Genetics without genes? The centrality of genetic markers in livestock genetics and genomics

2.1 Quantitative genetics

Quantitative genetics traces its roots to the early-twentieth-century work of Ronald Aylmer Fisher (1890–1962), John Burdon Sanderson Haldane (1892–1964) and Sewall Wright (1889–1988), each of whom contributed towards the reconciliation of Mendelian genetics and biometry (Provine 1971). The Mendelian conception of heredity centred on the transmission of discrete units—genes—that could be discerned in the genotype through observed differences in the phenotype (Johannsen 1909). Biometry focused instead on continuous variation, and statistical analysis of variants of small-effect. Though there were important differences between their approaches, Fisher, Haldane and Wright’s fundamental insight was that continuous variation could be explained with reference to a multitude of Mendelian factors, each of which may only contribute a small part to the overall observed variance. This was used to explain how variation is conserved across generations, and helped to inaugurate population genetics (Provine 1971).

Wright had an interest in animal breeding; he had worked at the United States Department of Agriculture (USDA) facility at Beltsville, Maryland from 1915 to 1925. While there, he worked with livestock breeders, responded to queries from farmers, worked on problems concerning quantitative analysis posed by other researchers at the facility and devised an ancestry-based evaluation scheme to help calculate the market value of quarantined cattle (Russell 1989, pp. 5–6).

Jay Lush (1896–1982), who had worked in animal breeding at the University of Texas and was based at Iowa State College from 1930, attended Wright’s course on statistical genetics at the University of Chicago in 1931. He also absorbed Fisher’s insights when the latter lectured at Iowa State over the summers of 1931–1936. Lush was concerned almost wholly with developing the quantitative basis for informing the selection of livestock animals for breeding, to improve upon mass selection based largely on phenotypes (Hill 2014). He published voluminously, included the highly influential Animal Breeding Plans (in successive editions published in 1937, 1943 and 1945), and was recognised for his teaching and training of many future leaders of livestock genetics. The programme of research begun in Iowa soon became established in other centres worldwide.

One of the key underpinnings of quantitative genetics as applied to livestock breeding is the infinitesimal model, which was originally formulated by Fisher in 1918.⁵ Simplified, the infinitesimal model posits “an effectively infinite number of unlinked loci with infinitesimal effects” (Bulmer 1980, p. 150). The power of the model does not derive from any congruence to a supposed underlying reality, but because “it has very powerful simplifying statistical properties and avoids the need to specify individual gene effects, information on which has until recently been impossible to obtain” (Hill 2014, p. 4).

Quantitative genetic approaches to livestock breeding have been produced by researchers attentive to the needs and problems of breeders: namely, the creation of methods to index animals, calculate accurate Estimated Breeding Values and conduct selection experiments to test theory.

2.2 Classical and molecular approaches to genetics

Apart from population and quantitative genetics, a distinct Mendelian tradition remained, however, instantiated in the ‘classical genetics’ founded in the work of Thomas Hunt Morgan’s laboratory at Columbia University in New York from the 1910s to the late-1920s. Working with the fruit flies Drosophila melanogaster, researchers in this laboratory identified particular mutants: flies with different characteristics to the presumed norm or ‘wild type’, for example different eye colours or wing morphologies.⁶ They presumed that genic differences underlay these various mutations. In millions of flies, they measured the frequencies with which particular mutations manifested together. The laboratory then used the data on these associations between mutations—the linkages—to map the genes, which were presumed to be stable entities that were transmitted from one generation to the next (Waters 1994).⁷

This linkage mapping is based on the fact that during meiosis in the production of gametes (sex cells: sperm and ova), parts of chromosomes break and then recombine. Sometimes when they do this, part of one pair of chromosomes (the part being a chromatid) can become part of the other pair. As any two loci (specific places on a chromosome) that are closer together are more likely to remain on the same chromatid than two loci that are further apart, measurements of the co-location of loci can be used to conduct a linkage analysis that uses the presence or absence of particular loci to estimate the relative positions of the loci on the chromosome (Kohler 1994).

