Heritability of Face Recognition

To access facial information about identity certainly belongs to the highest visual skills and is part of “visual intelligence” (Hoffman 1998). Faces per se are crucial for nonverbal communication and necessary for directing ones attention (e.g. by gaze direction) and for effective social interactions. Face perception provides a multitude of different information and awareness of that provides critical information about social status, health status, physical attractiveness, gender, age, allowing considerations about one s, life style, nutritional condition, or eventually premature aging. Sexual attractiveness is perceived not only consciously but also unconsciously. An elevated but still physiological level of testosterone in males results in a receding forehead/hairline and is significantly more common in politicians representing “alpha leader”. Woman mate preferences differ according to their cycle-based fertility status regarding among other cues men s facial masculinity (Gangestad et al. 2010). Also, faces allow on-the-spot diagnosis of many genetic syndromes (Gorlin et al. 2001). Face perception (to see a face as a face) is predominantly triggered by the T-shape order of eyes, nose, mouth (Tsao and Livingstone 2004). Face perception is followed by face processing which finally may result in face recognition. Humans are extremely competent to recognize someone by the face alone. This high cognitive skill is very robust and also very rapid and allows individualizing a face out of thousands of familiar and unfamiliar faces. The false-negative and false-positive rates of these multiple daily decisions appear to be extremely low. Face perception is characterized by its simultaneous sensitivity and insensitivity to subtle changes. On the one hand subtle changes complicate the representation of invariant aspects of faces necessary for face recognition. On the other hand these changeable aspects play a central role for social communication (Haxby et al. 2002). Constitution will fluctuate depending on changing life style or physical exercise which is reflected in the face like facial rash; facial skin might be clean, bright, youthful; lips might be dry, cracked, puffy; eyes might be dark, bruised appearance below the eyes. However, face recognition ability is not significantly hampered by these changes.

It could be shown that these genes are switched on by the gene PAX6 (review van Heyningen and Williamson 2002) (Fig. 1).

Development of the brain
As the optic nerve is rather part of the brain than a "nerve" it is only consequent to assume that the visual processing system of the visual cortex of the brain evolved hand in hand with the information gathering system -the eye (Fig. 1). In the last decade a slowly increasing number of papers support a substantial genetic control and environmental influence. This is true for the total brain volume which differs between ethnic groups as well as partial morphometric measures including total gray and white matter (Baare et al. 2001, Joshi et al. 2011, Hulshoff Pol et al. 2006, cortical thickness (Schmitt et al. 2008, Lenroot et al. 2009, 2010 or visual fine structure (in the cat, Kashube et al. 2002). Also the white matter integrity and connectivity is beside some environmental input under genetic control (Pfefferbaum et al. 2001, Chiang et al. 2009, 2011. Total lack of visual experience affects the fine structure of visual cortex neurons (Wallace et al. 2006). In case of face recognition ability early visual input is necessary for developing full expertise (LeGrand et al. 2003(LeGrand et al. , 2006. But maturation starts before the onset of vision and there is ample evidence that genetics also plays an important role during development of neural circuitry of face processing (Luo et al. 2008, Polk et al. 2007, Tropea et al. 2011. It is suggested that full quantitative maturity is reached latest at 5-7 years of age (Crookes & Kone 2009).

Processing and recognition of faces
In general, facial information processing comprises a number of different processes targeted at extracting different kinds of information. The resulting processing complexity was first addressed at a functional level by conceptual models of face recognition which aimed at separating different processing stages, e.g. by Hay and Young (1982) in a linear hierarchic model and by Bruce and Young (1986) in a box-and-arrow model which is still considered a standard reference model (Fig. 2). As a box-and-arrow model the focus is on a conceptualization of functional steps but no reference is provided as to how these steps could be achieved by specific information processing and where in the brain these processing modules could be located. The latter has been clarified to some extent by more recent imaging studies revealing a distributed neural system of face recognition (Haxby et al. 2000, see Fig. 3).

