung et al. [54] Fangue et al. [55] McKenzie et al. [56] Whitehead et al. [57] Baumgart et al. [58] Harel et al. [59] Hartmann et al. [60]
class Mammalia
Peromyscus maniculatus (North American deer mouse)	inhabits high altitude has evolved an enhanced aerobic capacity in hypoxia	mitochondrial basis for adaptive variation in aerobic performance	Cheviron et al. [61] Cheviron et al. [62] Lui et al. [63] Mahalingam et al. [64] Nikel et al. [65] Scott et al. [66] Scott et al. [67]
Heterocephalus glaber (naked mole-rat)	high longevity cancer resistant rarely suffering from age-related diseases	ageing age-dependent structural and functional changes of mitochondria	Bakeeva et al. [68] Buffenstein [69] Dammann [70] Holtze et al. [71] Khayal et al. [72] Kim et al. [73] Yu et al. [74]
Myotis myotis Myotis brandtii Myotis lucifugus (bats)	the only mammals capable of powered flight, with an extremely high metabolic rate, but associated with extended longevity	longevity/ageing heteroplasmy challenging the free radical theory of ageing	Ball et al. [75] Brunet-Rossinni [76] Brunet-Rossinni & Wilkinson [77] Dammann [70] Huang et al. [78] Jebb et al. [79] Munshi-South & Wilkinson [80] Seim et al. [81]
class Aves
Eopsaltria australis (Eastern yellow robin)	low nuclear differentiation with two highly divergent, parapatric mitochondrial lineages with sharp climate-correlated differences in their distributions	climate-driven mitochondrial selection mitonuclear coevolution	Lamb et al. [82] Morales et al. [83] Morales et al. [84] Sunnucks et al. [85]
class Hexanauplia
Tigriopus californicus	populations with extreme mitochondrial DNA divergence	thermal tolerance and its relationship to mitochondrial functions mitonuclear coevolution	Barreto & Burton [86] Barreto et al. [87] Burton et al. [88] Edmands [89]
			Ellison & Burton [90] Ellison & Burton [91] Foley et al. [92] Harada et al. [93] Willett [94]
class Bivalvia
species with doubly uniparental inheritance (DUI) of mitochondria	natural and evolutionary stable mitochondrial heteroplasmy transmission of sperm mitochondria across generations the longest-lived animal to date has DUI (Arctica islandica: the specimen ‘Ming’ was 507 years old; [95]	mitochondrial biology and inheritance heteroplasmy challenging the free radical theory of ageing longevity/ageing mitonuclear interactions genomic conflicts	Bettinazzi et al. [96] Bettinazzi et al. [48] Capt et al. [97] Ghiselli et al. [98] Ghiselli et al. [99] Ghiselli et al. [100] Ghiselli et al. [101] Iannello et al. [102] Milani [32] Milani & Ghiselli [103] Milani et al. [104] Milani et al. [105] Skibinski et al. [106] Zouros [107]

(a). Killifish (Fundulus spp., Nothobranchius furzeri): environmental adaptation, ageing

Members of the genus Fundulus have been investigated for adaptation to different temperatures [51–53,55] and different salinity [57]. Differences within and between species (different populations of Fundulus heteroclitus, and Fundulus grandis) showed the involvement of mitochondria in thermal adaptation. Indeed, temperature changes affect metabolism and mitochondrial performance, playing a role in whole-animal thermal tolerance and plasticity. These studies support a general role for OXPHOS and mitochondrial oxygen kinetics in differentiating aerobic performance and in influencing species responses to environmental change [53]. Comparisons of mitochondrial responses to thermal changes in different tissues investigated using high-resolution respirometry in subspecies adapted to different temperatures documented an extensive plasticity in mitochondrial performance following thermal acclimation in killifish, with the extent of responses differing between tissues [52]. Also, lipid remodelling in mitochondrial membranes appears to be a mechanism contributing to these changes [54]. Fundulus heteroclitus is also studied for cytonuclear disequilibrium that appears to act in reproductive isolation or selection against hybrids [56]. The species is abundant in estuaries and it shows adaptations to different water salinity. Comparative experiments were performed to track the transcriptomic and physiological responses to salinity variation, and to explore the regulatory mechanisms that may enable osmotic acclimation. Some genes showed population- and salinity-dependent patterns of expression during acclimation, including genes involved in mitochondrial function [57].