From the 1950s, the determination of the structure of DNA helped to establish it as the material basis of genes. From this foundation, research established how DNA is transcribed and translated to produce particular proteins with specific amino acid compositions, with the aid of intermediary molecules such as messenger RNA and transfer RNA. On the basis of this work, the gene was reconceptualised from the classical conception of a unit of distant control over some aspect of the phenotype, to a conception in which the gene consists of information for the precise specification of the structure of a protein (Morange 2000).⁸

While further research into the nature of the ‘molecular gene’ and its relationship to observable structure and function rapidly complicated simple models of gene expression and function, for instance through the recognition of complex webs of gene regulation and interaction (Rheinberger and Müller-Wille 2017), the molecular approach to genetics still focused on the identification and characterisation of genes. Whereas classical genetics inferred the presence of genes and different versions of them (alleles) from the existence of phenotypic variants, variants of molecular genes were associated with variation in the structure and function of macromolecules, which provide links from genetic variation to observable phenotypic variation.

This approach held the promise of prospective intervention in biological processes and mechanisms to specifically alter phenotypic outcomes. This promoted the search for genes (and variants and mutants thereof) that exhibit relatively large effects.

From the 1950s, molecular biologists and biochemists had furnished an increasing number of different kinds of biomarker, from which the presence or absence of particular genes (and alleles thereof) could be inferred. Many of these were assayed in the blood, and were presumed to be macromolecular products of these genes. Linkage could be inferred by the combinations of presence or absence of particular biomarkers, and perhaps also phenotypic indicators, as in the origins of classical genetics.

In the context of medical genetics, identifying and mapping the actual gene presumed to play a role in a physiological or pathological process of interest was of paramount importance. This still required markers to be identified, however. Research on the Human Leucocyte Antigen (HLA) system is an instructive example. The genes in this system, implicated in immune response among other functions, are densely packed and highly polymorphic. As antibodies could be used to assay the presence or absence of particular antigenic molecules coded for by specific variants of HLA genes, this region was ripe for mapping genes using the antigens as markers. Mapping of disease-related genes proceeded into the 1980s through the development of genetic markers linked to prospective genes (Harper 2008, pp. 200–211). Some of the techniques by which these markers were produced are discussed in the next section.

In the context of livestock genetics, the inference that, for a given trait of interest, a gene (or version thereof) must exist in a certain part of a given chromosome, need not culminate in the precise identification and localisation of that gene. As a therapeutic or corrective intervention into the individual animal is not necessarily the aim, it is sufficient to identify and map the genetic loci of interest, to be able to confirm that particular markers were sufficiently close to the locus of interest to serve as a proxy, so that a test could be developed to detect the marker when genotyping animals. This relied on the classical genetical insights into linkage and recombination.

2.3 Investigating halothane sensitivity: towards a candidate gene strategy

This was the approach taken by the geneticists who worked on the Halothane sensitivity locus in the 1980s. Halothane is a veterinary anaesthetic that had been identified as a test for Porcine Stress Syndrome (PSS). Pigs affected by PSS died suddenly when exposed to stress, and the meat from these pigs was of poor quality, and therefore unappealing to consumers. Prior to the full elucidation of the genetic basis of the syndrome, attempts were made to identify pre-symptomatic pigs by their response to exposure to halothane, responders becoming rigid within 3 min (Webb 1980). However, as the condition had been theorised to be recessive, requiring two copies of the version (allele) of the gene involved (Ollivier et al. 1975), only homozygous pigs carrying two copies would be symptomatic, and thus exposed by this test. Heterozygous pigs who carry one copy of the halothane-sensitivity allele would be undetected and therefore would not necessarily be excluded from breeding, enabling the lesion to be transmitted to the next generation.