Modular vs. distributed face processing
In 2001 two papers appeared in the same volume of the Science Magazine. One supporting a modular (Downing et al. 2001) the other a distributed fashion (Haxby et al. 2001) of processing and finally of recognizing faces in the brain. Cohen and Tong 2001 comment on this seemingly alternate or controversial findings: "We can imagine a variety of possible intermediate or alternative positions: a heterogeneous mix of special porpose modules and more distributed general mechanisms; representations that appear modular at one scale but distributed at finer scales; or representational structure that does not divide along the lines of common stimulus categories (such as faces versus objects) but rather is organized along more complex or abstract dimensions". Quiroga et al. (2005) even showed visual representation by single neurons in the medial temporal lobe. As the same neurons also report when showing the written name of the same individual these cells might simply be memory cells and might express maximal compact or sparse coding. Fig. 1. Hypothetical evolution of the visual systems of eye and brain. The morphogenetic pathways consist of the information-gathering system and the information-processing system. Both are not linear but rather a complex network. The "eye development pathway" is partially adopted from Gehring & Ikeo 1999, Gehring 2005. Evolution started with eye pigment coded by the Rhodopsin gene and accidentally by a universal master control gene PAX6. Successively new genes were acquired stochastically and evolutionarily optimized -this has been termed "intercalary evolution" (Gehring & Ikeo 1999). There is an estimate of around 2,000 genes required for eye morphology only (Halder et al. 1995). In analogy to the eye also the brain with the neuronal visual system might fit the concept of intercalary evolution. In a bilateral response both the eye and the visual cortex had been successively adapted and optimized. (MRT image of the head by courtesy of Stanislaw Milachowski, Department of Radiology, Westfälische Wilhelms University, Münster, Germany).

Fig. 2.
A two dimensional box-and-arrow diagram for illustrating a functional model for face recognition (modified from Bruce & Young 1986, Deffke 2005. The central idea of the model is the clear distinction between face recognition unit (FRU) and the person identity nodes (PIN). It is also assumed that the person identity nodes -for each individual there should be one -can be accessed via the voice, the name, or facial markers like the hair line. Dotted arrows denote the processing of familiar faces and dashed arrows the processing of unfamiliar faces.  Haxby et al., 2000).

Face specific neuronal processing vs. unspecific high resolution pattern recognition
A strongly debated question is whether and how face processing differs from non-face processing. It is argued that face recognition has a specific neuronal processing (Kanwisher et al. 2000) and that it is not just an instance of higher order object recognition. Others propose that face recognition is just a specific instance of visual expertise and the same neuronal mechanisms and areas are also recruited by bird, dog or car experts in their field of expertise (Gauthier et al. 2000, Gauthier andBukach 2007). While it is undisputed that faces are processed in specific areas, the debate is still ongoing as to whether these areas are exclusively recruited for the recognition of faces (domain-specificity). More generally, this debate can be framed as part of the more general nature vs. nurture discussion and is still ongoing (Pascalis et al. 2009, Park et al. 2009, 2010. In analogy to face recognition it has also been suggested that the human voice is an "auditory face" (Belin et al. 2004) based on auditory expertise (Chartrand et al. 2008). For more than a decade it is known that face recognition is a very quick function of the brain. As early as 100 to 120 ms there is a electrophysiological response to emotional faces or face categorization and after 170 there is an event related potential (called N170) associated with face recognition (Eimer & Holmes 2002, Liu et al. 2002. This supports the idea that special neuronal pathways for visual recognition of faces exist. This hypothesis of face-specific processing pathways is further substantiated by brain imaging studies (functional brain mapping studies, fMRI). Brain areas in the occipito-temporal cortex preferentially response to faces. In particular in the mid-fusiform area (FFA) there is a significant response when someone is seeing a face but much less to non-facial objects (Behrman et al. 2005). The same is true for the inferior occipital gyrus (IOG) (Hoffman & Haxby 2000). Interestingly also an evolutionary old brain area beyond the cortex and the amygdala responds significantly to faces (Behrman et al. 2007). The relevance for face processing of the FFA is also supported by patients who suffered a sudden loss of face recognition ability after a traumatic event covering this area (e.g. Steeves et al. 2009). In a recent study with a large sample of people who lack face recognition ability from childhood without any detectable brain damage some anatomical and functional differences to a control collection could be seen. In particular a diminished gray matter density in the bilateral lingual gyrus, the right middle temporal gyrus, and the dorsolateral prefrontal cortex. A decreased functional activity in the left fusiform face area and the dorsolateral prefrontal cortex. An enhanced activation in the left medial prefrontal cortex and the anterior cingulate. This is suggestive of a network dysfunction and anatomic curtailing of visual processing in the lingual gyrus (Dinkelacker et al. 2011). Experimental investigations of differences in face and object recognition are complicated by the fact that for most people faces are the only area of visual expertise. Thus, a direct comparison of neural recruitment for different areas of expertise, which would be needed to distinguish between the two conflicting views on face specificity, is most often not possible. From a theoretical viewpoint there are clearly differences in the processing of objects as stimuli belonging to a specific class or as stimuli idiosyncratically displayed by a specific individual (Stollhoff 2010(Stollhoff , 2011. By incorporating these differences into models of facial encoding, Stollhoff et al. (2011b) provided a link from deficits in face recognition (behavioral level), via a lack of holistic processing (computational level) to decreased structural neural network connectivity (implementation level). Whether or not this link also holds in the reverse direction such that a comparatively increased neural connectivity is specific to the processing of faces in contrast to the more general visual expertise still needs to be investigated.