The killifish Nothobranchius furzeri is becoming a model for the study of senescence, being naturally short-lived [60]. Shorter- and longer-lived individuals of the species differ in gene expression, especially in early life. Mitochondrial complex I expression resulted to be negatively correlated with lifespan, with the phenotype reversed by complex I inhibitors (reversion of ageing-related regulation of gene expression and extended lifespan) [58]. An integrative genomic and genome-editing toolkit using the de novo-assembled genome and the CRISPR/Cas9 technology was developed for this species. This approach will be important to target candidate genes related to ageing [108], including mitochondrial dysfunction [59]. Ageing in N. furzeri appears associated with a decline in mtDNA copy number, downregulation of mtDNA-associated genes and an impairment of mitochondrial function [60]. This species appears to be an ideal model to assess the role of physiological and environmental parameters on ageing and lifespan determination.

(b). Deer mice (Peromyscus spp.): environmental adaptation

At high altitude, small mammals have to support thermogenesis and aerobic exercise in spite of the reduced oxygen availability. To address this topic, many studies have been performed involving highland and lowland Peromyscus maniculatus [63,65]. The integration of data from different approaches, such as measures of whole-organism thermogenic performance, measures of metabolic enzyme activities and genomic/transcriptional profiles, was used to examine the adaptive variation of this complex trait in deer mice that are native to different elevations. Different species of the same genus have been compared for such research (P. maniculatus and Peromyscus leucopus) [61]. Highland deer mice have an enhanced thermogenic capacity under hypoxia compared with lowland conspecifics and closely related lowland species. This feature is largely owing to an increased capacity to oxidize lipids associated with elevated activities of muscle metabolic enzymes that influence flux through fatty-acid oxidation and OXPHOS pathways in high-altitude deer mice, and by concomitant changes in the expression [61]. Other than having higher respiratory capacities in high-altitude mice than in low-altitude mice, the former also showed higher mitochondrial volume densities [64]. The association between transcriptomic profiles and muscle phenotypes was also tested: several genes involved in energy metabolism were more expressed in highlanders, and the regulators of mitochondrial biogenesis were positively correlated with muscle oxidative phenotype [66]. In species that are distributed across steep environmental gradients, adaptive variation in physiological performance may be also owing to transcriptional plasticity in underlying regulatory networks [62].

(c). Naked mole-rats (Heterocephalus glaber): ageing

Naked mole-rats (Heterocephalus glaber) are the longest-living (approx. 30 years) rodents known to date [69], thus being an important model to examine mechanisms modulating ageing. Naked mole-rats show negligible senescence, no age-related increase in mortality and high fecundity until death [73]. In addition, they are resistant to cancer—both spontaneous and experimentally induced—thus challenging the theories that link ageing, cancer and redox homeostasis. Indeed, although characterized by significant oxidative stress—young individuals surprisingly have high levels of accrued oxidative damage [69]—the naked mole-rat proteome does not show age-related susceptibility to oxidative damage nor increased ubiquitination [73]. Traditionally, the main mammalian models used for the study of ageing have been mice and rats, i.e. short-lived species [70]. To reach a reliable generalization and to deal with life-extending mechanisms, a long-lived species such as H. glaber can be precious; accordingly, in the past two decades, long-lived mammals (i.e. mole-rats and bats) became a matter of investigation. These new models drew attention to the oversimplification problem affecting existing theories about ageing in Chiroptera, and to several features considered universal components of enhanced longevity in mammals [70]. Indeed, among mammals with high longevity considerable variation exists with respect to candidate regulatory mechanisms [70].

Mitochondrial morphology and abundance were investigated in naked mole-rats, leading to interesting findings. In contrast to mitochondria of other mammalian cardiomyocytes, where the internal space is filled with parallel rows of cristae, in naked mole-rats they have a chaotic pattern of wave-like cristae. In addition, there are some large mitochondria exceeding normal sizes by two to three times, with cristae assembled into groups with curved and ring-like structures [68]. Moreover, in these animals, mitochondrial number and size increase with age. They also possess a neotenic type of chondriome accompanied by specific features of OXPHOS, and a significant decrease in the level of matrix adenine nucleotides [71].