This, combined with the time-consuming nature of testing herds in this way, meant that alternative tests to identify carriers as well as symptomatic pigs were sought. As with the human medical geneticists who in the mid-1980s attempted to identify and map disease genes (Harper 2008; Lindee 2005), it became a pressing matter for the livestock and breeding industries to identify, localise and characterise the locus or loci associated with PSS.⁹ Then, using linked markers, animals could be genotyped with a view to removing the rogue allele from breeding populations.

The relationship of this locus with nearby loci was therefore of interest to geneticists. In the UK, Alan Archibald of the Animal Breeding Research Organisation (ABRO; one of the main institutional antecedents of the Roslin Institute: see García-Sancho 2015; Myelnikov 2017; Lowe, in review) obtained a research commission from the Ministry of Agriculture, Fisheries and Food to investigate the genetics of PSS, and in particular the group of linked loci it was deemed to be associated with. In a volume of the journal Animal Genetics edited by Archibald and Pat Imlah (of the Royal Dick School of Veterinary Studies, University of Edinburgh), papers included research on the order of the linked loci, the use of them in genotyping animals, and rates of recombination between the loci (Archibald and Imlah 1985). Data on recombination events between loci supported a consensus order of the loci, but the identity of the halothane locus was still unknown.

By 1986, Archibald recognised the potential importance of genetic markers, markers representing some section of DNA rather than a molecule circulating in the blood. He cited the use of Restriction Fragment Length Polymorphisms (RFLPs) to isolate and characterise a gene implicated in Duchenne muscular dystrophy (Archibald 1986). RFLPs were developed for use in human genetics. Linkage analyses require detectable markers of polymorphisms or discrete variants, which had been in short supply beyond serological tests for genes in the dense and polymorphic HLA region. RFLPs use restriction enzymes to cut sections of DNA; the resulting fragments are then separated by length using gel electrophoresis. The principle behind RFLP techniques is that variation or polymorphisms result in cuts being made at different places, and therefore the DNA fragments will end up in different places in the gel; on the membrane to which it is transferred; and in the resulting autoradiograph. The basis for using RFLPs was established in 1974 and 1975. In 1980, the basis for using RFLPs to construct a human genetic linkage map was outlined (Botstein et al. 1980).

From the very beginning, it was recognised that RFLPs need not include a gene. If an RFLP was linked to a gene or genes, however, this created the prospect that genes may be mapped—or tracked—without isolating or characterising the gene itself (Botstein et al. 1980, p. 317). In 1988, a gene linked to the halothane locus was mapped to a particular location on swine chromosome 6 using cDNA (complementary DNA) probes obtained from a pig genomic library (Davies et al. 1988). Following this and research on malignant hyperthermia, the human manifestation of the trait associated with the halothane locus in pigs, a candidate gene was proposed. In 1990, the malignant hyperthermia locus was mapped to a location on human chromosome 19 by inference from the linkage group established by the 1988 research on pigs to the corresponding area of the human genome. This location coincided with the position where the ryanodine receptor gene (RYR) had recently been localised (McCarthy et al. 1990).

In the same issue of Nature where this research appeared, David MacLennan and colleagues reported the results of a linkage analysis that they had conducted on families containing individuals deemed to be at risk from malignant hyperthermia. They analysed seven loci, including RYR, using seven different cDNA probes, and applied ten different RFLP enzymes to chromosome 19 DNA obtained from the family members. When they analysed the results in terms of recombination (and, by implication, co-segregation) frequencies, they found strong linkage between the malignant hyperthermia trait and RYR. This, combined with the other evidence already indicated, led them to conclude that RYR must be considered to be a candidate gene implicated in malignant hyperthermia in humans, and PSS in pigs (MacLennan et al. 1990).