High functioning and low functioning face recognition
There is a broad distribution of face recognition ability in humans. From daily experience we are not only aware of people with ordinary face recognition ability but also of some who are extremely good but also some who are very poor. An increasing number of papers support that the variability of face recognition ability in humans is even much higher than hitherto thought. In the context of studies with very poor recognizers (i.e. "face-blind" people or prosopagnosics, see chapter 2.4.4) there were self-reports of people who claimed that their ability is just opposite. These people with exceptionally good face recognition ability were called "super-recognizers" (Russell et al. 2009).

Heritability of neurocognitive functions and dysfunctions
There is an increasing but still surprisingly low number of isolated neurocognitive deficits with a proven or suspected genetic background. Such demonstrable impairments in visual, auditory, tactile, smell and other perceptions also help to characterize the normal basic neuronal mechanisms that have an impact on multiple brain modalities (Tab. 1).

Heritability of face perception and recognition
There is converging evidence that face recognition is highly heritable. The neurogenetic background and heredity of visual intelligence can be achieved through an increasing number of very different approaches.

Evolution of face perception and recognition
"Evolution is best understood as the genetic turnover of the individuals of every population from generation to generation" (Mayr 2001). Visual expertise of face recognition is evolutionarily conserved in distant species telling us about a strong genetic background. There are many reasons for improving face recognition over non-facial recognition. One reason might be described by "preparedness" (Seligman 1970(Seligman , cit. in Öhman 2005. Hunter-gatherers sitting in a palaeolithic cave certainly will have an increased chance to survive when they can decide immediately whether the face of someone looking inside is a family member or a stranger and a putative enemy. Better survival means better reproduction. Depending on the selective pressure even a slight advantage can drive evolution. Since ancient times it is of common knowledge that elephants never forget a (human) face -at least when associated with aversive events. Sheep underwent behavioral and electrophysiological tests for visual face recognition of faces (Kendrick et al. 2001). In a number of trials these sheep could accurately discriminate not only individual sheep faces but also human faces even after 600 days. Such a high visual expertise is also reported in social insects with much simpler nervous systems. The paper wasps (Tibbetts and Dale 2004) and honeybees (Dyer et al. 2005) are able to recognize conspecifics by facial cues only. In honeybees this ability is also shown for human(!) faces. It should be mentioned that honey bees have less than 1 million neurons in contrast to 100 billion neurons in humans (Pakkenberg and Gundersen, 1997). As the experiment only tested recognition of specific face images, and not the ability to generalize across different views, successful completion of the experiment would have been possible without an engagement of all of the cortical processes normally involved in face recognition.

Gender bias for face recognition
Sex differences of a variety of cognitive functions are known (Rehnman and Herlitz 2006). Women outperform men in their ability to recognize female faces. When shown male faces both perform on similar levels (Rehnman and Herlitz 2006). The authors have no explanation for the neurophysiology of this female driven own-sex-bias. On the other hand there is a positive correlation of verbal IQ and face recognition in men but not(!) in women (Herlitz and Yonker 2002). Nothing indicates that this sex-biased advantage is a result of social or cultural learning only. As gender and a variety of secondary sexual characteristics are strongly genetically determined also genetic modification of sex-biased face recognition might be assumed. A causal interpretation still requires caution. Epigenetic mechanisms also might in part or fully explain these differences. The term epigenetic refers to modification of the DNA that does not alter the DNA sequence. The DNA stores the genetic information and the expression of genes results e.g. in the processing of structural and regulatory proteins. It is now known that this basic genetic information stored in the DNA is just optional. The reason is that the genetic information underlies large modifications (e.g. by methylation). This is also called "imprint". By this not the gene sequence is altered but the way the specific information is processed, e.g. by enhancing or silencing gene expression. Thus epigenetics enables different read-outs from a fixed template (review Egger et al. 2004, Fagiolini et al. 2009).