RNA sequencing was used to compare liver gene expression profiles between naked mole-rats and wild-derived mice. Genes associated with redox processes and mitochondria were expressed at higher relative levels in naked mole-rats. Also, a protease inhibitor and a mitochondrial complex II subunit, both ageing-related genes, were found strongly over-expressed in the naked mole-rat, suggesting alterations in mitochondrial and oxidation-reduction pathways in the species [74]. A combination of bioinformatic algorithms with nucleotide genomic signal processing and hierarchical cluster methods were also used to investigate differential phylogenetically convergent genetic traits related to senescence of the long-lived naked mole-rat and various species. Two ageing-related mitochondrial genes were selected based on the classification of ageing-related genes in the Human Ageing Genomic Resources database. Cytochrome b and cytochrome c oxidase subunit I are both related to neurodegenerative disease [72].

(d). Bats (Myotis spp.): heteroplasmy, longevity and ageing

The ‘free radical theory of ageing’ [109] posits that a high metabolic rate causes mitochondrial heteroplasmy and progressive ageing (but see [103]).The extended longevity of bats, despite their high metabolic rate (they are the only mammals with the ability of powered flight), may provide insights into mechanisms of ageing. Predictions of the free-radical theory of ageing as an explanation for differences in lifespan were tested in Myotis lucifugus (little brown bat), Blarina brevicauda (northern short-tailed shrew) and P. leucopus (white-footed mouse)—maximum lifespan potential: 34, 2, 8 years, respectively—by comparing whole-organism oxygen consumption, hydrogen peroxide production and superoxide dismutase activity in several tissues. Mitochondria from M. lucifugus produced half to one-third the amount of hydrogen peroxide per unit of oxygen consumed compared to mitochondria from B. brevicauda and P. leucopus, respectively. These results are similar to those reported for birds that, like bats, show high metabolic rates and longevity. These results seem to provide support, at least partially, for the free radical theory of ageing as an explanation to longevity of bats [76]. A series of genome-wide comparative analyses between bat (Myotis myotis) and non-bat mammals was performed to investigate the molecular mechanisms underlying longevity. Bat-specific and differentially transcribed micro RNA and messenger RNA that function in key longevity pathways were detected. As suggested for M. myotis, bats may possess unique regulatory mechanisms for resisting tumorigenesis, repairing cellular damage and preventing oxidative stress, all of which probably contribute to their extraordinary lifespan [78].

In this connection, analyses were done to see if bats exhibit increased mitochondrial heteroplasmy with age. In M. myotis, the majority of heteroplasmy was at a low frequency, and nucleotide changes consisted of transitions. Oxidative mutations were present in only a small number of individuals, there was no significant increase in heteroplasmy with age, and heteroplasmy was found to be dynamic (from recaptured individuals), without a uniform increase over time. All that considered, the data acquired for M. myotis appeared not to follow the predicted increase in heteroplasmy as posited by the free-radical theory of ageing, questioning its general validity [79]. On average, bats and birds live substantially longer than non-flying mammals of similar body size. The combination of small body size, high metabolic rates and long lifespan they share does not seem to support oxidative theories of ageing. However, large-scale comparative analyses on a few emerging model species have identified several mechanisms for resisting oxidative damage in mtDNA and cellular structures in both bats and birds [80].

(e). Eastern yellow robin (Eopsaltria australis): environmental adaptation, mitonuclear coevolution

Diversifying selection on metabolic pathways can reduce intraspecific gene flow and promote population divergence. An opportunity to explore this arises from mitonuclear discordance observed in an Australian bird Eopsaltria australis. Across 1500+ km, nuclear differentiation is low, whereas two highly divergent, parapatric mitochondrial lineages show a discordant longitudinal geographical pattern and experience different climates. Eopsaltria australis is one of several animals that show functional mitonuclear interactions despite discordance between mitochondrial and nuclear genomes. The two mitolineages show sharp climate-correlated differences in their distributions, suggesting that the mitochondrial introgression and divergence were driven by natural selection [85]. Evidence of positive selection has been found in genes of OXPHOS complexes, predicted to cause differences in electrostatic subunit-subunit interactions influencing coupling efficiency of the complex [83,85]. This selection may reflect local environmental adaptation, a by-product of other selective processes, or genetic incompatibilities [83].

