Congenital anomalies from a physics perspective. The key role of "manufacturing" volatility
Alex Bois, Eduardo M. Garcia-Roger, Elim Hong, Stefan Hutzler, Ali Irannezhad, Abdelkrim Mannioui, Peter Richmond, Bertrand M. Roehner, Stephane Tronche
aa r X i v : . [ phy s i c s . b i o - ph ] M a y Congenital anomalies from a physics perspective.The key role of “manufacturing” volatility
Alex Bois , Eduardo M. Garcia-Roger , Elim Hong , Stefan Hutzler , Ali Irannezhad , Abdelkrim Mannioui , Peter Richmond , Bertrand M. Roehner , St´ephane Tronche Version of 3 May 2019
Key-words: Congenital anomalies, malformations, infant mortality, manufacturing defects
1: Aquatic facility, Pierre and Marie Curie Campus, Sorbonne University, Paris, France.Email: [email protected]: Institut Cavanilles de Biodiversitat I Biologia Evolutiva, University of Val`encia, Spain.Email: [email protected]: Neuroscience Laboratory, Sorbonne University and INSERM (National Institute for Health andMedical Research).Email: [email protected]: School of Physics, Trinity College, Dublin, Ireland.Email: [email protected]: School of Physics, Trinity College, Dublin, Ireland.Email: [email protected]: Aquatic facility, Pierre and Marie Curie Campus, Sorbonne University, Paris, France.Email: [email protected]: School of Physics, Trinity College Dublin, Ireland.Email: peter [email protected]: Institute for Theoretical and High Energy Physics (LPTHE), Pierre and Marie Curie campus,Sorbonne University, Centre de la Recherche Scientifique (CNRS). Paris, France.Email: [email protected]: Aquatic facility, Pierre and Marie Curie Campus, Sorbonne University, Paris, France.Email: [email protected]
Abstract
Genetic and environmental factors are traditionaly seen as the sole causes of con-genital anomalies. In this paper we introduce a third possible cause, namely ran-dom “manufacturing” discrepancies with respect to “design” values. A clear way todemonstrate the existence of this component is to “shut” the two others and to seewhether or not there is remaining variability. Perfect clones raised under well con-trolled laboratory conditions fulfill the conditions for such a test. Carried out for fourdifferent species, the test reveals a variability remainder of the order of 10%-20% interms of coefficient of variation. As an example, the CV of the volume of E.colibacteria immediately after binary fission is of the order of 10%.In short, “manufacturing” discrepancies occur randomly, even when no harmful mu-tation or environmental factors are involved. If the pathway is particularly long orrequires exceptional accuracy, output dispersion will be high and may lead to mal-formations. This effect will be referred to as the dispersion effect . We conjecturethat it will be particularly significant when major changes occur; this includes theearly phase of embryogenesis or the steps leading from stem cells to differentiated(organ-specific) cells.The dispersion effect not only causes malformations but also innocuous variabil-ity. For instance monozygotic (MZ) twins resemble each other but are not strictlyidentical. It is not uncommon to see only one of the twins of a MZ pair showing acongenital defect (see Appendix A).Not surprisingly, there is a strong connection between congenital defects and infantmortality. In the wake of birth there is a gradual elimination of defective units andthis screening accounts for the post-natal fall of infant mortality. For reasons whichare not yet fully understood, this fall continues until the age of 10 years. Neither dowe understand why, as a function of age, the downward trend of human infant mor-tality follows a power law with an exponent around 1 (whereas for fish it is about 3,see Bois et al. 2019a). Apart from this trend, post-natal death rates also have humpsand peaks associated with various inabilities and defects.In short, infant mortality rates convert the case-by-case and mostly qualitative prob-lem of congenital malformations into a global quantitative effect which, so to say,summarizes and registers what goes wrong in the embryonic phase.Based on the natural assumption that for simple organisms (e.g. rotifers) the man-ufacturing processes are shorter than for more complex organisms (e.g. mammals),fewer congenital anomalies are expected. Somehow, this feature should be visibleon the infant mortality rate. How this conjecture can be tested is outlined in ourconclusion.
Contents
Introduction: the “manufacturing dispersion”effect
Variability in biochemical reactionsFrom technical systems to living organisms: a physics perspectiveBroad reach of congenital anomaliesRandomness of the dispersion effectRationale for an output dispersion effectControl proceduresWhy defect statistics give a biased pictureAre complex organs more affected by output dispersion?The most frequent defects have close links with normalityWeak role of genetic factors in birth defectsIdentification of output dispersion through phenotype variationsOutline of the paper
Fault-tolerant design
Tolerance issues in the industrial production of mechanical devicesThe tolerance system as a way to mitigate the effects of manufacturing defectsOutput dispersion in biological systems
Salient features of embryonic mortality
Implication of geometrical abnormalities for development of the embryoAge-dependent embryonic mortalityGeneral observations about embryonic mortalityAvian species. Fish species. Human fetal deathsHow can one explain that the death rate is highest at the beginning of embryogenesis?A conjecture about embryogenesis in unicellular organismsInfluence of temperature on hatching rate
Salient features of the two phases of human mortality
Human infant mortality for all causes of deathThe age of 10 seen as an equilibrium point between screening and wear-outInfant mortality for specific causes of death
Conclusion
Main resultsRationale for cross species comparisons
Appendix A. Estimating the strength of genetic factors
Mutations and repair mechanismsThe twin methodology for assessing the strength of genetic factorsStrength of genetic factors in cancer
This paper is the first leg of an exploration in three parts; the two others areBois et al. (2019a,b). Despite the connections the three papers can be readindependently from each other.
Introduction: the “manufacturing dispersion” effect
In a characteristic way the abstract of a recent paper about birth defects begins withthe following sentence: “The causes of birth defects are complex and include ge-netic and environmental factors and/or their interactions” (Chen et al. 2018). Inother words, genetic and environmental factors are seen as the only sources of birthdefects. Here we add a third source referred to as a “manufacturing dispersion” ef-fect. Its introduction is motivated by several reasons which are outlined in comingsubsections.
Variability in biochemical reactions
Thousands of biochemical reactions are required for the growth of any living organ-ism even if it is a single cell. Taken together they constitute what one may call amanufacturing process. For each of these reactions there is a set of optimal param-eters in terms of temperature, pH, concentration of enzyme-catalyst, orientation andshape of interacting molecules and so on. It is clear that mutations and environmentalfactors may disrupt this process. However, in the present paper we develop the ideathat even if all parameters are set at their optimal design values nevertheless therewill be a dispersion of the outcomes. It has four main causes. (i) Initial conditionsmay not be identical. (ii) Even if initial conditions are very similar, there will be“butterfly effects” (due to the nonlinearity of the reactions) which will greatly am-plify any initial dissimilarities no matter how tiny. (iii) The parameters defining thereactions are never exactly at their optimum values. (iv) Random quantum fluctua-tions cannot be avoided. Note that this last effect is probably smaller than the others.Even if at each step the volatility is small, a succession of steps will result in a cu-mulative effect which, eventually, may lead to noticeable congenital anomalies.As an illustration of this kind of variability consider an observation made at the levelof individual cells. According to a recent study (Wallden et al. 2016, p.729,733,Fig. 4B,C), isogenic E. coli cells (i.e. having same genotype) growing in a uni-form and invariable environment display significant variability in volume at birth(i.e. volume immediately after binary fission) and in individual growth rates. Thecoefficients of variation are fairly substantial, of the order of CV ≃ , re- In practice this means: (i) a time interval sufficiently short to ensure that the likelihood of mutations is negligiblecompared to other reactions. (ii) constant optimal environmental conditions of the kind maintained in controled laboratoryexperiments. spectively. Actually, variability at cell level has already been recognized and studied(at least qualitatively) in the 1910s and 1920s as will be documented later on.
From technical systems to living organisms: a physics perspective
In this paper we examine biological systems from the perspective of reliability engi-neering. Such a comparative approach is rather uncommon in biology; in contrast,comparative analysis plays a key-role in experimental physics. Therefore, it is per-haps not surprising that it is tried by physicists and biologists who share a similarturn of mind .Why should it be useful to establish a link between technical and living systems? Inphysics it is natural to take systems that we understand pretty well as starting pointsfor the investigation of phenomena that remain mysterious .One should not focus only on similarities, differences may also be revealing. Arather obvious illustration is that, whereas in engineering the duplication of criticalcomponents is a common technique for improving reliability, mammals have onlyone heart not to speak of many other vital functions for which there is no backup.For instance, urinary retention can occur for many reasons whether physiologicalor neurological and, if not remedied, may lead to death within a few hours. Yet,there is no backup mechanism. We are told that Tycho Brahe, one of the foundingfathers of modern astronomy, died that way. This example is of interest because,whereas adding a second heart would require a considerable design change, creatinga supplementary bladder outlet would be a fairly simple matter. Broad reach of congenital anomalies
Malformations versus deficiencies This paper is mainly about congenital anoma-lies We prefer this expression to birth defects for two reasons: (i) Many anomaliesdo not appear in the form of malformations but as deficiencies, e.g. insufficient pro-duction of insulin in Type 1 diabetes. (ii) Many congenital anomalies do not appearat birth nor even in childhood but much later in the course of life; anomalous heartvalves are an example that will be discussed later on. Behavioral anomalies may alsoappear only later in the course of life. Having said that, we will sometimes also use“birth defects” which has the advantage of being shorter.Anomalies of the immune system It should be observed that in fact it is difficultto separate mortality due to congenital anomalies from other causes of death. Evencancer or mortality from infectious diseases may be attributed to congenital anoma- Not long ago, in an email of 31 December 2018, Prof. Bert Vogelstein, a biologist renowned for his work on cancertold us: “We need more physicists thinking about cancer”. Such a statement was certainly an encouragement. Many such cases can be found throughout the history of physics. One of the most recent examples is how electromag-netism, more precisely quantum electrodynamics (QED) was used as a guide for building a theory of strong interactions,namely quantum chromodynamics (QCD). lies of the immune system. In this respect one should remember that even in majorepidemics such as the Spanish influenza pandemic of October-November 1918 lessthan 10% of the population was affected in the sense of being hospitalized and onlyabout 0.4% died which means that most persons were protected by their immunesystem. Only a few were not.Behavioralanomalies The behavior of living organisms is to a large extent genet-ically controled. As an example consider the case of a broody hen. From the eggslaid by the hen to the hatching of chicks 21 days later there is a succession of stepswhich is quite remarkable.(i) The process starts when the eggs are fertilized by the rooster inside the hen’sbody. (ii)
Physiological changes.
The beginning of the process is also marked byphysiological changes: the body temperature of the hen increases and the feathersunder her body fall off . (iii) Making a nest.
The hen makes a nest about 5cm deepby scratching the ground. (iv)
Storage of the eggs.
As the hen will brood a set ofabout 6 to 10 eggs, over several days she will lay eggs and store them in the nest.A delay of up to 6 days will make little difference in hatching time. As soon as anegg has been laid, it will cool down and the content will contract whereby the aircell is created. It will play a crucial role during hatching because it is always on thisside that the chicks will pierce the shell. (v)
Sitting on the eggs.
While sitting onthe eggs, the hen will have to turn them in order to prevent the embryonic chicksfrom sticking to the shell. As well as turning them she will also move the eggs onthe outside of the nest into the middle and the middle ones out so all are evenlywarmed. A graph presented in the section on embryogenesis shows that in terms oftemperature there are very strict requirements. (vi)
Cleaning the nest.
The hen willhave to keep the nest clean and tidy which, in particular, means that non-fertilized orbroken eggs must be discarded. (vii)
Taking breaks.