Twin studies
As already delineated there is no doubt that the morphological and functional brain development is under genetic control. Yet, knowledge about the heritability of cognitive skills is still poor. Best described is the genetic impact of superior general intelligence (Haworth et al. 2010). A classical approach to behavioural genetic traits which do not follow simple Mendelian inheritances are twin studies. Such studies allow an estimate of the respective environmental and genetic contribution to a given phenotype. Identical or monozygotic twins are genetically almost similar. Hence, the more discordant they are for a behavioural trait the less heritable and the more environmental effects account for the phenotype. In other words, best proof for high heritability are concordant monozygotic twins which are grown up in different families. By such studies the heritability of general cognitive ability (mostly called g) was estimated. Results from different studies vary. Metaanalyses of such studies give heritability estimates of 50%, i.e. half of the total variance in g is due to genetic difference between individuals (Harworth et al. 2010). In a functional magnetic resonance imaging twin study a significant influence of genetic similarity on the cortical response is shown for face and place stimuli but not for objects like chairs. Monozygotic twins showed a significantly more similar neural activity pattern than dizygotic twins (Polk et al. 2007). In a recent twin study it was shown that such genetic differences can account not only for face perception but also for face recognition ability.

Hereditary prosopagnosia (i.e. non-functioning face recognition)
The strongest support for heritability of face recognition per se comes from the counterpart. There is a dysfunction in the neuronal processing of face recognition which is called prosopagnosia. Prosopagnosia is a specific inability to recognize familiar faces. The synonym face blindness is misleading as the perception of human faces is preserved. The term prosopagnosia was coined by Bodamer in 1947 as an assimilation of the Greek words prosopon ( όσω ν) meaning face and agnosia (αγνωσία) meaning not knowing, not recognizing or ignorant. He came across the phenomenon of acquired prosopagnosia when examining soldiers suffering from brain injuries. In a detailed review of the literature he states that the first well documented clinical descriptions of an agnosia would be from Quaglino (1867)(cited by Bodamer 1947). Della Sala and Young (2003) come back to that case showing that this observation is very typical for prosopagnosia. The oldest convincing report is from Wigan (1844) describing a man who is not able to remember faces: "He would converse with a person for an hour, but after an interval of a day could not recognise him again. Even friends, with whom he had been engaged in business transactions, he was unconscious of ever having seen. Being in an occupation in which it was essential to cultivate the good-will of the public, his life was made perfectly miserable by this unfortunate defect, and his time was passed in offending and apologizing". These cases have in common that they are acquired by traumatic events or strokes.
Only recently, an increasing number of cases with an idiopathic form of prosopagnosiamostly called congenital or developmental form -in the absence of any exogenous event are reported (review Behrmann and Avidan, 2005). Until 2003 three reports about familial segregation were published clearly supporting the heredity of "non"-face recognition (De Haan 1999, Galaburda andDuchaine 2003, McConachie 1976) (Fig. 4).
In contrast to the rare acquired form of prosopagnosia, the inborn form of prosopagnosia is among the most common anomalies in humans with a prevalence of around 2% (Bowles et al., 2009, Kennerknecht et al. 2006, 2008a. Most interestingly these cases almost always run familiar following a simple mode of segregation patterns (Kennerknecht et al. 2008b). We therefore coined the term hereditary prosopagnosia (Schwarzer et al. 2006, Grüter et al. 2007, Kennerknecht et al. 2006. Such recurrent disorders of higher visual function may help to delineate normal neuronal processing. The function of a gene can be tested in the laboratory by genetically engineered mice either by turning off gene expression by targeted mutation or rudely by knocking out the gene physically ("knock-out mouse"). Mice have gradually become a model system in vision research (Luo et al. 2008, Wang et al. 2010, Zhang et al. 2007. From a genetic approach the prosopagnosic subjects might serve as "natural knock-out" humans for loss-offunction experiments for the module "face recognition". So far no genes or candidate genes are known to be associated with the development of visual processing or in particular with face processing and face recognition. However, formal genetic studies of familiar transmission of dysfunctions of the visual intelligence are powerful tools to guide molecular genetic dissection. mother also claims not to be able to recognize familiar faces. She suggests that as a child she had less difficulty, than her daughter because she lived in a small community, and went to a school where no uniform was worn. She remembers believing that a family friend was two different people until one day he put on his spectacles in her company. She has found ways, over time, of overcoming the handicap, for example, by using tone of voice for recognition." (2) De Haan (1999) Galaburda and Duchaine (2003): "TA´s [see arrow] son, mother and grandmother also have prosopagnosia, it clearly has a genetic basis".
Besides functional testing the diagnosis can also be established by highly informative narratives about situations where family members or very close friends have been overlooked. This can further be substantiated by the finding of compensation strategies for overcoming such problems. E.g. normal sighted (emmetropic) prosopagnosics apologize for having forgotten their glasses or for having been kept in thought. Such compensation strategies are otherwise very rare in control collections. The question is whether the prosopagnosic phenotype can be unambiguously distinguished from poor face recognition ability? (Fig. 5A). If there are overlapping features it might be argued that these could be due to individual development of a more or less effective compensation strategies and/or clinical distinct severity of symptoms and/or part of varying methods of assessment (Minnebusch et al. 2007, LeGrand et al. 2006, Stollhoff et al. 2010, 2011 (Fig. 5B).
Prosopagnosics and controls show only as a group statistically significant differences by behavioural and physiological tests. On an individual level there is always an overlap in the performance of some control and prosopagnosic individuals in a varying number of tests. Moreover there is no single test available today which unambiguously allows to differentiate between a poor recognizer (at the lower end of the normal face recognition distribution) and a prosopagnosic (at the extreme low end). In principal, studies of familial segregation of variable face recognition might help to address this topic. As prosopagnosia is of increasing scientific interest it is not surprising that intra familial variability of face recognition was described recently (Schmalzl et al. 2008, Kischka 2011 (Fig. 6).