It proved to be so. With the gene, and the particular mutation (RYR1) responsible for PSS identified (Fujii et al. 1991), genotyping tests could be developed to detect it in individuals (Dalens and Runavot 1993).¹⁰ This offered breeders the opportunity to manage the responsible gene, and potentially to eradicate it (Fujii et al. 1991; O’Brien and Ball 2013; Otsu et al. 1991).

Within the space of a decade, linkage mapping had progressed to the identification and characterisation of a single gene responsible for a condition that was causing considerable losses to the livestock industry. This provided the impetus for two linked strategies. One was the identification of candidate genes related to production traits of interest, and the development methods to genotype them. The other was a more comprehensive identification and mapping of markers in the genome, deemed important in part to aid in the mapping of quantitative trait loci (QTL), locations in the genome associated with variation in traits of interest. Animals could then be genotyped using nearby markers, and therefore the livestock geneticists did not need to methodologically orient themselves towards genes.

The latter strategy underpinned support for major projects to populate pig chromosomes with maps of ordered and, in some cases, physically localised or anchored markers, as will be discussed in the next section. Major projects based on sets of reference families included PiGMaP, funded by the European Commission (1991–1996), one led by the United States Department of Agriculture Agricultural Research Service Meat Animal Research Center (USDA-ARS MARC), and a Nordic collaboration.

3.1 RFLPs and other techniques for mapping markers

PiGMaP began with two parallel mapping approaches. The first was to detect RFLP loci using homologous (mainly cDNA) probes derived from pigs and heterologous (again mainly cDNA) probes derived from humans and other non-porcine mammals. This work was done on crosses of distinct pig breeds with known pedigrees. Crossing distinct breeds helped to generate polymorphisms that could be used in linkage analyses (Archibald et al. 1995). The second approach was physical mapping, the assignment of genes to chromosomes using techniques of in situ hybridisation, first using radioactive probes, then increasingly moving to Fluorescent In Situ Hybridisation (FISH). The technical challenge presented by the development and mapping of markers meant, according to Chris Haley, a leading researcher and organiser in PiGMaP, that “there was benefit in different groups developing different markers and validating them using both physical and linkage mapping approaches.”¹¹

Increasingly, the number of markers used in linkage mapping (also known as genetic mapping) became dominated by hypervariable markers: first minisatellites, and then microsatellites. These are based on areas of the genome featuring repetitive sequences of DNA: particular patterns or ‘motifs’ that are repeated a number of times over. They are not genes. The difference between minisatellites and microsatellites is in the length of the motifs in base pairs, typically the former from ten to dozens of base pairs, the latter ranging up to ten. There are many more microsatellites than minisatellites in the genome, and microsatellites are more widely distributed across chromosomes. Consequently, microsatellites came to be of more importance to mapping and other applications.

The different number of repeats contained in microsatellites is valuable for linkage analysis due to the large number of alleles it may exhibit as a result, leading often to differences between individuals and heterozygosity within an individual (Gulcher 2012). This variation leads to these markers having a high polymorphic information content. It was in 1989 that the potential use of “simple sequence length polymorphisms” as polymorphic markers was recognised (Tautz 1989), and that the relatively new Polymerase Chain Reaction (PCR) could be used to amplify these variable repeats, using specially-designed custom primers (Weber and May 1989). This could then be used as a basis for linkage analysis that would be less complicated than the RFLP approach, involving fewer stages, and be faster, cheaper and more sensitive (Weber 1990).

Microsatellites, as previously noted, are found all over the genome. They disproportionately tend to be in noncoding regions, however, though the exact patterns of distribution beyond that varies among different kinds of noncoding region, and across different taxa (Tóth et al. 2000). There is, therefore, no presumed functional correlation between the presence, absence or variant of a microsatellite and any variation in a phenotypic trait.