The hen will leave the eggs one,two or three times a day (each time for about 15 minutes) to find food and water andto defecate. Nonetheless, a hen will usually lose weight while brooding. (viii)
Lastthree days.
Toward the end of the incubation and particularly during the last threedays the embryos start to produce significant levels of metabolic heat. Therefore,brooding should be relaxed. When the chicks start to break their shells the hen mustgive them enough room. (ix) After hatching the chicks remain close to or underneaththe hen thereby sharing her body heat. For the same reason, after hatching in anincubator chicks are kept warm by infrared lamps. They are fully feathered only atsix weeks of age. In natural conditions the hen will show her chicks how to identifyand peck food.It is by purpose that we have described this process in some detail to show how eas- Injections of the hormones prolactin, luteinizing, and oestradiol to non-broody hens induces broodiness. ily it can be disrupted or become sub-optimal (in the sense of a reduced hatchingrate). Usually the disruptions which may trigger anomalies remain hidden to the out-side observer. On the contrary, in the brooding process inappropriate environmentalconditions which at each step may derail the process occur in full view and can beidentified . This gives an intuitive view of the notion of manufacturing volatility.Just as for phenotype characteristics, there is also a substantial variability in brood-ing ability and behavior. Some hens are very good at brooding while others are not .For instance, first time brooders might not stay broody for very long.In the same way as birth defects are nothing but amplified forms of normal variabil-ity, similarly some forms of behavior become sufficiently extreme to be labelled as“abnormal”. Here are a few examples. • Sometimes, hens will go broody without eggs underneath them. In some casesthey may continue to sit on empty nests for 2 or 3 months. • In many bird species males and females alternate sitting on the eggs. Curiously,the same behavior was observed for two hens (Buibaku et al. 2010). During the daythey were alternating: one was incubating from morning till noon while the otherwas out to eat, drink, dust bath and rest. Roles were interchanged around midday.During the night the two hens were jointly brooding. However, the result was notsatisfactory in the sense that of the 22 eggs they were brooding only one was able tohatch.
Randomness of the dispersion effect
The main defining characteristic of the dispersion effect is its randomness. However,this word does not mean that anything can happen and that nothing can be predicted.In fact, there are predictable consequences. For instance the dispersion does notmanifest itself in the same way in a process that requires high accuracy than inone which does not. Several illustrative examples are described below. Whetherthe dispersion occurs at the beginning or at the end of a pathway will also make adifference. Rationale for an output dispersion effect
There are several motivations for introducing the dispersion effect. In other words, the analysis of the brooding process offers an excellent observational opportunity to explore theinteraction between genotype and environmental conditions. To our best knowledge, this field of research has not yetbeen explored in a systematic way. A simple test consists in putting an egg in front of a hen. If she pulls the egg under her she may be well “gifted”. For instance, for the eyes even a small disymmetry may result in strabismus. For the ears synchronization require-ments are less critical. In molecular biology the term “pathway” has a technical meaning in reference with the expression of genes. Here, weuse the word more broadly as referring to a succession of steps realizing a given function. It can be a cascade of chemicalreactions or also a succession of actions. An illustration is the feeding function which requires an organism to see theprey, then to identify and catch it and finally to eat and digest it. (1)
Most birth defects are unexplained.
For most birth defects the factor re-sponsible is not known. A recent publication in the “British Medical Journal” (Feld-camp et al. 2017) tells us that in a total of 5,504 birth defects in 270,878 childrenborn in the state of Utah in 2005–2009, the etiology is unknown for 3,390 whichrepresents 80% of the cases. Of the 1,104 cases for which the etiology is known,844 are due to chromosomal abnormalities which are mostly trisomy 13, 18 and 21.In our conception most defects occur randomly, so it is hardly surprising that manyremain unexplained.(2)
Variability in true twins.
Many articles (e.g. Ahmed et al. 2017) give the(misguided) impression that most malformations can be attributed to specific genes.If this were true, the twins of monozygotic pairs would have the same birth anoma-lies. In fact, as shown in Appendix A, the discordant cases (where the two twinsdo not have the same defect) are 4 times more frequent than the concordant cases(where they share the same defect).At this point it is necessary to say a word about epigenetic changes, a notion whichrefers to how genes are expressed rather than to their identity. The present-day con-sensus is that to be considered as epigenetic a trait has to be heritable at least fora number of generations. This is certainly a wise rule for otherwise any differenceoccurring between true twins could (somewhat arbitrarily) be attributed to epigeneticfactors.(3)
Variability of offspring in uniparental reproduction.
Inheritance fromtwo parents is a difficult problem. The study of true twins is one way to over-come this difficulty. The study of reproduction from a single parent is another. Uni-parental reproduction was much studied between 1900 and 1930 particularly at the“Zoological Laboratory” of John Hopkins University; see the studies of Ruth Stock-ing (1913,1915), Ralph Middleton (1915), Herbert Jennings (1916), Bessie Noyes(1922). Uniparental reproduction (also called asexual reproduction) occurs in twocases.The simplest is the reproduction by fission of unicellular organisms. In her thesis(Noyes 1923) Bessie Noyes cites four species of protozoans for which inheritabilitywas studied.The same kind of investigation can be made for multicellular organisms (i.e. meta-zoans) with uniparental reproduction. For instance, in rotifer species during its lifetime of a few days one female can generate successively of the order of 10 offspring.Although they are in a sense clones of their mother, they present a substantial vari-ability (Noyes 1923).It is true that one can never exclude that a somatic mutation (i.e. a DNA alteration)occurred during the embryogenesis of offsprings. Yet, it is well known that errors inprotein synthesis are far more frequent than errors in DNA replication (Drummond et al. 2009).(4)
Dispersion of outputs.
The three previous points explain that there is roomfor a third source of birth defects but it does not describe what this source could be.It is simply the fact in any manufacturing process there are two parts: (i) The designphase (ii) The implementation of the design. For living organisms it is the DNA-RNA code which represents the design instructions destined to the manufacturingprocess.In real life, a design is never carried out with absolute accuracy. If a table is designedwith a width of 3m, in reality its width will be comprised between W = 2 , mmand W = 3 , mm. For most practical usages such small discrepancies are of noconsequence. However, if one wants to bring the table into a room whose door has awidth of 3m, then the W table will get through whereas the W table will not.This is a static view. As soon as there is a nonlinear process evolving in time (whichis the case of most biochemical reactions) there will be butterfly effects throughwhich small initial differences are amplified.(5) Crucial role of early discrepancies
In 2015 it was shown that mutationswhich eventually lead to cancer cells may occur at different stages of the transfor-mation of undifferentiated stem cells into mature differentiated cells (Tomasetti etal. 2015). This discovery provided a natural explanation for the fact, known sincethe 1920s (Greenough 1925, Patey et al. 1928), that cancer cells which have a lowdegree of differentiation are also the most malignant, that is to say, result in early re-currence and death. Indeed, a mutation occurring early in the differentiation processwill impact and derail all following stages.There is a similar feature with the embryo itself in the sense that organs in the ear-liest stage of their development are most sensitive to teratogenic (i.e. causing de-velopmental malformations) factors at the time of their appearance. This point isshown very clearly in a paper by Uchida et al. (2018); in this study various shocks(e.g. heat shocks) were applied in different stages of the embryo development ofzebrafish, frogs and chicken. In all cases embryonic lethality was the most severewhen the shock was applied in the earliest stage.This observation has a natural interpretation in the manufacturing framework; itsays that a small defect in a component A used in the early stages of a productionchain may have quite detrimental consequences because it may hinder the appropri-ate working of components introduced later on in the process and with which A isfunctionally related.The manufacturing conception developed in this paper is consistent with (yet broader To use for living organisms the expression “manufacturing process” may seem odd. However, our objective is pre-cisely to watch living systems from the perspective of technical reliability science. than) the mechanism identified in Tomasetti et al. (2015) and which the authorsdescribe as follows:“The concept underlying the current work is that many genomic changes occursimply by chance during DNA replication rather than as a result of carcinogenicfactors.” Therefore, one expects a correlation between “the lifetime number ofdivisions among the stem cells within each organ and the lifetime risk of cancerarising in that organ.”Each division bringing about a further step in the differentiation process also repre-sents a new manufacturing challenge which makes it more prone to output dispersionthan mere divisions into identical daughter cells. Whether the discrepancy occurs bymutation or by output dispersion, its impact will be more severe if it occurs early inthe differentiation chain.In medical language, such early cell anomalies are labeled as pre-cancerous condi-tions. They are characterized by the presence of abnormal cells, yet in low proportionand in shapes which are not very different from normal types.In the manufacturing of living organisms mechanical operations play a role (see be-low) but most pathways consist of a succession of chemical reactions. The previousargument remains valid however. Conditions of concentrations, temperature, acidityor other parameters are never 100% optimum; as a result, the outputs will have adispersion around optimal design values.(6) Critical processes are the most affected.
Under the term “critical pro-cesses” we understand processes which require high synchronicity and accuracy.Whenever two sheets must grow at the same speed in order to join seamlessly, evena slight discrepancy may affect the closure. Examples of defects of this kind are: • Spina bifida, a defect of closure around the spine. From the open to the closedform there is a broad range of severity for this defect. Spina bifida occulta is a closedform which is quite frequent; it affects 15% of newborn according to estimates butcauses no sypmtoms. About this case one can read the following assessment: “Theexact causes of spina bifida occulta are not well understood. Both genetic and en-vironmental factors seem to play a role”. Our thesis is that there are no causes; itis a purely random effect. The fact that slight defects are much more common thansevere defects is consistent with a dispersion mechanism. An explanation based onmutations is less satisfactory. It is true that severe forms may affect the reproductibil-ity rate and therefore the transmission of possible genetic factors but there would belittle difference in this respect between light forms and very light forms. • Cleft lip and palate or more generally facial cleft. • The positioning of the eyes (i.e. iris+pupil+lens) also requires high accuracy Table 1: Incidence of birth defects in high accuracy processes.
Birth defect Description Prevalence(per 1,000)“All” birth defects Cases with “geometrical” defectsStrabismus Eyes not properly synchronized Heart valves defects Abnormal joints of cuspids Cleft palate and/or cleft lip Facial sheets do not join well Spina bifida (open) Defect in spine closure . Spina bifida occulta Slight defect in spine closure
Among children with trisomy 21Strabismus Eyes not properly synchronized
Heart Serious congenital heart defects
Notes: Prevalence is defined as the total number of births affected by the problem in a time interval of severalyears compared to the total number of live births in the same time interval. All these cases are characterized by“mechanical” or “geometrical” defects. The cuspids designate the leaflets which form the valve. In most valvesthere should be three leaflets; when two leaflets stick together it is a bicuspid defect. There can also be 1 or 4cuspids but these defects are fairly rare. Incidentally, the fact that the prevalence of the four causes mentionedis higher than the “all defect” prevalence estimate shows that the “all defect” notion does not include somelight cases (e.g. light strabismus or spina bifida occulta) or defects which manifest themselves only later in thecourse of life (e.g. light valve defects). Most often spina bifida occulta (i.e. not visible) causes no symptomsand is only identified through X-ray imaging.Trisomy 21 (that is to say three chromosomes number 21 instead of two) results in over-production of theproteins under the control of the 310 genes located on this chromosome. This disrupts many mechanismsand particularly those requiring high accuracy: brain (100% are more or less affected), heart (40% seriouscongenital anomalies), eyes (strabismus affects 35%), ears (hearing loss affects 70%).