A)
B) Fig. 6. Intrafamilial variability of face recognition. A) White symbols denote not impaired; black symbols denote impaired on familiarity tasks; grey symbols denote normal on familiarity task, impaired on more subtle measures of face processing (Schmalzl et al., 2008). B) White symbols denote unimpaired; shaded symbols denote face recognition ability below average; black symbols denote prosopagnosia (Kischka 2011).
Further studies will show whether the phenotypic variability follows a continuum described by a bimodal distribution (Fig. 7A) or simply by a Gaussian distribution (Fig. 7B). At the low end there are 2.5% prosopagnosics and it is assumed that at the high end there are the same number of super recognizers (Russell et al. 2009). So far this seems highly suggestive of a regular bell-shaped distribution.

Genetic considerations
Human face recognition is highly heritable however it´s genetic basis is unclear. Based on literature and own data an attempt is made for the genetic dissection of this higher level perception and cognition skill. Different genetic traits support one or the other finding and there is ample evidence that the complex neuronal processing also has a complex genetic background. More than 2,000 genes are involved in eye development only (Halder et al. 1995). A screening for genes that wire the (visual) cortex has been started in the mouse (review Lokmane and Garel 2011). One such approach is to look in transgenic mice for defects during embryogenesis and the functional consequences on the developing cerebral cortex. These genetic studies allow conclusions on the etiology of cerebral disorders in mice and also in humans and help to dissect the molecular pathways responsible for normal brain development.
A) B) Fig. 7. Face recognition ability might be continuously distributed either in a (A) bimodal way with respect to non-functioning face recognizing (prosopagnosia) or (B) alternatively, low and high performers are just tail-enders of a normal distribution.
a. Mendelian inheritance Already the first published cases of congenital prosopagnosia showed familiar segregation and are fully compatible with a monogenic disorder of dominant phenotypic expression (Fig. 3). Dominant in that case means that the phenotype is already expressed in the heterozygous status i.e. when only one of the two inherited parental copies (alleles) shows the mutation. This is surprising as behavioral phenotypes are part of complex genetic mechanisms, environmental influences and epigenetic modifications. One explanation might be that already one dominant acting gene mutation in the molecular cascade for the processing of face recognition interrupts or disturbs the resulting neurocognitive function. Along with such an assumption all our observations of familiar cases perfectly fit. We have meanwhile more than 100 families with recurrent prosopagnosia in two to four generations (the first 38 families of them are reported in Kennerknecht et al. 2008b). In favor of an autosomal dominant inheritance are (1) vertical transmission of the disorder, (2) males and females are equally impaired, and (3) father-to son transmission. As the father only transmits the Y-chromosome to a son (but an X-chromosome to the daughter) X-linked inheritance is excluded (Fig. 8). Among 38 pedigrees (Kennerknecht et al. 2008b) only 6 exceptions were observed which are still compatible with the concept of autosomal dominant inheritance. In four families one generation is "skipped", i.e. an obligate carrier does not manifest prosopagnosia (normal transmitter) but one parent and the child(s). In two other families there is evidence of a de novo mutation as only one family member is impaired. It can be argued that already mutations in only one gene of the gene cascade or gene network might be sufficient in disrupting the processing of face recognition. This does not necessarily mean that all obligate carriers of such a mutation must manifest prosopagnosia. "Generation skipping" is a common phenomenon in autosomal dominant traits and is described by incomplete penetrance. Yet, it cannot be differentiated whether these are sporadic cases or isolated familial cases which are the only carriers of the mutation in a family. The genetic background (which is similar but not identical in a family) and putative epigenetic mechanisms might modify gene expression (Fig. 9). Such formal genetic considerations are extremely helpful in dissecting moleculargenetic etiology of face recognition.  Table 2. Schematic genotype-phenotype correlation of assumed digenic (genes A and B), multiallelic inheritance of face recognition. The higher the figure the more it contributes to the phenotype.
According to these assumptions the prosopagnosic mother might have the genotype A0/A4 and B4/B5 and the normal father A4/A5 and B4/B5. In a two-factor-cross of the uncoupled genes the filial generation (F1 generation) predicts a variety of genotypes and resulting phenotypes (Tab. 3).  Table 3. Genotypic and phenotypic segregation ratios of a two-factor-crossing.