Comparison of fixation indexes in the nuclear genomes between Eastern yellow robin populations across their biogeographic range revealed the existence of two genomic islands of divergence against a background of low differentiation [85]. A strong genetic differentiation and sequence divergence was detected in a region of chromosome 1A mirroring the geographical pattern of mtDNA divergence. Such a region is enriched for genes performing mitochondrial functions. Molecular signatures of selective sweeps in this region alongside those in the mitochondrial genome suggest a history of adaptive mitonuclear co-introgression, consistent with mitonuclear coevolution as an important mechanism for population divergence and local adaptation [84]. Studies on a large set of Australian birds were performed to investigate whether Pleistocene climate changes drove mitochondrial selection and evolution. Climate was a significant predictor of mitochondrial variation in eight out of 17 species [82].

(f). Tigriopus californicus: environmental adaptation, mitonuclear coevolution

The copepod Tigriopus californicus shows a very rapid mtDNA sequence evolution [94], reaching extreme mtDNA divergence between populations with exceptionally high levels of amino acid differentiation compared with other taxa [89,88]. Hybridization between allopatric populations expose functional incompatibilities between genomes that have not coevolved and places intrinsic selective pressures on many nuclear genes involved in mitochondrial functions [86]. Ellison & Burton [90] measured fitness, mitochondrial function and enzyme activity in hybrid lines of T. californicus, and showed that only the OXPHOS complexes composed of both nuclear and mitochondrial subunits suffered a loss of activity. Moreover, they observed a positive correlation between OXPHOS enzyme activity and mitochondrial function, and between mitochondrial function and fitness. Moreover, the analysis of transcript levels of genes involved in the OXPHOS pathway for a series of parental and inbred hybrid lines showed that only genes located on the mtDNA differed among lines. Lines bearing certain genotypic combinations of mtDNA-mitochondrial RNA polymerase showed a diminished capacity to upregulate mitochondrial genes in response to hypo-osmotic stress. This suggests that disruption of coadaptation between nuclear and mitochondrial genes contributes to the phenomenon of hybrid breakdown [90,91,110]. Tigriopus californicus nuclear genome shows elevated protein evolutionary rates and putative positive selection in genes predicted to interact with mtDNA, mtRNAs and the corresponding proteins encoded. Thus, the rapid mitochondrial evolution appears to drive compensatory nuclear evolution within populations, thereby providing a potentially important mechanism for reproductive isolation [87].

As reported above for certain non-model vertebrates, variation in thermal tolerance plays a key role in determining the biogeographic distribution of organisms, and variation in mitochondrial function probably contributes to variation in thermal tolerance. Populations of T. californicus were also studied in this concern, showing a latitudinal thermal gradient along the coast of California (USA), suggesting thermal adaptation. The obtained data supported population-specific rates of ATP synthesis at chronic temperatures with a role for mitochondria in setting thermal range limits, and indicate that divergence in mitochondrial function is probably a component of adaptation across latitudinal thermal gradients [93].

Tigriopus californicus differs from standard animal models used in genetics because it lacks heteromorphic chromosomes but has a polygenic sex-determination mode. Reciprocal F₂ hybrids between two strongly differentiated populations were used for quantitative trait loci mapping. Although many studies on standard speciation models found the strongest genetic incompatibilities to be nuclear-nuclear (i.e.: X chromosome-autosome), in T. californicus the strongest deleterious interaction was mito-nuclear. This system thus is important to study the genetics of reproductive isolation in comparison to standard model systems [92].

(g). Bivalve molluscs: heteroplasmy, mitochondrial inheritance, mitonuclear interactions and genomic conflicts