Source: Child health, USA 2014, Table 1: National prevalence estimates of selected major birth defects;Gunton et al. (2015); for spina bifida occulta: estimate of the “National Institute of Neurological Disordersand Stroke”. because the two eyes must move in a synchronized way. For each eye positioning re-lies on two muscles (one on each side) whose actions must be perfectly coordinated.As it is not easy to achieve such high accuracy requirements it is hardly surprisingthat, as shown in Table 1 strabismus is one of the most frequent birth defects (2% ofbirths). • Heart valve defects are almost as frequent as strabismus. More details will begiven later.
Control procedures
In industrial production there are control procedures all along the supply and produc-tion chains. There are certainly similar control procedures in the making of livingorganisms. Although we do not know them very well there has been progress in thisdirection in recent decades. For instance, the role played by the non-coding region2
In industrial production there are control procedures all along the supply and produc-tion chains. There are certainly similar control procedures in the making of livingorganisms. Although we do not know them very well there has been progress in thisdirection in recent decades. For instance, the role played by the non-coding region2 of the genome (which represents 98.5%) is becoming clearer.Spontaneous abortion can be seen as a control mechanism but the occurrence of livebirths with severe malformations (e.g. anencephaly, that is to say newborns withouta brain, whose prevalence is about 120 per million births) shows that this control isinsufficient. It is true that apoptosis (that is to say programmed cell death) is a localcontrol mechanism, but it is surprising that massive defects at macro level are notidentified and corrected. In our industrial analogy it would mean producing aircraftwithout wings .The dispersion conception would also suggest more frequent defects in highly com-plex organs than in simpler ones. However, before we discuss this point we need toassess the reliability of defect statistics. Why defect statistics give a biased picture
The statistics of birth defects released by hospitals give a picture which is biased in(at least) three respects.(1) Very serious defects usually will lead to early abortion or still births. This factcan be illustrated by the following data. In 13,614 births that occurred in an hospitalof Rajasthan (India) in 2012 there were 431 stillborn and 13,183 live-births. Amongthe stillborn, 18% had a birth defect whereas only 0.64% of the live-births had adefect. (Vyas 2016). Thus, many serious cases will not be included if birth statisticsare restricted to live-births.(2) Many slight defects will not be recorded because they will give rise to symp-toms only much later. This can be the case even for heart defects; for instance lightvalve defects or stenosis (i.e. narrowing) will be noticed only at the age of 40 or50. It is the same problem for many other internal defects. Whereas polydactily (i.e.more than 5 fingers) can be detected visually just by inspection, many slight defectsof internal organs may never appear or appear only later in life.(3) For a complex organ like the brain, there is no well defined border line betweenwhat is normal and what is not. Thus, the fact that some persons can sing very wellwhile others cannot will not be considered as a congenital defect. Even more seriousdefects (such as a propensity to autism) will appear only later on in life; as a result therespective role of genetic, environmental or dispersion factors will remain unclear.For that reason, although the brain is by far the most complex organ of a human bodyit will be left aside in the next subsection where we discuss the role of complexity.
Are complex organs more affected by output dispersion?
The manufacturing process of an airliner requires more accuracy and controls than It can be argued that this is an anthropocentric view for indeed the ability to fly may not be the main purpose. Afterall there are insects and birds which have wings but cannot fly. the production of bicycles. Similarly, in a human body some organs are more com-plex than others. Obviously, the heart is a more complicated device than the bones ,the skin or even the liver. Therefore, the fact that heart defects are the most frequentcongenital malformation comes as a nice confirmation of the dispersion conception.In contrast, defects based on mutations are not expected to follow the same rule. Itseems natural to admit that the number of mutations (including harmful mutations)is proportional to the number of genes involved in the manufacturing of each specificorgan. As each gene codes for a specific protein one would have to admit that thenumber of proteins is in relation with the complexity of an organ. If data are availablesuch numbers could provide a useful metric for estimating the complexity of variousorgans. The most frequent defects have close links with normality
Defects, particularly minor defects, are usually “in line” with normal organs. In orderto explain what we mean by this expression let us consider polydactily defects. Canthe 6th finger appear anywhere?Firstly, one can observe that the additional finger is never perpendicular to the hand.Can it appear anywhere in the plane of the hand? Observation shows that it is muchmore likely to appear on each side of the hand (that is to say next to the thumb orlittle finger) than next to the three inner fingers. In other words the 6th finger is morelikely to appear as an addition to the normal blue print rather than as a drastic changein the normal design.A similar observation can be made for the heart valves. Consider for instance theaortic valve which is located at the beginning of the aortic artery. Whereas normallyit has three leaflets the defect which is by far the most frequent is when two of themstick together. The prevalence of this so-called bicuspid aortic valve (BAV) defectis between 1% and 2%. In contrast, the quadricuspid aortic valve (QAV) is a rarecongenital anomaly with an incidence of only 0.01% (Schaeffer et al. 2007).Why is the first defect more in line with the normal valve than is the quadricuspid?The BAV originates from the fusion of two existing leaflets whereas the QAV re-quires the creation of an additional leaflet with corresponding changes to the threeothers in order to make room for the new one. Such a defect would require significantdesign changes.
Weak role of genetic factors in birth defects
At first sight it may seem that the dispersion effect is only of marginal importance It is true that “complicated” has no obvious meaning. Even a single cell is very “complicated”. In addition it can beargued that the bone marrow is very essential. What we mean here is that seen from outside a pump (which is what theheart is) is more difficult to design and build than a table leg. compared to the genetic and environmental factors. For a better assessment we use amethodology based on the observation of pairs of twins.How similar are monozygotic twins? The fact that they may look “alike” is not suffi-cient proof of their similarity. This can be illustrated by a case reported in Williamson(1965, p.166). In a study of family characteristics of congenital malformations donein Southampton (UK) the author reports the case of twins who were “similar in haircolor, eye color, head shape, finger nail shape, teeth pattern and many other features”but one of these twins was a hydrocephalic (too high pressure of fluid in the brain)male while the other was a normal male.It is true that no valid conclusion can be drawn from a single case but this kind ofobservation is confirmed by a recent study of 6,752 monozygotic (MZ) twins and13,310 dizygotic (DZ) twins in California observed from 1957 to 1982 (Yu et al.2019).MZ twins share 100% of their genome whereas DZ twins share on average 50%of their genome (Yu et al. 2019, p.18). In Appendix A we explain a method forassessing the role of genetic factors. When applied to the data given in Yu et al.(2019) it leads (see Appendix A) to the conclusion that genetic factors play in facta fairly weak role in major congenital malformations. This leaves free space for (i)environmental factors and (ii) for the dispersion effect described above.Is it possible to discriminate between (i) and (ii)? For birth defects the only envi-ronmental factors which can play a role are those which affect the mother. Manyfactors of that kind were considered by researchers, e.g. age, level of education,birthweight, birth order, season of birth, smoking of the mother. It appears that onlysmoking of the mother is significantly associated with congenital defects (Yu etal. 2019). However, why should smoking of the mother affect one twin and not theother? Identification of output dispersion through phenotype variations
Observation of uniparental reproduction offers a fairly direct view of the effect ofoutput dispersion. It allows the notion of “pure line” (also called “inbred line” or“inbred strain”) to be defined in a rigorous way as being formed by the offspring ofa single individual. In contrast, for sexual reproduction a strain is considered inbredwhen it has undergone at least 20 successive endogenous matings (brother-sister orparents-offspring) but even at this point the individuals are only nearly clones. Thatis why in the first half of the 20th century there have been many investigations ofuniparental reproduction. In order to measure more accurately the influence of this factor it would be useful to do a comparative analysiscovering a sample of countries with highly different levels of tobacco consumption. Fig. 1 gives two illustrations. They are followed by a table which lists causes ofcongenital anomalies.
Weight (mg) . mm (Phascolus vulg.) (b) Difflugia corona Mother Children
Fission (before separation) (a) Brown beans
Fig. 1 Two examples of output dispersion in uniparental reproduction. Top:
Dispersion in the weight of418 bean seeds in a pure line obtained from a single grandmother seed through self pollination. The histogramis well described by a Gaussian distribution of mean m = 455 mg and standard deviation σ = 70 mg which givesa coefficient of variation CV = 17 . . For all nineteen pure lines totaling 5,494 beans CV = 19 . . Theseexperiments were done by Wilhelm Johannsen in 1900-1902. Bottom:
Dispersion in the aspect of
Difflugiacorona , an unicellular protozoan living in water. As reproduction is by fission (as shown on the left-hand sidefor two pairs differing in size) all 3 descendants of the first individual (at the left) are clones. However, thereare variations in their aspect at time of fission, particularly in number of spines on the shell. Note that naturalself-pollination is not exactly the same thing as asexual uniparental reproduction; the later produces real cloneswhereas in the former (when performed naturally) there is a high degree of inbreeding which however may besomewhat less than 100%.
Sources: Johannsen 1903 (p.22-28), Jennings 1916 (p.438-439).
The main difference between the two experiments shown in Fig.1 lies in the numberof successive generations that can be observed. For Johannsen’s beans there wasonly one harvest per year whereas under good conditions the protozoans reproducedat intervals of 3 to 5 days, that is to say almost one hundred times faster than thebeans. Another difference is that the second experiment relied mainly on resultsexpressed in integers: either the number of spines whose range is 0-7 or the numberof teeth around the mouth which is an integer smaller than 17.A study with a similar objective was published in 1915 by Ms. Ruth Stocking whichwas based on variations occurring in paramecia ( Paramecium caudatum ), a largeunicellular organism which lives in fresh water. Here again, as reproduction is byfission (and does not involve conjugation episodes), the descendants of each single The mouth cannot be seen on the picture describing the fission process because it is located at the separation betweenthe mother and daughter cells. individual will constitute a pure line. The study focused on the shape of the parame-cia. A recapitulation figure (p. 408) shows a bewildering diversity of forms from thestandard ellipse to strange shapes with many tentacles. Table 2: Mechanisms related to congenital anomalies.