Paternal gametes
According to the resulting genotypes in the filial generation (see Tab. 3) the genotypes of the family members with variable face recognition ability (see Fig. 3) are simulated (Fig. 10). Fig. 10. Observed phenotypic distribution in a large family with assumed genotypes as derived from a hypothetic two-factor-cross (see Tab. 3).
The observed segregation ratio comes close to the expected segregation ratio (Tab. 4).

Face recognition ability Expected segregation [N]
Observed segregation in the family of Fig. 7 Table 4. Genotypic and phenotypic segregation ratios of a two-factor-cross of the family described in Fig. 10.

c. Multifactorial inheritance
In a multifactorial mode of inheritance a phenotype is not entirely defined by one or more gene(s) (i.e. monogenic or polygenic) but also significantly influenced by environment and/or experience. In short, endogenic and exogenic contribution is necessary. Twin studies are an ideal approach to evaluate the relative contribution of nature and nurture. In monocygotic twins who almost completely share the same genetic background a high concordance or discordance of a given phenotype stands for a high or low genetic influence and vice versa for a low or high environmental influence. So far twin studies show a significant genetic influence on functional brain organization as documented by functional brain imaging studies (Polk et al. 2007) and by face recognition memory tests (Willmer et al. 2010).

Conclusion
Face recognition ability is a high level perception and recognition skill. It remains unclear whether it is neurologically processed in a modular or in a distributed manner in the cortex or subcortical structures. Also an open question is whether face recognition ability differs from object recognition or whether it is just the top functioning end of visual recognition. At all levels a consistent finding however, is the high heritability of face recognition. So far there is no gene cloned which functions in the cascade of the visual cortex. Yet, the genetics of cerebral visual function has been shown by different approaches. Here we argue that prosopagnosia may open a window on the physiology and genetics of normal face recognition (= visual intelligence). From a large collection of prosopagnosic families phenotypic heterogeneity is obvious. Preliminary molecular genetic data also indicate genetic heterogeneity. This might be due to altered gene expression of one or more genes of the gene network associated with brain development. Depending on the impairment different familiar segregation patterns are expected either following classical modes of inheritance (Mendelian phenotype), polygenic or multifactorial inheritance and/or epigenetic modifications (Fig. 11).
[Major control gene]  Fig. 11. A hypothetical network of genes and their respective expression that contributes to face recognition ability. Exogenic/environmental and epigenetic influences are not shown but should also play a role. A) Suggested allelic variance of the respective genes will differentially attribute to the phenotype. In a given population the face recognition ability might describe a Gaussian distribution. B) In a small subgroup one or more allelic drop-outs (e.g. by loss-of-function mutations) might occur which then result in the phenotype of poor recognizers or prosopagnosics.
By means of a permanent functional contact it can be considered that eye and brain development are simultaneously driven forward by evolution (co-evolution) as well as backward (retrograde evolution). During this process a plethora of genes had entered the network by gene duplications and successive evolutionary adaption of the duplex to its new task (intercalary evolution).

Acknowledgment
This project is supported by the University of Münster, Germany and approved by the local ethical committee (protocol No 3XKenn2).