There is a striking contrast here with human genetics, where in the context of searching for disease-related genes, the shift from using RFLPs and microsatellites towards identifying SNPs was associated with a switch in focus from family-based genetic linkage research towards studying populations comprised of unrelated people (Rajagopalan and Fujimura 2018). Mapping markers in the service of identifying genes, a functional interest, therefore implied that the use of different kinds of markers were associated with fundamental changes in practice. In the case of pig genomics, the common project of the quantitative and molecular geneticists enabled these different kinds of markers to instead work for and with the mapping practices that they wanted to pursue, rather than the material or technical nature of the marker directing efforts towards one particular approach.

As genetic and physical mapping continued into the mid-1990s, the methodology and tools to identify and map QTL were also developed. The existence of QTL are posited by quantitative genetic theory, with a multitude of loci expected to each contribute a small degree to the overall variance in a trait. In the second iteration of the PiGMaP project (1994–1996), five different centres across Europe conducted genotyping using panels of 100 microsatellite markers derived from those mapped in PiGMaP and at USDA-ARS MARC. They also recorded phenotypic data on 2945 animals, each with known pedigrees from the third generation of reference populations. Several QTL bearing on livestock production traits, such as growth, fatness and meat quality, were detected as a result.¹² Chris Haley has observed that “building maps using different reference populations allowed a unified map to be produced and the homogeneity of that map between populations to be confirmed.”¹³

Work to identify genes continued parallel to this. This included some groups who were involved in the more marker-centric work. Genes were mapped by a variety of methods, including linkage analysis, in situ hybridization and somatic cell hybrids (which will be discussed below). Candidate genes could be inferred in much the same way as RYR was for malignant hyperthermia; an alternative strategy was to discern ‘positional candidate genes’ through a presumed association with a known QTL. The existence of a gene could be inferred by the presence of a nearby QTL that had exhibited some association with variation in a given phenotypic metric. In some cases, this resulted in the identification of genes and mutated versions of them that were causally associated with observed phenotypic variations, for example IGF2 (insulin-like growth factor 2; Jeon et al. 1999; Nezer et al. 1999), which is associated with leanness, and ESR (estrogen receptor gene; Legault et al. 1996), associated with increased litter size.

The expectation that this positional candidate gene strategy would lead to a detailed mechanistic understanding of the genes and pathways involved in various traits, and that this would form the basis of genetic change in livestock, did not, however, come to pass. In academia, Max Rothschild was a prominent exponent of this strategy, and was involved in discovering genes, including ESR, which formed the basis of patents exclusively licenced to a pig breeding company, the Pig Improvement Company (PIC; Rothschild and Plastow 2002). By 2007, having commented on the range of QTL so far identified, he and his co-authors confided that, “[p]erhaps more interesting is the fact that only a limited number of these QTL have been further investigated to the point that a known causative mutation has been implicated or proven” (Rothschild et al. 2007).

The ESR research and application thereof was actually an example of Marker-Assisted Selection (MAS): the patents were for genetic markers indicative of polymorphisms of the gene, and methods for identifying these kinds of markers, rather than for the gene or any presumed causative mutation itself (Rothschild and Plastow 2002). The research and its putative application could therefore live without genes. In the context of practices already centred on the identification and mapping of markers, and the continued presence of quantitative genetics approaches, gene-hunting diversions could be readily re-routed towards a marker-centred approach amenable to quantitative genetic methods. This was made possible by the intersection of quantitative and molecular approaches. A marker-centred historiographical lens allows us to discern this dynamic in the way that a gene-centred approach, which might instead interpret the effort to find and use candidate genes as a failure, may not.

The pursuance of MAS was a key part of the strategy of PIC, a company with close links to researchers at publicly-funded research institutions. To aid in the development and implementation of MAS, as well as alternative approaches favoured by other companies, the livestock genetics community continued to find new ways of identifying and mapping markers. They also sought to develop the methods to make direct use of those markers in selective breeding, rather than just using them to identify genes. This was, for instance, a key difference with initiatives such as the Human Genome Mapping Project or the Human Genome Project, in large part because of differences in intended translational goals.