Mechanism Passed Identification test ExampletooffspringYes/NoDesign glitch Mutation in DNA of gem cells Yes Genome sequencing Trisomy 21 Mutation in DNA of somatic cells No Non inheritable Cancerabnormal cellsManufacturing glitch Environmental interference No Epidemiological studies Effect of nicotine Random output dispersion No Uniparental reproduction StrabismRepair mechanism Apoptosis (programmed cell death) No Finger separation in embryoNotes: Four comments are in order. • It is the word “random” which characterizes the difference between items 3 and 4. It means that dispersionin outputs occurs even in optimum conditions, i.e. when no harmful environmental factor is present. • Mutation and repair mechanisms can hardly be separated for most often we can see only their combinedeffects. If the cells resulting from a somatic mutation are quickly eliminated through apoptosis nothing willappear. • Uniparental inheritance tests allow a distinction between (2) and (3)+(4). If, as seems natural, the amountof somatic mutations increases with time, their contribution to congenital anomalies should be fairly small.Moreover, when (3) can be excluded in the controlled environment of a laboratory experiments, then (4) seemsthe most likely mechanism for the abnormalities shown in the text. • Epigenetic mutation was not included in the table for its status does not seem clearly defined. For instance,one of its mechanisms involves the addition of methyl radicals CH to the molecules composing the DNA butwhat triggers this addition remains unclear. How can one account for the variations observed in those experiments? Standardfactors are listed in Table 2. Item 1 is clearly excluded because the changes were notinheritable. Item 2 seems unlikely. If somatic mutations are random and independentfrom one another their number must be proportional to the number of cells and tothe time interval. Thus, for unicellular organisms observed at fission time this effectshould be minimal.What can be said about item 3? With a little imagination one can easily suggest This statement just results from basic probability theory. Peto’s paradox relies on what happens not at cell level butat the level of the organism (“Why don’t all whales have cancer?”). A mutated cell will lead to cancer only if it is notremoved by the immune system. Is the immune system of the mice used in laboratories not affected by the fact that theyare pure line mice? possible environmental factors. Thus, for beans one can mention the position ofthe beans in the pods and the location of the pods on the plant. However, whyshould such discontinuous factors lead to almost perfect Gaussian distributions? Forthe protozoa which were raised in laboratory conditions and identical medium it ismore difficult (yet not impossible) to cite environmental factors. In a general way,however, in order to make a convincing case for a specific environmental factor,evidence must be provided showing that in a series of tests it has indeed the claimedeffect. Otherwise it would be just an ad hoc explanation.It is surprising that item 4 is almost never mentioned. In particular, we did not findit in the numerous papers of the 1910s and 1920s analyzing asexual reproduction.Yet, is it not a natural mechanism? It can easily account for continuous variability asdescribed in Johannsen’s paper because its randomness leads naturally to Gaussiandistributions. Through the Central Limit theorem of probability the occurrence of arandom discrepancy X i at each step i of a multi-step pathway gives a nearly Gaussiandistribution for the sum of the X i (at least if the X i are independent).Through the hole and shaft mechanism described below item 4 can also account forvariability by leaps, as happens for spine numbers or a similar effect for tentaclenumbers in Lashley (1915).In principle if the manufacturing process is known it should be possible to computeand predict the variability of the output (except if butterfly effects play a major role).In other words, this framework can really be tested. Although in the present paperwe limited ourselves to qualitative or semi-quantitative tests, subsequently it shouldbe possible to find cases simple enough to allow modeling. Outline of the paper
The paper proceeds through the following steps.(1) First, we explain why random output fluctuations are inevitable in any produc-tion process. It is only thanks to a sound management of defects that an assemblageof several (defective) parts can be made workable. Depending on the specific indus-try, those management systems use different ways. We will focus on the tolerancesystem in use for mechanical systems because it is probably the easiest to under-stand.(2) Secondly, we explain in what respects the two phases of human mortality,the “wear-in” and “wear-out” phases, bear close resemblance with the failure modesdefined in reliability engineering.(3) If simple technical devices can give us a better understanding of how toachieve minimal manufacturing defects, is it not natural to try the same approachfor living systems? For instance, is the shape of the age-dependent infant mortalityof simple living systems similar to or different from that of humans? This leads us in our conclusion to outline an agenda of cross-species investigations. Fault-tolerant design
In order to make industrial production able to cope with output discrepancies in thesupply chain appropriate systems have been developed. In the following subsectionwe explain briefly the tolerance system for mechanical devices. In recent decadesmuch attention has also been given to electronic semiconductor systems because ofthe high complexity reached by such systems which may have millions (or even bil-lions) of components (Dubrova 2013). In a broad way, the purpose is always the sameand can well be summarized by the title of a paper written by John Von Neumann in1952, namely: “Synthesis of reliable organisms from unreliable components” (VonNeumann 1956).
Tolerance issues in the industrial production of mechanical devices
First of all, it should be realized that mechanical operations involve inherent outputvariations. This was already mentioned earlier in an informal way; let us see moreprecisely how the tolerance system can deal with it.Two holes made on a lathe with the same drill bit (say of 10mm diameter) in analuminum cylinder will in fact not have the same diameter. The boring operationwill introduce a small but unavoidable random error. For instance, the diameter ofthe holes may be . mm and . mm respectively; naturally, the measurementintroduces an additional uncertainty which will be ignored here for the sake of sim-plicity.One may think that this small difference is of little importance but suppose that thishole is destined to receive a shaft which has a diameter of . mm. This will bepossible for hole 1 but not for hole 2. In short, even small discrepancies may preventassemblage.As already mentioned, in embryo-genesis there is a somewhat similar problem whentwo separate sheets are expected to join. In such cases even a small discrepancy ingrowth velocities may disrupt normal closure. This may create a defect of the neuraltube which results in a birth abnormality called “Spina bifida”, a Latin expressionwhich means “spine split in two”. Similarly, disruption of the closure of the left andright facial sheets may result in what is called a cleft lip and cleft palate. We comeback to this point below.A related case is the genesis of the furcula. In humans the furcula consists of twoseparate bones called clavicles or collarbones. On the contrary in birds it is a singleV shaped bone called furcula (latin for small fourk) or wishbone. Located in the upper chest of birds it is an essential structural element which allows them to movetheir wings; it also acts as a mechanical spring during flight. On day 13 of the 21-daylong embryogenesis of chicks the left and right collar bones meet and close togetherto form the furcula. It can be predicted that even small discrepancies can preventgood working of this critical element. The tolerance system as a way to mitigate the effects of manufacturing defects
Mechanical engineers have developed a system of standardized tolerances . In thiscontext a tolerance is a specification which gives not only the nominal dimensionbut also the allowed margin. As an example, for the previous hole, the specificationwould be: 10 +0.015-0 mm, meaning that it may be up to 0.015 mm larger than thenominal dimension, but 0 mm smaller (that is to say it should not be smaller than10mm).The task of the engineer is to give for every dimensions appropriate tolerances sothat, if respected, the device will work. For each separate part the technician whomakes it will check whether or not it is “within tolerances”. If it is not, it will bediscarded and replaced by a suitable one.There are similar tolerance systems for electrical elements such as capacitors or re-sistances. The specification (often written on the element itself) may indicate thenominal value (e.g. 100 Ω ), the margin of error (e.g. ± ), the temperature range(e.g. 5 to 35 Celsius degrees).One could summarize the specification procedure by saying that the science of engi-neering is to make working devices with spare parts which, strictly speaking, are alldefective in the sense that their values differ from the nominal values (but are withintolerance margins). This mechanical example is useful because it allows a clear un-derstanding of the problem but since living organisms are not made with nuts andbolts, nor with resistors, one must explain how this should be adapted. Output dispersion in biological systems
At first sight one may think that the two cells produced in the fission of a parentcell are exactly identical. The previous discussion suggests that in fact they arenot, but does not explain the why and how. Basically, biological processes consistin a succession of physico-chemical reactions. In order to give an intuitive feelingof why such reactions are sensitive even to fairly small condition changes we willmake three points. (i) First we emphasize the relatively high frequency of errors inprotein folding. (ii) Secondly, we explain how spatial factors play a great role inreactions involving enzymes. (iii) Thirdly, we consider a simple reaction whose highsensitivity to temperature may be familiar to many readers.(1) It has been recognized that “errors arise at all steps of protein synthesis, from Defects D e f ec t s D e f ec t s Defects
Dispersion ofparameter 1 Dispersion ofparameter 2
Parameter 1 P a r a m e t e r Withintolerance W i t h i n t o l er a n ce Withintolerancefor the twoparameters
Fig. 2 Within and out of tolerance areas when a process depends on two parameters.
In this schematicrepresentation it is assumed that a process depends simultaneously on two parameters, each of which has aGaussian distribution. The green dot represents the (optimal) design values of each parameter. As illustrationsone can mention the following cases. (i) The green dot corresponds to the ideal center of a hole that is drilledinto an aluminum cylinder. Actual centers in 50 successive realizations are represented by the red crosses.Although never exactly at the design location, the effective centers may be close to it and fall within thetolerance domain represented in yellow (note that it may have another shape than a simple disk). (ii) For achemical reaction parameter 1 may be the concentration of one component and parameter 2 the concentration ofthe other. Then, the green dot corresponds to the optimum concentrations. For the process to unfold successfullyboth parameters must be within tolerance which means that all cases which fall in the magenta region will notwork well and may lead to defects. For a process which has more than two parameters the acceptable zonewould be reduced even further. Such additional parameters could be for instance the temperature and pH. transcription to protein folding, and have widespread phenotypic consequences”.Due particularly to the “fragility” of protein folding mechanisms “errors in pro-tein synthesis are orders of magnitude more frequent than DNA-replication errors”(Drummond et al. 2009). This review paper contains a table which lists a number oferrors along with their estimated frequency.(2) One hallmark of the present paper is to emphasize the role of geometrical andpositional factors. Here is another case of that kind. We know that enzymes (mostenzymes are special kinds of proteins) act as catalysts of chemical reactions. In fact,they are highly sophisticated catalysts in the sense that they can play this role notonly for one specific reaction but for several. In addition their activity can be mod-ulated according to needs. In other words, they are a kind of multipurpose controlstation, somehow like the control room of a power plant. The multipurpose capabil-ity comes from the fact that at their surface they have several so-called active siteswhere the reaction will take place; each active site is coupled with a so-called al-losteric (meaning “other place”) site which will bind with control molecules that can be either activators or inhibitors. Needless to say, if a control molecule is attachednear but somewhat off the right location its regulation function will not be well im-plemented. With allosteric sites that are particularly cramped there can be situationssimilar to the hole and shaft case where even a small shift can greatly affect the en-zyme and therefore the reaction that it is supposed to catalyze. To make things evenmore complicated, one should add that many enzymes do not work well if they arenot bound to helper molecules called cofactors.(3) Our third illustration is a process which may be familiar to many readers. As iswell known, a mayonnaise is made by slowly adding oil to egg yolk, while whiskingvigorously with a fork. An emulsion will form made of small oil droplets. Thesedroplets are strongly held together by van der Waals intermolecular attraction forceswhich cause the high viscosity of mayonnaise (Depree et al. 2001). Addition ofmustard contributes to the taste and further stabilizes the emulsion.This, at least, is the theory.In fact, the operation may fail (i.e. no emulsion forms) for various reasons.(i) It fails when the oil is added too quickly.(ii) It fails when the temperature of the oil is too high; as a matter of fact, it worksbest when the oil and egg come directly from the refrigerator.(iii) Another reason for failure may be the presence on the fork of traces of aproduct which prevents the formation of the emulsion.In short, we have here a simple physico-chemical process which has fairly stricttolerance specifications. If two or several processes are involved either successivelyor at the same time, the tolerance area is further reduced (Fig.2). Salient features of embryonic mortality
In previous sections it was suggested that a manufacturing process which involvesmajor innovations is more prone to faults than mere cell reproduction by fission.That is why, for instance, the transition from stem cells to fully differentiated cells isa more challenging task than duplication.The process by far the most innovative is the transition from a zygote, i.e. a fertilizedcell, to a fully developed embryo. Within a fairly short fraction of the order of 10%of the embryonic period, a completely new organism will be created and each stepis highly dependent upon the satisfactory outcome of previous steps. In other words,this is a critical development process in which major faults are expected to occurwith significant probability.
Implication of geometrical abnormalities for development of the embryo
Fig.3a shows position anomalies occurring in the early steps of embryogenesis and2
Fig.3a shows position anomalies occurring in the early steps of embryogenesis and2 Fig.3b indicates that they have adverse implications as revealed by the fall in hatch-ing rates.
Fig. 3a Cleavage abnormalities in haddock embryos.
Examples of abnormalities occurring in the first stepsof embryogenesis. The growth process starts with one fertilized cell, then subsequent steps every 20mn with2,4,8,16, . . . cells. The pictures show that early defects can occur already in the 8-cell step. Normal developmentis shown on the left-hand side and abnormal development on the right-hand side. The red segment correspondsto 1mm. Apart from the two cases shown here three other sorts of abnormalities are described in the samepaper, namely (i) unequal sizes of the cells (ii) cellular outcrops where one or two cells protrude from the maingroup of cells. (iii) Separation of the 8 cells into two disconnected sets. In the following figure it is shownthat such abnormalities result in lower hatching rates that is to say in increased embryonic mortality.
Source:Adapted from Rideout et al. (2004,p.219) H a t c h i n g r a t e ( % ) ASYMMETRY BAD JUNCTURES H a t c h i n g r a t e ( % ) Fig. 3b Hatching rates for embryos involving malformations.