More than a hundred species of bivalve molluscs are known for their unique mode of mitochondrial transmission, known since 1994 as doubly uniparental inheritance (DUI) (see [107] for a review). DUI species present two mitochondrial lineages, one transmitted by females (F-type), and one by males (M-type), this is achieved by having homoplasmic gametes: eggs for the F-type, spermatozoa for the M-type. Adult animals are heteroplasmic to various degrees, with interspecific mtDNA nucleotide divergence reaching exceptionally high values ([101] and references therein). The unusual mechanism of inheritance and the natural heteroplasmy provide an interesting point of view about mitochondrial biology in general. For example, the ‘division of labour hypothesis’ [111] posits that because mtDNA is maternally inherited, female gametes would prevent oxidative stress and mtDNA damage by repressing OXPHOS, thus being quiescent genetic templates. Interestingly, DUI species have swimming spermatozoa transmitting their mtDNA for hundreds of million years, with no apparent consequence for species survival [103,100]. The presence of stable and natural heteroplasmy of highly divergent mtDNAs makes DUI animals ideal systems to investigate genomic conflicts [104,105], and how an individual, a tissue, a cell, an organelle can deal with multiple mitochondrial variants [101]. So far, the DUI system was investigated mainly at the mtDNA level [107], and with RNA-Seq [98,99,97,102], even if more recently mitochondrial phenotypes of different kinds has been analysed. For example, differences between F- and M-types in OXPHOS activity—inferred by bioinformatics tools [106], or by direct test on oxygen consumption and bioenergetics [96,48]—were recently examined. Such divergent phenotypes may be produced by amino acid substitutions at specific sites of OXPHOS subunits differentiating the F-type from the M-type [106]. According to these new findings, the membrane potential may be involved in mitochondrial segregation into the germline [32], and/or with sperm-specific adaptations [99,96,48]. These studies evaluated mitochondrial functions associated with sex-linked mtDNAs in DUI species—among which Arctica islandica, the oldest animal known to date [95]—, compared DUI species with bivalve species showing the common strict maternal inheritance of mitochondria, and compared homoplasmic with heteroplasmic tissues. The results suggest that the inheritance of sex-linked mitochondria not only could better fulfil different energetic requirements of sperm and eggs, but could also be associated with success in facing heteroplasmy and ageing [96,48,99].

5. The comparative method in the ‘-omics’ era: future perspectives

In the last 15 years, biology has changed radically. New technologies have reshaped the way data are collected and, most importantly, the amount of data that it is possible to obtain at an increasingly lower cost. Above all, high-throughput sequencing (HTS) granted virtually all laboratories around the world access to volumes of data unthinkable before. As a result, ‘big data’ became the foundation of most of the research in life sciences, and ‘data mining’ became one of the most used methods. While such change represents a breakthrough innovation that has greatly improved the potential of solving increasingly complex problems, it also has downsides. The deluge of data generated gave rise to what Sydney Brenner called ‘factory science’ in his 2007 Vanderbilt Discovery Lecture entitled ‘The Next 100 Years of Biology’ (video available at: http://hdl.handle.net/1803/1088), citing his words: ‘I like to call it low-input, high-throughput, no-output biology’. This departure from hypothesis-driven research towards model-independent data mining was prophetically discussed in two papers by John Allen in 2001 [112,113]. Allen used Karl Popper's metaphor of ‘the bucket and the searchlight’ [114] indicating two approaches to knowledge: the big-data approach, represented by the bucket, as opposed to the hypothesis-driven research, represented by the searchlight. It is true that HTS has made more systems accessible, but maybe it has unified the approach, extinguishing other methodological approaches in its wake. Are we moving from ‘a few model systems’ to ’a few model technologies’, and if so, which bottleneck is more consequential? We do not wish to demonize HTS and new technologies which will probably be the pillars of new discoveries in the next few years. That said, we need to use such augmented technical power wisely.

HTS has surely made research on non-model species more accessible, but there are still limitations, and often the optimism about the revealing potential of genomics/transcriptomics data is excessive. We briefly list the major difficulties the researchers have to face when applying the comparative method.

(a). Sequencing disparity

New sequencing technologies were supposed to facilitate the study of organismal biodiversity, by making non-model species more accessible. This is not happening. David et al. [115] analysed species representation of nonhuman eukaryotes in the NCBI Sequence Read Archive, and found out that while species richness has been increasing, species evenness is decreasing. They reported a growing bias in favour of a minority of species: the top 1% most studied organisms represent a larger proportion of the experiments. In the future, it will be necessary to reverse such a trend, trying to explore diversity much more extensively. Initiatives such as the Global Invertebrate Genomics Alliance (GIGA) aims to promote standards that facilitate comparative approaches to genomics and collaborations across the scientific community [116]. It might sound ridiculous to split animals into vertebrates (including a subphylum of Chordata) and invertebrates (including all the other animal phyla), but given the disparity of genomics resources available—mostly vertebrates—it has a practical use. During the Biogenomics 2017 Conference in Washington DC the Earth Biogenome Project was announced, which aims to sequence, catalogue, and characterize the genomes of all eukaryotic biodiversity over a period of 10 years [117]. This project might seem overly ambitious, but 10 years is a long time for science and technology, and this impressive collaborative effort will probably reduce sequence disparity and enhance our ability to do comparative analyses.