Chris Haley has attested that the unified concept of the marker, with the actual processes of detection and the exact material differences between different kinds of marker abstracted away, was key to helping people from disparate disciplinary and research traditions to work together.¹⁴ The different markers became equivalent units, amenable to—and using the methods of—both quantitative genetics and molecular genetics. This was because the identification and localisation of markers did not require them to be imbued with the kinds of theoretical connotations, for instance concerning the nature and distribution of gene effects, that would otherwise prevent these two approaches from adopting a common orientation to these objects. The marker therefore enabled different communities to contribute towards the common task of mapping. It also functioned as a unit with which comparisons with other maps of the pig genome, and the genomes of other species, could be made.

3.2 Markers as a basis for comparison

To take advantage of the resources available from better-resourced communities working on human genetics, pig geneticists were keen to identify correspondences between the human and pig genomes. Recognising homologous regions of the respective genomes of human and pig could aid in the identification of candidate genes, as with RYR, and also in selecting which probes to use from human sources to aid the mapping of markers in the pig genome. The use of human genome data did not mean that the purposes of its use would be the same. For instance, for pig geneticists the search for markers would not necessarily be a precursor to the localisation and characterisation of genes, as was conceived in human genome research.

Comparative mapping was a key strategy of pig genomics from its inception. It required initial marker and mapping data to be able to begin to ascertain the comparative relationships on which inferences could be built. Rather than simply providing an initial skeleton on which pig-specific data would provide the flesh, it opened up an ongoing iterative process, once sufficient development of maps enabled comparisons of correspondences between pig and human genomes to be drawn. The identification of ever more precise patterns of correspondence (which included the creation of comparative maps), enabled maps to be populated with progressively more markers. This, in turn, enabled the further elaboration of relations of homology between areas of the pig and human genomes.

An example of this derives from some of the early work undertaken using labelled probes derived from human DNA libraries, then called heterologous painting or chromosome painting, now called Zoo-FISH. The probes were derived from mapped areas of the human genome, and would fluoresce when hybridised to corresponding pig DNA, providing a signal when photographed. Human genome data were also used to identify potential primers to amplify markers in particular regions of the pig’s genome; markers do not simply present themselves, but require some means of teasing them out from the genome.

These methods enabled researchers to determine areas of synteny—conserved order between chromosomes, in this case between pig and human chromosomes—and the pig chromosomes could then be sorted and identified. As well as this, a further refinement of the inferential techniques was advanced by the finding that conserved synteny did not necessarily imply conservation of gene order on the chromosome (Rettenberger et al. 1995). Markers were therefore used as indicators to assess homology relationships.¹⁵

Synteny and the relative positions of markers were also detected through methods based on the hybridisation of porcine cells (and the DNA contained within) with those of more genetically well-characterised species such as mice, rats and hamsters: somatic cell hybrid mapping. The basic principle was to use the presence of known markers from better genetically-characterised species to identify areas of synteny and linkage, the latter based on measurements of the co-retention of markers.

One example of this, which became an important mapping method and tool from the late-1990s, was radiation hybrid mapping. In radiation hybrid mapping, cells are irradiated, and then fused with recipient cells of another species. The irradiation breaks apart the DNA, fragments of which then hybridise to corresponding parts of the recipient cell’s chromosomes. The presence or absence of particular markers can then be ascertained and measured, and used to physically map markers to particular parts of the genome, and discern linkage relationships through their co-retention (or otherwise) on individual fragments. The higher the initial dose of radiation, the smaller the fragments of DNA, and therefore the higher resolution the mapping.

The first whole genome radiation hybrid panel for mapping (the IMpRH panel) was developed in 1998 through a collaboration between teams at the Institut National de la Recherche Agronomique (INRA) institute at Castanet-Tolosan near Toulouse and at the University of Minnesota in the USA (Yerle et al. 1998). The same collaborators used the IMpRH panel to produce a map consisting of 757 markers in 128 linkage groups. They noted that this method was not as reliant upon highly polymorphic markers as previous maps, and that the search for QTL would be progressed by further “increasing the amount of comparative mapping information between the swine and human expression maps”, as well as using genome libraries to isolate new markers of interest, once the genomic region of interest is identified in this way (Hawken et al. 1999).