Hatching rates were measured for 12 samplescontaining various proportions of defective embryos. The coefficients of linear correlation are equal to r = 0 . and r = 0 . , respectively. Similar correlations are obtained for size and outcrop anomalies. Source: Adaptedfrom Rideout et al. (2004, p.222)
Age-dependent embryonic mortality
In demography age-specific death rates are a key-variable . In the embryonic phasethey are paralleled by mortality rates as a function of post-fertilization age which,therefore, should also be seen as a key-variable. Curiously, it attract little attentionso far; as a result, such data are available for only few species. Fig. 3c presents dataobtained by three high-accuracy studies for bird and fish species. The graph alsoshows human data, albeit with the drawback of starting 4 weeks after conception. From a physical perspective the resolution of demographic phenomena into age-specific components is similar tofrequency analysis of physical phenomena; for more details see Berrut et al. (2017).
110 0 5 10 15 20 25 30
Age of embryo (days post fertilization) D e a t h r a t e ( p e r da y a n d , f e r til e e gg s ) ChickenTurkeyEuropean perchHuman
Fig. 3c Embryonic mortality rates.
The bottom and left-hand side scales are for birds. The top and right-hand side scales are for fish (ages are also expressed in days). The scales for humans (not shown) are asfollows: the age scale starts with the 4-7 weeks gestational age interval and ends at 50 weeks. The verticalscale (expressed in rates per 1,000 pregnancies) starts at 2 and ends at 150. Note that the data provided by vitalstatistics agencies usually start only at 20 weeks. The present data for the intervals between 4 and 20 weekswere obtained through a special study covering a 4-year period (1953-1956). Note that the age scale of thechicken case has been extended from 21 to 24 days to facilitate the comparison with the turkey case. Notethat the perch curve is made of straight lines because there are too few data points to use a smoothing option.
Sources of the data: Chicken (broiler): Pe˜nuela et al. (2018, p.6505), number of fertilized eggs ( n ) = 3 , ;turkey: Fairchild et al. (2002, p. 262), n = 51 , ; European perch: Alix (2016, p. 161), n = 13 , ;humans: French et al. (1962, p.840,844), n = 3 , . General observations about embryonic mortality
What can be said about the role of mutations and environmental factors?For the animal experiments described in Fig.3c all embryos were raised in identi-cal conditions so that exogenous factors can hardly explain why some embryos areaffected by severe anomalies while others are not.Mutations are certainly responsible for some anomalies but under the assumption ofa uniform mutation rate it seems difficult to explain the huge changes affecting thedeath rate. For turkey or chicken eggs why should there be more lethal mutations onday 1 than on day 11?For all four species, there is a sharp fall of the death rate between fertilization andthe subsequent leveling off. For the turkey, perch and human cases the death rate isdivided by a factor of about one hundred whereas for chicken the factor is about 30. However this last factor is affected by a substantial uncertainty because of the smallnumbers of deaths; indeed, between days 8 and 14 the daily death numbers are allsmaller than 6 with three of them being zero or one .The fact that for the avian cases there is a second peak on the right-hand side whereasno similar peak appears in the two other cases is due to the fact that birds have topierce the shell of their eggs which is a difficult task. If early neonatal death rateswould be included into the embryonic phase there would also be a left-hand sidepeak in the perch and human cases. In other words, this difference is related to howone defines the end of embryogenesis.As a last point we wish to compare the absolute magnitude of the death rates atthe beginning of the embryo development. For this comparison we leave apart thehuman case for reasons which are explained below.For turkeys the first data point which is an average for the first three days stands at14 per day and per 1,000 fertile eggs. For chicken the average for the first three daysstands at 22 which is close.For the European perch the data point for the first day stands at 167 that is to sayabout 10 times higher than for the birds in 3 days. The interpretation of this differ-ence remains an open question at this point. Avian species
To the two avian cases shown in Fig. 3c one can add that a similar pattern was ob-served for several other avian species, e.g. pigeons, doves, ducks, grouse, pheasantsand quail (Romanoff 1949).The fact that some of these deaths are due to fairly random conditions can be illus-trated by the case of malpositions. It has been observed that one half of all chickembryos which die between day 18 and 20 were in abnormal positions (Hutt 1929).In order to understand the reason one should recall that the lungs of chicks start towork shortly before they begin to break the shell of the egg. However, to make thatpossible they must have access to the air cell which is on the blunt tip of the egg. Iffor some reason their head cannot move in time to the right location the chicks willdie. Moreover, to pierce the eggshell is quite a challenge . If, for some reason, theeggshell is too hard or too thick the chick may be unable to break it. Fish species This is in spite of the fact that the experiment involved 3,240 eggs and that 471 of these embryos died. As the turkeyexperiment involved 10 times more eggs its results are more reliable. For that purpose the chick is using a special “tool” in the form of a so-called egg-tooth which is a sharp temporarystructure on the top of the beak. There is also a special “hatching muscle” which serves the purpose of activating theegg-tooth. The embryonic phase of fish can be studied easily due to the fact that the fertilizationof the eggs occurs outside of the body of the female. For that reason one can getreliable death data even for the very early part of the cycle. For instance, for zebrafishas the first division of the fertilized embryo occurs less than an hour after fertilizationone should be able to get hourly death rates. Unfortunately, such investigations didnot attract much attention so far. To our best knowledge the case of the Europeanperch described in Fig.3c is an unparalleled study of fish embryonic mortality.
Human fetal deaths
The study described in French et al. (1962) took place in the island of Kauai in thestate of Hawaii. During the four years of the study there were 3,083 pregnancies,273 fetal deaths and 2,777 live births. These are of course small numbers due to thefact that the island’s population was only 30,000. The reason for doing the study inthis place was the existence of a well organized network of medical personnel.Very early fetal deaths can only be noticed by the women themselves. That is whythis part of the death rate curve must be recorded through special surveys involving adevoted network of physicians and medical personnel. Standard fetal death statisticsas provided by hospitals include only pregnancies which lasted more than 20 weeks.In the three other cases of Fig.3c the procedure was to observe a sample of N eggsin the course of time and for each subsequent day to record the number of survivingembryos. Clearly, it was not possible to use the same procedure here. As pregnanciesand fetal deaths were recorded in a continuous way the whole process required moreintricate and less transparent computations. How can one explain that the death rate is highest at the beginning of embryo-genesis?
Here is a tentative interpretation of the fact observed in Fig.3c that the death rate ishighest on the first day of the embryogenesis.In principle the organism of the mother produces embryos equipped with all that theyneed to grow. But, as for any real process, there are necessarily faults and defects.The embryos in which some important ingredients are missing will be unable to growand instead will die. As these faulty embryo are gradually eliminated the death ratewill decrease just as observed.At present this mechanism is purely speculative but the interesting point is that it canbe tested. How?Consider for instance the case of zebrafish embryos. Two hours after fertilization theembryo has about 64 cells. If the embryo is able to reach this point it means that itis well equipped, at least for the cleavage phase. In contrast, one would expect thefaulty embryos to be eliminated very shortly after the beginning of the embryogene- sis. This means that the death rate should be highestin the very first hours. In otherwords, this explanation can be tested by measuring the embryonic death rate every 2or 3 hours during the first 24 hours. A conjecture about embryogenesis in unicellular organisms
In unicellular organisms is there a process similar to embryogenesis which precedesthe birth of a new organism? Formally no, but functionally yes. For instance inthe prokaryotic bacterium
Caulobacter crescentus the initiation of replication startssome 2 hours before division actually occurs (Laub et al. 2000,p.2145). This phase(which consists of successive so-called G , S and G transitions) can be consideredas a kind of embryogenesis during which the new organism is made ready for au-tonomous survival.Naturally, the success rate of the complex transformations which take place is cer-tainly not 100.00%. For instance, it has been shown (Cryms et al. 1999) that hyperexpression of one gene (named podJ ) involved in a crucial transition at the beginningof the replication process causes a lethal cell division defect. Thus, it is conceivablethat random fluctuations in the concentration of this protein will lead to a percentageof failures.This means that, in the same way as there is an embryonic death rate, there will bea predivisional death rate. The magnitude of this death rate will give an estimate ofthe sensitivity of the process to random variability. The more sharp requirements areincluded in the design of the process, the higher the expected failure rate.In the wake of the division, as indeed in a more general way after any major tran-sition, one expects a phase of infant mortality during which the death rate of thedaughter organisms will start from an inflated level and then decrease as the screen-ing progresses. Those organisms for which the replication process has been carriedout to its end but which nevertheless are not completely fit for an autonomous exis-tence, will die. Influence of temperature on hatching rate
Fig.3d shows a striking influence of temperature on the average mortality rate duringthe 21-day long of the embryonic phase of chicks. In terms of hatching rate whichis perhaps more suggestive (but less apropriate for cross-species comparison) thereis an increase from 10—5 at 35.8 degrees to 88% at 38.1 degrees and then a fall to50% at 39.8 degrees.It can of course be argued that the temperature is an environmental parameter but thisis just a label and would not help to explain the behavior seen in Fig.3d. It is clearthat it is only through a better understanding of the manufacturing process that wecan hope to predict the shape of the mortality curve; needless to say, the temperature Temperature (degree Celsius) E m b r y o n i c m o r t a lit y r a t e ( p e r da y a n d f e r til e e gg s ) . d e g r ee s N u m b e r o f da y s f r o m f e r tli z a ti o n t o h a t c h i n g Fig. 3d Average embryonic mortality rate for chicks from fertilization to hatching.
Apart from the mor-tality rate, the graph shows also the length of the embryonic phase. On account of the fact that the speed of mostchemical reactions increases with temperature one is not surprised by the shortening of the embryonic phase.On the contrary, the fact that the mortality rate exhibits a sharp minimum requires an explanation. A good testof our understanding of this manufacturing process will be our ability to predict (at least approximately) theoptimal temperature.
Source of the data: ISA (2009). is an essential variable in this process.
Salient features of the two phases of human mortality
Our main goal in this section is to show that the curves of age-specific infant mortal-ity rates provide, so to say, a global quantitative summary of the various congenitalanomalies that appear in the embryonic phase.
Human infant mortality for all causes of death
As our starting point we consider infant death rate curves for humans as shown inFig.2a,b .Three striking features of infant mortality rates appear in Fig.4a,b but before wedescribe them in detail we wish to attract the attention of the readers on two aspects.(i) Fig.4a shows that the death rates exhibit little fluctuations. (ii) Fig.4c shows thatthe pattern of death rates remains fairly stable even when the death rate level changesconsiderably as happened between 1923 and 1960. Moreover, an examination across More details about infant mortality can be found in Berrut et al. 2016. several countries shows that these curves remain much the same in all developedcountries. As an illustration, one can look at the death rate curves for the UK shownin Berrut et al. (2016)One may think that the first point is hardly surprising because the death rate is anaverage over a large sample comprising thousands of deaths for each age interval.However, averaging alone cannot explain the absence of fluctuations as is demon-strated by the fact that weekly or monthly death rate curves show fairly large fluc-tuations. This suggests that evolution as a function of age is much more stable thanchanges in the course of (calendar) time. As a matter of fact, it will be seen in Bois etal. (2019a) that this stability is greater for young-age deaths than for old-age deaths.