(b). What to compare and how?

When choosing a model or extrapolating findings to other organisms, it is extremely important to know the phylogenetic context [5,17]. Not only knowing the phylogenetic relationships among organisms provides information about relatedness and divergence time, but it is fundamental to assess character evolution, polarity and homology: all essential data for the comparative method. For most of the animal taxa, there is a lack of phylogenetic framework—even for taxa closely related with model organisms—and this results in a problematic assessment of biodiversity and evolutionary relationships. Not being able to reconstruct the evolutionary history of the organisms undermines the comparative method.

Another fundamental resource we need when performing comparative analyses is gene annotation, which is the assignment of an identity and a function. Annotation works by translating the information available for model or well-known species, to the species subject of study [118]. Such translation is mostly based on sequence similarity—with all the obvious limitations—and nowadays, it is usually performed through fully automated procedures. The massive amount of genomics data generated and the consequent lack of human curation is causing a decrease in annotation quality [119]. Annotation is a difficult and time-consuming task, but unfortunately, the reward for such effort is almost non-existent: it would be difficult to publish a curated genome annotation in a relevant journal. In general, producing data are much more rewarding than curating them; in Sydney Brenner's words: ‘We presently have huge amounts of data that are largely unassimilated. I have discovered that while there are rewards for collecting data and distributing data, there is little, if any, support for organizing it’ [120, p. 9].

Strictly related with annotation, there is another big challenge—perhaps the biggest one—for comparative biologists, that is the concept of homology. Homology refers to the historical continuity of characters and it is a fundamental concept in biology, nevertheless, it is ‘conceptually highly elusive’ [121]. The main difficulty with homology is basically the topic of this Issue: the loose connection between genotype and phenotype. To make meaningful evolutionary studies, we need to compare genes that share ancestry and that retained the same function through evolutionary time. The so-called ‘ortholog conjecture’ [122] states that orthologues—homologous genes originated by a speciation event—retain the same function in the course of evolution, while paralogues—homologous genes originated by a duplication event—tend to evolve new functions. Currently, the typical procedure in comparative analyses is strictly based on the ‘ortholog conjecture’, and consists in comparing putatively orthologous sequences identified by sequence similarity [123]. As usual, the reality is much more complicated. The multi-domain architecture of most eukaryotic proteins, gene duplication, recombination events, exon shuffling, gene fusion and fission are only some of the several processes that go under the spot-on definition of ‘molecular tinkering’ [124], which together with hybridization and introgression make the notion of homology at the molecular level an ‘elusive concept’ (for a thorough discussion on homology and related concepts, see [125]). A possible solution to this problem is to compare gene regulatory networks instead of individual genes. Such ‘character identity networks' as defined by Günter Wagner [121] would enable more meaningful comparisons.

(c). The importance of ‘basic research’

The representativeness of a model cannot be established in advance. Steel [126, p. 4] defined the ‘extrapolator's circle’, a paradox of model organisms and comparative research: ‘establishing the suitability of the model would require already possessing detailed knowledge of the causal relationship in the target, in which case extrapolation would be unnecessary’. The choice of models is an iterative process: we make starting assumptions, test them against our knowledge, and then adjust as necessary, given our new discoveries [5]. The key aspect of a model organism is whether a community of scientists is working with it. Implementing the comparative method using as many models as possible basically means doing ‘curiosity-driven’ science (also known as basic research). Right now, basic research is underfunded, and mostly overlooked. We hope that, albeit more than 90 years later, an increasing number of biologists will start following the words of, among others, August Krogh and Hans Krebs. That would bring more balance between the currently predominant mechanistic/reductionist approach and the comparative approach.

Data accessibility

This article has no additional data.

Authors' contributions

F.G. and L.M. contributed equally to this work.

Competing interests

We declare we have no competing interests.

Funding

This work was supported by the Italian Ministry of Education, University and Research (MIUR) SIR Programme grant no. RBSI14G0P5 funded to L.M., MIUR FIR2013 Programme grant no. RBFR13T97A and the Canziani bequest funded to F.G., and ‘Ricerca Fondamentale Orientata’ (RFO) funding from the University of Bologna to L.M. and F.G.