3.3 Using markers

If a marker is identifiable by some test, and that marker is indeed linked (technically, in ‘linkage disequilibrium’) with a gene that may be implicated in variation in a trait of interest, in principle one could use genotyping tests for this and other markers of interest to inform selective breeding decisions. Similarly, if the gene and particular variants of it have been identified, as for example RYR1, then its presence could be tested for.

Using a new method of DNA fingerprinting first published in 1995, Amplified Fragment Length Polymorphisms (AFLPs) could also be identified (Vos et al. 1995). PIC identified AFLPs as a potential rich source of markers, for use in MAS as well as in identifying QTL for economically important traits. In collaboration with the Dutch company that developed AFLPs, Keygene NV, PIC ran a three-year European Commission-funded project as part of the BIOTECH2 programme of Framework Programme 4 to develop the means to use AFLPs in animal breeding.¹⁶

Physical mapping also continued apace, with the use of the successive radiation hybrid panels to map increasing numbers of markers. Between April 1999 and November 2005, 7138 markers were mapped using the two INRA-maintained radiation hybrid panels.¹⁷

MAS was thought to have the potential to be used in selection programmes: within breeds to capitalise on the variation existent in a given breed; or between breeds, for example to introduce (introgress) a desired gene variant or allele from one breed to a recipient breed. In addition to the identification and mapping of markers, extensive phenotypic data and information on the pedigree of the animals would be required to develop an effective programme of MAS (Dekkers and van der Werf 2007).

PIC claims to have implemented MAS for pork quality traits in 1998. The strategy of using a select set of markers persisted into the late-2000s. One estimate of the impact of MAS in the European Union was that in the livestock industry as a whole, MAS had been used in the breeding of between 40 and 80% of breeding females, and that the total direct economic contribution to the livestock industry had been 207–560 million euros (Papatryfon et al. 2008).¹⁸

In the annual reports of Genus plc, the leading animal breeding company into which PIC became incorporated, the last mention of a strategy based on the identification of particular markers from the total set of genetic markers they had identified, was in 2009.¹⁹ In the 2011 annual report, a new approach was cited: genomic selection.

Historian of animal breeding Margaret Derry has questioned the value of MAS, viewing it as a Mendelian interlude between the ‘black-box thinking’ and infinitesimal model that dominated breeding before the late-1980s, and now again in the genomic era (Derry 2015, pp. 131–159). The Mendelian characterisation is not, however, strictly true for many of the markers mapped and used to inform selective breeding decisions, at least if the Mendelian interest in transmission across generations is excluded. Where an actual gene was not the basis of selection, a marker could be used that was close enough to a gene whose presence might be inferred, but not necessarily confirmed, let alone characterised. The marker was characterised by simply existing as a marker first and foremost, with no presuppositions concerning its role in constructing the phenotype, or being a marker for a particular phenotypic effect, change or trait. As geneticist Michael Goddard told us, “for us animal breeders it doesn’t matter, we regard them all as markers.”²⁰

What was being searched for was a way to link the marker to some form of phenotypic variation on which there was data. Even where this is found, it may not be presumed to be causative of anything mechanistically or be the ‘real’ marker of the phenotypic trait. It may just be statistically and probabilistically associated with the phenotypic variation because it is linked to a particular stretch of DNA that is mechanistically and functionally associated. But it is not important that the ‘real’ marker or ‘causative’ gene ever be identified. The marker is neither like a classic Mendelian gene, nor a gene that is a defined sequence of DNA that is transcribed into RNA and then translated into protein (sensu Francis Crick, as explicated by Griffiths and Stotz 2013).