110 -60 -40 -20 0 20 40 60 80
Age (day) M o r t a lit y r a t e ( p e r da y a n d p e r , li ve b i r t h s ) Fetal mortality B i r t h -1
110 1 10 10 Age (day) D e a t h r a t e ( p e r da y a n d , ) B i r t h -1 Age (day) D e a t h r a t e ( p e r ye a r a n d , ) I n f a n t m o r t a li t y ph a s e A g i n g ph a s e Fig. 4a,b Infant and adult mortality rates for humans (United States). (a) is for 1923 in the US and theinset is for the same data in log-log coordinates. Fetal mortality corresponds to the average level of late fetalmortality (6 to 9 months pregnancy). (b) extends until 30,000 days which represents 82 years. The three mainfeatures of infant mortality are the following: (i) The sharp spike at birth. (ii) The decrease of infant mortalityrate between birth and the age of 10 followed by subsequent increase. (iii) The fact that, as a function of age t , the decrease follows an hyperbolic law of the form: µ = A/t γ with γ of the order of 1. Note that despitethe huge fall of the death rate between 1923 and 1960 the structure of the two phases did not change much. In1923 γ = 0 . ± . , whereas in 1960: γ = 1 . ± . (the error bars are for a confidence level of 95%).The change in the slope from 1923 to 1960 is due to the fact that early mortality is almost time independent(because mostly due to malformations) whereas the mortality at the age of 10 has decreased considerably. Inthe interval (0 , the infant mortality rate is defined as: µ b = (1 /x )∆ x/ ∆ t , where x =number of live births, ∆ x =number of deaths in the age interval ∆ t ; this definition is standard for the interval (0 , but here we extendit to the age interval (0 , . In the expression of the adult mortality rate µ , the denominator x is replaced bythe number x ( t ) of individuals alive at the beginning of the age interval ∆ t . Actually, as long as the total infantdeaths remain under 10%, using the adult definition at all ages would not make much difference because inthis case the infant age groups are anyway close to x . A last comment is in order to say that in the presentpaper the expressions “death rate” and “mortality rate” are used as synonyms; sometimes “death” is preferredto “mortality” just because it is shorter (that is why it is used in the small inset graph). Sources: 1923 (a) Underone year: Linder et al. 1947 p.574, (b) Over one year: Linder et al. 1947, p.150 (gives in fact 1920); 1960 (a)Under one year: Grove et al. 1968, p.210-211, (b) Over one year: Grove et al. 1968, p.318. Now we describe the three salient features of the shape of the infant death ratecurves.(1) The most impressive feature is certainly the very sharp spike which coincideswith birth. It means that the death rate is high immediately after birth but decreasesrapidly in subsequent days and weeks.(2) In Fig.4a this decrease seems to level off after the age of 60 days. In fact,the decrease does not stop but simply becomes slower . This fall is described bya power law which continues until the age of 3,600 days that is to say about 10years. If one considers that the maximum life span is about T max = 100 years thiscorresponds to 10% of T max. After the age of 10 years the death rate increasessteadily and exponentially up to T max in accordance with Gompertz’s law.Although in medical language, infant mortality is understood as the first year afterbirth, in the present paper “infant mortality” refers to the whole phase during whichthe death rate decreases. This definition follows a well established usage in reliabilityscience.(3) During the infant mortality phase, the human death rate decreases in anhyperbolic way of the form: x ( t ) = A/t γ where the exponent γ is of the order of 1. The age of 10 seen as an equilibrium point between screening and wear-out
If one attributes the downward part of the mortality curve to a screening processthrough which individuals with congenital malformations are eliminated and its up-ward part to wear-out, it makes little sense to assume that the first effect stops at theage of 10 while the second starts at that age. Certainly the screening continues after10 and the wear-out starts immediately after birth. In this perspective, 10 becomesthe equilibrium point between the two effects.
Infant mortality for specific causes of death
The graphs of Fig. 4c,d show infant mortality for specific causes of death, namelyviral and bacterial diseases (of which tuberculosis was the most important instancein the early 20th century). Fig 4b, Fig. 4c,d show a broad downward trend but inaddition for specific age intervals there are peaks denoting mortality surges. In fact,these peaks are also visible on the “all causes” curves but only with poor accuracybecause they are overshadowed by the general trend of all other causes.The reason for these peaks is not yet clear but it is likely that they relate to the gradual By this expression we mean that a fall from 1,000 to 100 will take place between day 1.5 and 7, whereas from 10 to1 it will take from day 150 to day 700 (approximately). Although the distinction between power law and exponential is well known in biology it is not seen in the same wayas in physics. It is of course obvious that an exponential falls off faster than a power law, but one must realize how massivethe difference is. y = 1 /x, y = exp( − x ) : x = 10 → y = 0 . , y = 0 . This makes the two functions really different in nature. For instance, the exponential form of Gompertz’s law absolutely Age (day) D e a t h r a t e ( p e r m illi o n li ve b i r t h s a n d ye a r o f ag e ) Viral: 1973-1975Viral: 1989-1993 1010 -3 -2 -1 Age (Year) D e a t h r a t e ( p e r m illi o n li ve b i r t h s a n d ye a r o f ag e ) Tuberculosis: 1932-1936Tuberculosis: 1950-1952
Fig. 4c,d Infant mortality rates for viral diseases versus tuberculosis. Left
Although there is a generaldiminution of the death rate from 1973-1975 to 1989-1993 the peak of the first curve (in blue) has an amplitude(ratio of top rate to base rate) of 2 whereas the second curve (in red) has an amplitude of 5. The error barsgive the standard deviation of the average of individual years in the respective age intervals.
Right
Oneobserves the same phenomenon as in the graph for viral diseases, namely an overall diminution coupled witha higher peak in the more recent time interval: for 1932–1936 the peak has an amplitude of 3 whereas for1950-1952 its amplitude is 6.5. It should be noted that these peaks are also visible on the total mortality curvesbut in attenuated form which means that one needs high accuracy measurements to detect them.
Sources: VitalStatistics for the United States for the appropriate years; Berrut et al. (2017) establishment of the immune system. Shortly after birth the newborn is protectedby the antibodies contained in the breast milk of the mother but this protection isgradually replaced by the child’s own immune system. Moreover, the immunityprovided by the mother first during pregnancy and then shortly after birth dependson the diseases that the immune system of the mother had to face.In other words, these surges in infant mortality can tell us something about specialevents in infant development that would not be visible otherwise.
Conclusion
Main results
The considerable variety of birth defects, whether lethal or non-lethal, attests thatcontrol mechanisms can be overwhelmed in many ways. However, the relatively lowfrequency of each of these defects (mostly under 1 per 1,000) attests that most of thetime the “manufacturing process” works fairly well. forbids anybody to reach the age of 130 years. Defined as: µ b = (1 /x )∆ x/ ∆ t , where x =number of life births, ∆ x =number of deaths in the age interval ∆ t . In this paper we have introduced the idea of a third source of congenital anomaliesbesides the genetic and environmental factors. It was called “manufacturing disper-sion” because it consists in the accumulation of small output defects in the successivesteps of a development process. Such a mechanism was shown to be responsible of asubstantial variability even with the two other factors are inactive. This would solvethe mystery of the large proportion of defects for which no specific source can beidentified (as noted at the beginning of the paper).We have described a number of circumstances which are likely to amplify outputdispersion: complex organs, processes which require perfect synchronization in timeand space, rapid and drastic transformations.Clearly one would like to get a better understanding of the basic mechanisms ofmanufacturing dispersion. It is for that purpose that in a forthcoming paper (Bois etal. 2019b) we propose two simple physical models which provide a clearer insightthan in vivo biological organisms..
Rationale for cross species comparisons
The dispersion hypothesis led to the prediction that “simple” organisms should haveless lethal congenital anomalies than complex organisms like mammals. As an illus-tration consider the following example.In humans, within a few days after birth, heart and lungs defects are the main causesof death (see Fig.5); lung problems are particularly critical for preterm newborn.In contrast, for rotifers these two causes are completely non-existent for the simplereason that rotifers have neither heart nor lungs. Because of their size (about 0.2mmin length and 0.03mm in diameter) rotifers, like all other aquatic organisms of similarsize or smaller, receive their oxygen by diffusion through their skin. There is ofcourse a similar diffusion process for larger animals but whereas the concentrationjump, ∆ c , is the same, the skin thickness, ∆ x , may be 100 times larger, thus givinga diffusion gradient, ∆ c/ ∆ x some 100 times smaller. Size also makes blood uselessbecause oxygen can be brought by diffusion to all parts of the body.In short, for rotifers one does not expect the kind of sharp peak immediately afterbirth as observed for humans. Is there nevertheless an infant mortality phase duringwhich the death rate decreases? Only observation can tell us. That is why rotifermortality will be studied in a companion paper (Bois et al. 2019a).Incidentally, it can be observed that the diffusion mechanism works not only formicroscopic organisms but also for centimeter-size organisms on the condition thatthey are formed of thin layers. That is the case for: (i) sponges consisting of a singlecell layer or (ii) jelly fish whose body is a layer not more than a few cells thick. Inall these organisms gases, nutrients, and wastes are exchanged by diffusion. Thus,2
The dispersion hypothesis led to the prediction that “simple” organisms should haveless lethal congenital anomalies than complex organisms like mammals. As an illus-tration consider the following example.In humans, within a few days after birth, heart and lungs defects are the main causesof death (see Fig.5); lung problems are particularly critical for preterm newborn.In contrast, for rotifers these two causes are completely non-existent for the simplereason that rotifers have neither heart nor lungs. Because of their size (about 0.2mmin length and 0.03mm in diameter) rotifers, like all other aquatic organisms of similarsize or smaller, receive their oxygen by diffusion through their skin. There is ofcourse a similar diffusion process for larger animals but whereas the concentrationjump, ∆ c , is the same, the skin thickness, ∆ x , may be 100 times larger, thus givinga diffusion gradient, ∆ c/ ∆ x some 100 times smaller. Size also makes blood uselessbecause oxygen can be brought by diffusion to all parts of the body.In short, for rotifers one does not expect the kind of sharp peak immediately afterbirth as observed for humans. Is there nevertheless an infant mortality phase duringwhich the death rate decreases? Only observation can tell us. That is why rotifermortality will be studied in a companion paper (Bois et al. 2019a).Incidentally, it can be observed that the diffusion mechanism works not only formicroscopic organisms but also for centimeter-size organisms on the condition thatthey are formed of thin layers. That is the case for: (i) sponges consisting of a singlecell layer or (ii) jelly fish whose body is a layer not more than a few cells thick. Inall these organisms gases, nutrients, and wastes are exchanged by diffusion. Thus,2 Perinatal respiratory and cardiovascular disorders (P20-29) -3 -2 -1 -3 -2 -1 D e a t h r a t e ( p e r ye a r a n d m illi o n pop u l a ti o n ) slope=-1.56
20 40 60 80
Age (year)
Fig. 5 Infant and adult mortality due to lung and heart congenital abnormalities.
There are two notewor-thy features. (i) The high slope of 1.56 is associated with a high initial mortality rate (in the first year it is 168times higher than for spina bifida) and reveals a drastic screening process. (ii) As a result, this cause of death isnearly eliminated which explains that the adult death rate is not increasing with age but instead fluctuates moreor less randomly at a very low level of less than 100 annual deaths.
Source: CDC Wonder: Detailed mortality1999-2017. as a conjecture, one would expect their infant mortality curve to start similarly as theone of rotifers.More broadly, it is in order to test such predictions that we started a research programconsisting in the measurement of infant mortality across species.
Appendix A. Estimating the strength of genetic factors
It is probably not far from the truth to say that nowadays some 90% of the researchpapers in biology are to some extent focused on genetics. This is surprising because,as explained in a review paper published in the “New York Times” (Kolata 2006), de-mographic and epidemiological research shows that for most human characteristics(e.g. lifespan or diseases) there is only a very loose genetic influence.Here we are interested in birth defects. Because they are not affected by all lifeincidents (which differ from person to person) one may think that there is a firmerground for genetic influence. Currently, it seems to be a well accepted axiom thatmost malformations have a genetic origin. At least this is the implication of paperslike the study by Ahmed et al. (2017) which, for all separate variants of fingermalformations, lists the genes which seem responsible. Under such an assumption,monozygotic twins should have the same malformations. We will see below that thisis far from true. Before focusing on the twin methodology let us briefly examine some aspects ofharmful mutations leading to anomalies.
Mutations and repair mechanisms
In a living organism harmful mutations can occur at three levels. (i) Germ cells.(ii) Stem cells, i.e. cells no yet differentiated into specific organ types. (iii) Fullyfunctional differentiated cells existing in various organs. The last two types are calledsomatic mutations for they are not passed on to children.In a long term perspective the most serious cause of concern are of course the germcell mutations because, unless there is a repair mechanism, they will be passed overfrom generation to generation and will accumulate . So, the existence of effectiverepair mechanisms has been a natural assumption among biologists long before itwas eventually demonstrated in a work honored by a Nobel award in 2015. If thereare repair mechanisms it means that the static picture with a rigid connection betweendefective genes and abnormalities must be replaced by a dynamic vision. The twin methodology for assessing the strength of genetic factors
A methodology based on twin data which permits to ascertain the role of geneticfactors in the occurrence of malformations (or more generally of any disease or trait)has been developed by several authors, e.g. Hrubec et al. (1981) and Tishler et al.(2007). However, as the method is used differently in each specific application, wesummarize in this appendix the variables and reasoning which are most convenientfor our purpose.Before giving a formalized presentation for a large sample of twin pairs it may beuseful to describe a specific case consisting in the occurrence of breast cancer inmonozygotic twins. A team of Czech researchers followed 5 monozygotic pairsof twins over a long time period of up to two decades. They made the followingobservations (Hlad`ıkov´a et al. 2013). • Pair 1=(breast cancer at age 54 versus ovarian cancer at age 43) • Pairs 2,3,4,5=(breast cancer at a median age of 44 versus no cancer)The authors conclude that “environmental factors play an important role in breastcancer development”. Instead of mysterious “environmental factors” such outcomescan also result from a random dispersion of manufacturing outputs.Next, we consider this problem in a more general way.The starting point is a dataset for a sample comprising M monozygotic (MZ) twinsand D dizygotic (DZ) twins. Secondly, one focuses on the frequency of a specific There may be many external mutation factors but one that has existed without any doubt since the beginning of lifeon Earth consists in high energy cosmic rays. congenital malformation. This leads to define and compute the following variables. • Concordant pairs, i.e. pairs in which both twins have the malformation; wedenote their number by c m and c d respectively for MZ and DZ twins. • “Discordant” pairs, i.e. pairs in which one child has the malformation but notthe other; we denote their number by d m and d d respectively for MZ and DZ twins.In addition, we denote the probability of the malformation in the general populationby p . A typical order of magnitude for p is 1 per 1,000 that is to say: p = 10 − .Ideally, for a malformation that is 100% genetically determined, among MZ twinsthere should be no discordant pairs, i.e. d m = 0 . Thus, if we introduce the ratio g m = c m / ( c m + d m ) it will be equal to 1.In contrast, for DZ twins there may be some discordant pairs, i.e. d d > . Thus, for g d = c d / ( c d + d d ) one gets: g d < , in other words: g d < g m ; this last inequalityis also expected to hold at least approximately for malformations in which geneticdetermination is less than 100%.For a malformation which has no genetic basis at all, the probability for both twinsto have it would be p , whereas the probability for only one having it would be: p (1 − p ) ; as usually p is of the order of one per thousand the factor − p can beapproximated by 1.Thus, c m = M p , d m = M p → g m ≃ p / ( p + p ) = p/ ( p + 1) ≃ p Naturally, in this case the expectations for DZ twins are the same as for MZ twins.In short, the strength of genetic factors can be estimated in two ways:(i) How close is g m to 1? It turns out that for most congenital malformations g m is smaller than . . In the previous cancer example, c m = 0 because even for pair 1there are different cancers , thus g m = 0 .(ii) How much is g m larger than g d ? This can be expressed by the ratio: g ′ = g m /g d . In the cancer example: g ′ = 0 .These conclusions are summarized in Table A1a.Inserting the values of c m , d m , c d , d d given in Yu et al (2019, Table 2) one gets theresults shown in Table A1b.The estimates show that for all malformations the strength of genetic factors is farfrom 100%; in other words there is room for other factors than heredity particularlyfor environmental factors and output dispersion. According to the g m criterion, thestrength of genetic factors rank as follows (from high to low): oral cleft, club foot, If one is only interested in whether there is cancer or not then c m = 1 and g m = 1 / . . Table A1a: Twin variables for estimating the strength of genetic factors in malformation occurrences.
MZ MZ MZ DZ DZ DZ MZ/DZConcord. Discord. Ratio Concord. Discord Ratiopairs pairs pairs pairs c m d m g m c d d d g d g ′ = g m /g d c m d m = 0 g m = 1 c d d d > g d < g m g ′ >
0% genetic
N p N p g m = p N p N p g d = p g ′ ≃ Notes: N is the population of the sample. MZ means monozygotic (true twins) and it corresponds to the index m . DZ means dizygotic and it corresponds to the index d . “Concord.” means “Concordant” (corresponds to thevariable c ). “Discord.” means “Discordant” (corresponds to the variable d ). As an example of the notations,the variable c d represents “concordant pairs of dizygotic twins. g m and g d have the following definitions: g m = c m / ( c m + d m ) , g d = c d / ( c d + d d ) . p is the probability of the malformation in the general population;it is assumed that p ≪ (usually p ≃ − ). High strength of genetic factors is associated with g m close to 1and g ′ higher than 1, whereas low strength is associated with g m much smaller than 1 and g ′ close to 1. Table A1b: Estimates of the strength of genetic factors in malformation occurrences.
Birth p MZ MZ MZ DZ DZ DZ MZ/DZdefect per Concord. Discord. Ratio Concord. Discord Ratio , pairs pairs pairs pairs c m d m g m = c d d d g d = g ′ = g m /g d c m c m + d m c d c d + d d g ′ = g m /g d Oral cleft .
3% 4 . Spina bifida .
9% 0 33 0% − Club foot .
6% 4 . Strabism
18 33 161 17% 27 412 6 .
5% 2 . Average . .
7% 4 .
10% 3 . Notes: Although for oral cleft and spina bifida the numbers of cases are somewhat too small the fact that amongMZ pairs there are much more discordant pairs than concordant pairs (which translates in a value of g m muchlower than 1) shows a loose genetic determination. The results for g ′ are only significant for strabismus; for theother defects there are too few DZ cases. Sources: The data are for 6,752 monozygotic twin pairs and 13,310 dizygotic twin pairs from the Californiatwin program covering 1957–1982 (Yu et al. 2019) . strabismus, spina bifida; according to the g ′ criterion the ranking is: club foot, oralcleft, strabismus (not defined for spina bifida).In Table A1b we see that: (i) g m > p , (ii) g d < g m and (iii) g ′ > which suggests thatgenetic factors play a role in the malformations. However, the fact that on averagefor the 4 malformations g m = 0 . ± . which is well below 1 show that genetic determination is rather weak. In other words, other factors may be at work. Strength of genetic factors in cancer
So far, we have examined birth defects. Although it is at birth that these defectsbecome visible, in fact they appear earlier during pregnancy. On the contrary, cancerappears late in the course of life. Therefore, one can expect important contributionsof somatic mutations and environmental factors. It is for the purpose of comparisonthat we study this case.
Table A1c: Estimates of the strength of genetic factors in cancer.
MZ MZ MZ DZ DZ DZ MZ/DZType of Concord. Discord. Ratio Concord. Discord Ratiocancer pairs pairs pairs pairs c m d m g m (%) c d d d g d (%) g ′ = g m /g d Specific cancers
Lung .
0% 3 112 2 .
6% 0 . Stomach .
9% 4 138 2 .
8% 0 . Colon .
6% 13 191 4 .
3% 1 . Breast
22 257 7 .
9% 23 467 4 .
7% 1 . Cervix
30 242 11 .
0% 27 412 5 .
1% 2 . Prostate
19 137 12 .
2% 7 299 2 .
3% 5 . Average .
8% 3 .
6% 2 . All cancers
182 1306 12 .
2% 257 2351 3 .
6% 1 . Notes: The variables c m , c d , d m , d d , g ′ are defined in the text. “Concord” means “Concordant” (i.e. samedisease in each twin of a pair); “Discord” means “Discordant”. In the “Specific cancers” cases “concordant”means the same specific kind of cancer whereas in the “All cancers” row “concordant” means “any kind ofcancer”. The cancers are ranked by order of increasing values of g ′ , that is to say increasing strength of geneticfactors. The “All cancers” row includes more cases than the 6 types listed in the table. Sources: The data are for 23,386 twin pairs from the “Swedish Twin Registry” covering the years 1959–1961and 1970–1972 (Ahlbom et al. 1997) . Table A1c gives estimates for the strength of genetic factors in cancer. Whether ornot cancer can be seen as resulting from a congenital defect of the immune system isa matter of perspective. On average the estimates show that the genetic componentis weaker than for the malformations given in Table A1b.When the concordance of monozygotic and dizygotic twin pairs are approximatelyof same value, i.e. g ′ ∼ , it suggests a small influence of genetic factors. In such asituation one must check if this common value is higher than what would be expectedon a purely random basis.For all cancers except cervix and prostate cancer, on account of g m ≃ g d there islittle genetic influence. As the prevalence for all cancers is about p = 6% in the population over 15, Table A1c shows that g m = 12 . is (slightly) higher than therandom threshold of . This suggests that family similarities may play a role, e.g.obesity, stress due to living or working conditions and so on. Acknowledgments
The authors express their gratitude to Drs. Kun Wang and LucWestphal for their help and interest.
References
Ahlbom (A.), Lichtenstein (P.), Malmstr ¨om (H.), Feychting (F.), Hemminki (K.),Pedersen (N.L.) 1997: Cancer in twins: genetic and nongenetic familial riskfactors. Journal of the National Cancer Institute 89,4,287-293.Alix (M.) 2016: Etude de la variabilit´e de l’embryog´en`ese chez la perche com-mune: d´eveloppement d’approches alternatives. [Embryogenesis of the Eu-ropean perch (Perca fluviatilis): alternative approaches.] PhD Thesis presentedat the University of Lorraine on 15 December 2008.Berrut (S.), Pouillard (V.), Richmond (P.), Roehner (B.M.) 2016: Deciphering infantmortality. Physica A 463,400-426.Berrut (S.), Richmond (P.), Roehner (B.M.) 2017: Age spectrometry of infant deathrates as a probe of immunity: Identification of two peaks due to viral and bac-terial diseases respectively. Physica A 486,915-924.Biobaku (K.T.), ADELEYE (O.E.) 2010: Two hens mutually brooding: a rare be-haviour in
Gallus domesticus . Science World Journal 5,3,21-22.Boas (H.M.) 1918. Inheritance of eye color in man. American Journal of PhysicalAnthropology 2,15-20.Bois (A.), Garcia-Roger (E.M.), Hong (E.), Hutzler (S.), Ali Irannezhad (A.), Man-nioui (A.), Richmond (P.), Roehner (B.M.), Tronche (S.) 2019a: Infant mortal-ity across species. A global probe of congenital abnormalities. Preprint April2019.Bois (A.), Garcia-Roger (E.M.), Hong (E.), Hutzler (S.), Ali Irannezhad (A.), Man-nioui (A.), Richmond (P.), Roehner (B.M.), Tronche (S.) 2019b: Physical mod-els of infant mortality. Implications for biological systems. Preprint June 2019.Burdett (I.D.J.), Kirkwood (T.B.L.), Whalley (J.B.) 1986: Growth kinetics of indi-vidual
Bacillus subtilis cells and correlation with nucleoid extension. Journalof Bacteriology 167,1,219-230.Chen (J.), Huang (X.), Wang (B.), Zhang (Y.), Rongkavilit (C.), Zeng (D.), Jiang (Y.),Wei (B.), Sanjay (C.), McGrath (E.) 2018: Epidemiology of birth defects basedon surveillance data from 2011-2015 in Guangxi, China: comparison acrossfive major ethnic groups. BMC [Bio-Med Central] Public Health 18,1008. Crymes (W.B.), Zhang (D.), Ely (B.) 1999: Regulation of podJ expression during the
Caulobacter crescentus cell cycle. Journal of Bacteriology 181,13,3967-3973.Danielsen (R.), Aspelund (T.), Harris (T.B.), Gudnason (V.) 2014: The prevalence ofaortic stenosis in the elderly in Iceland and predictions for the coming decades:The AGES-Reykjavk study. International Journal of Cardiology 176,3,916-922.Depree (J.A.), Geoffroy (P.S.) 2001: Physical and flavor stability of mayonnaise.Trends in Food Science and Technology 12,5,157-163.Drummond (D.A.), Wilke (C.O.) 2009: The evolutionary consequences of erroneousprotein synthesis. Nature Reviews Genetics 10,10,715-724.Dubrova (E.) 2013: Fault-tolerant design. Springer, Berlin.Fairchild (B.D.), Christensen (V.L.), Grimes (J.L.), Wineland (M.J.), Bagley (L.G.)2002: Hen age relationship with embryonic mortality and fertility in commer-cial turkeys. The Journal of Applied Poultry Research 11,3,260-265.Feldkamp (M.L.), Carey (J.C.), Byrne (J.L.B.), Sergey Krikov (S.) Botto (L.D.)2017: Etiology and clinical presentation of birth defects: population basedstudy. British Medical Journal 357,j2249.French (F.E.), Bierman (J.M.) 1962: Probabilities of fetal mortality. Public HealthReports 77,10,835-847.Garcia-Roger (E.M.), Mart´ınez (A.), Serra (M.) 2006: Starvation tolerance of rotifersproduced from parthenogenetic eggs and from diapausing eggs: a life tableapproach. Journal of Plankton Research 28,3,257-265.Gerhard (G.S.), Kauffman (E.J.), Wang (X.), Stewart (R.), Moore (J.L.), Kasales(C.), Demidenko (E.), Cheng (K.C.) 2002: Life spans and senescent phenotypeson two strains of Zebrafish (Danio rerio). Experimental Gerontology 37,1055-1068.Gopakumar (G.), Jayaprakas (V.) 2004: Life table parameters of
Brachionus pli-catilis and
B. rotundiformis in relation to salinity and temperature. Journal ofthe Marine Biological Association of India 46,1,21-31.Greenough (R.B.) 1925: Varying degrees of malignancy in cancer of the breast.Journal of Cancer Research, 1925,453-463.Grove (R.D.), Hetzel (A.M.) 1968: Vital statistics rates in the United States, 19401960.United States Printing Office, Washington, DC.Gunton (K.B.), Wasserman (B.N.), DeBenedictis (C.) 2015: Strabismus. Primarycare 42,3,393-407.Hirschmann (W.B.) 1964: Profit from the learning curve. Harvard Business Review,42,1,125-139.Hlad`ıkov´a (A.), Plevov´a (P.), Mach´a˘ckov´a (E.) 2013: Breast cancer in monozygotic twins [in Czech]. Klin Onkol. 26,3,213-217.Hrubec (Z.), Neel (J.V.) 1981: Familial factors in early deaths: twins followed 30years to ages 51-61 in 1978. Human Genetics 59,1,39-46.Hutt (F.B.) 1929: Studies in embryonic mortality in the fowl. I. The frequency ofvarious malpositions of the chick embryo and their significance. Proceedingsof the Royal Society of Edinburgh 49,II,118-130.ISA (Institut de S´election Animale) 2009: From egg to chicken. Hatchery manual.Boxmeer (Netherlands). [A 46-pages handbook published by ISA which is asubsidiary of the “Hendrix Genetics Company”.]Jennings (H.S.) 1916: Heredity, variation and the results of selection in the uni-parental reproduction of Difflugia corona . Genetics 1,407-534.Jian (X.), Tang (X.), Xu (N.), Sha (J.) YouWang (Y.) 2017: Responses of the ro-tifer Brachionus plicatilis to flame retardant (BDE-47) stress. Marine PollutionBulletin 116,1-2,298-306.Joe (B.B.) 2004: Growth response of
Euglena gracilis and
Selenastrum capricornu-tum in response to pH. Semantic Scholar (22 March 2004).Johannsen (W.) 1903: Ueber Erblichkeit in Populationen und reinen Linien. EinBeitrag zur Beleuchtung schwebender Selektionsfragen. [About heredity inpopulations and pure lines. A contribution to the understanding of pendingquestions about selection.] Gustav Fisher, Jena.Johnston (R.K.), Snell (T.W.) 2016: Moderately lower temperatures greatly extendthe lifespan of Brachionus manjavacas (Rotifera): Thermodynamics or generegulation? Experimental Gerontology 78,12-22.Joshi (P.S.) 1988a: Influence of salinity on population growth of a rotifer,
Brachionusplicatilis . Journal of the Indian Fisheries Association 18,75-81.Joshi (P.S.) 1988b: Mass culture of
Brachionus plicatilis . Master of Science Disser-tation, University of Bombay (84 p.) [available on line]Kioumourtzoglou (M.-A.), Coull (B.A.), O’Reilly (E.J.), Ascherio (A.), Weisskopf(M.G.) 2018: Association of exposure to diethylstilbestrol [DES] during preg-nancy with multigenerational neurodevelopmental deficits. Journal of the Amer-ican Medical Association (JAMA), Pediatrics. 172,7,670-677.Kobitski (A.Y.), Otte (J.C.), Takamiya (M.), Sch¨afer (B.), Mertes (J.), Stegmaier (J.),Rastegar (S.), Rindone (F.), Hartmann (V.), Stotzka (R.), Garc´ıa (A.), Wezel (J.van), Mikut (R.), Str¨ahle (U.), Nienhaus (G.U.) 2015: An ensemble-averaged,cell density-based digital model of zebrafish embryo development derived fromlight-sheet microscopy data with single-cell resolution. Nature Scientific Re-ports 5,8601. Knight (K.) 2016: Zebrafish larvae learn to hunt using lateral line in the dark. Journalof Experimental Biology 219,465-466.Kohler (I.V.), Preston (S.H.), Lackey (L.B.) 2006: Comparative mortality levelsamong selected species of captive animals. Demographic Research 15,14,413-434.Kolata (G.) 2006: Live long? Die young? Answer isn’t just in genes. New YorkTimes 31 August 2006.Lashley (K.S.) 1915: Inheritance in the asexual reproduction of hydra. The Journalof Experimental Zoology 19,157-210.Laub (M.T.), McAdams (H.H.), Feldblyum (T.), Fraser (C.M.), Shapiro (L.) 2000:Global analysis of the genetic network controlling a bacterial cell cycle. Science290,2144-2154.Linder (F.E.), Grove (R.D.) 1947: Vital statistics rates in the United States,1900-1940. United States Printing Office, Washington, DC, 1947.Middleton (A.R.) 1915a: Heritable variations and the results of selection in the fis-sion rate of
Stylonychia pustulata.
Journal of Experimental Zoology 19,451-503.Middleton (A.R.) 1915b: Heritable variations and the results of selection in the fis-sion rate of
Stylonychia pustulata.
Proceedings of the National Academy ofSciences, 4 November 1915.[This paper is a summary of the previous one.]Noyes (B.) 1923: Experimental studies on the life history of a rotifer reproduc-ing parthenogenetically (
Proales decipiens ). Journal of Experimental Zoology35,2,225-255.[It is a thesis submitted to John Hopkins University, Baltimore.]Papoulis (A.) 1965: Probability, random variables, and stochastic processes. McGraw-Hill, Tokyo.Patey (D.H.), Scarff (R.W.) 1928: The position of histology in the prognosis of car-cinoma of the breast. The Lancet 211,5460,801-804.Pearl (R.), Miner (J.R.) 1935: Experimental studies on the duration of life. XIV.The comparative mortality of certain lower organisms. The Quarterly Reviewof biology 10,1,60-79.[The paper contains data for Hydra fusca,Pearl (R.) Park (T.), Miner (J.R.) 1941: Experimental studies on the duration of life.XVI Life tables for the flour beetle
Triboleum confusum Duval . The AmericanNaturalist 75,756,5-19.Pe ˜nuela (A.), Hernandez (A.) 2018: Characterization of embryonic mortality in broilers. Revista MVZ (Medecina, Vetenaria,Zootecnia) C ´ordoba 23,1,6500-6513.Pouillard (V.) 2015: En captivit´e. Vies animales et politiques humaines dans lesjardins zoologiques du XIXe si`ecle `a nos jours : m´enagerie du Jardin desPlantes, zoos de Londres et Anvers. [In captivity. Zoo management and ani-mal lives in the zoological gardens of Paris, London and Antwerp from the 19thcentury to 2014.]. PhD thesis. Universit´e libre de Bruxelles and University ofLyon 3.Romanoff (A.) 1949: Critical periods and causes of death in avian embryonic devel-opment. The Auk 66,3,264-270.Rombough (P.J.) 1998: Partitioning of oxygen uptake between the gills and skin infish larvae: a novel method for estimating cutaneous oxygen uptake. Journal ofExperimental Biology 201,11,1763-1769.Rideout (R.M.), Trippel (E.A.), Litvak (M.K.) 2004: Predicting haddock embryoviability based on early cleavage patterns. Aquaculture 230,215-228.Sahin (T.) 2001: Larval rearing of the Black Sea Turbot, Scophthalmus maximus (Linnaeus, 1758), under laboratory conditions. Turkish Journal of Zoology25,447-452.Schaefer (B.M.), Lewin (M.B.), Stout (K.K.), Gill (E.), Prueitt (A.), Byers (P.H.),Otto (C.M.) 2007: The bicuspid aortic valve: an integrated phenotypic classi-fication of leaflet morphology and aortic root shape. British Medical JournalHeart 94,12.Shylakhovenko (V.A.), Olishevsky (S.V.), Kozak (V.V.), Yanish (Y.V.), Rybalko (S.L.)2003: Anticancer and immunostimulatory effects of nucleoprotein fraction of
Bacillus subtilis . Experimental Oncology. 25,119-123.Stocking (R.J.) 1915a: Variation and inheritance of abnormalities occurring afterconjugation in
Paramecium caudatum.
Journal of Experimental Zoology 19,387-449.Stocking (R.J.) 1915b: Variation and inheritance in abnormalities occurring afterconjugation in
Paramecium caudatum.
Proceedings of the National Academyof Sciences, 4 November 1915.[This paper is a summary of the previous one.]Sun (Y.), Hou (X.), Xue (X.), Zhang (L.), Zhu (X.), Huang (Y.), Chen (Y.), Yang (Z.)2017: Trade-off between reproduction and lifespan of the rotifer Brachionusplicatilis under different food conditions. Scientific Reports 7,1.Tishler (P.V.), Carey (V.J.) 2007: Can comparison of MZ- and DZ-twin concordancerates be used invariably to estimate heritability? Twin Research and Human2