Adaptive-filtering of trisomy 21: risk of Down syndrome depends on

Naturwissenschaften
DOI 10.1007/s00114-006-0165-3
SHORT COMMUNICATION
Adaptive-filtering of trisomy 21: risk of Down syndrome
depends on family size and age of previous child
Markus Neuhäuser & Sven Krackow
Received: 27 April 2006 / Revised: 4 July 2006 / Accepted: 14 July 2006
# Springer-Verlag 2006
Abstract The neonatal incidence rate of Down syndrome
(DS) is well-known to accelerate strongly with maternal
age. This non-linearity renders mere accumulation of
defects at recombination during prolonged first meiotic
prophase implausible as an explanation for DS rate increase
with maternal age, but might be anticipated from chromosomal drive (CD) for trisomy 21. Alternatively, as there is
selection against genetically disadvantaged embryos, the
screening system that eliminates embryos with trisomy 21
might decay with maternal age. In this paper, we provide
the first evidence for relaxed filtering stringency (RFS) to
represent an adaptive maternal response that could explain
accelerating DS rates with maternal age. Using historical
data, we show that the proportion of aberrant live births
decrease with increased family size in older mothers, that
inter-birth intervals are longer before affected neonates than
before normal ones, and that primiparae exhibit elevated
levels of DS incidence at higher age. These findings are
predicted by adaptive RFS but cannot be explained by the
currently available alternative non-adaptive hypotheses,
including CD. The identification of the relaxation control
mechanism and therapeutic restoration of a stringent screen
may have considerable medical implications.
M. Neuhäuser (*)
Department of Mathematics and Technique,
RheinAhrCampus Remagen,
Südallee 2,
53424 Remagen, Germany
e-mail: [email protected]
S. Krackow
Institute for Biology, Humboldt University Berlin,
Invalidenstrasse 43,
10115 Berlin, Germany
Keywords In utero screening . Aberrant embryos .
Spontaneous abortion . Maternal age . Menopause .
Chromosomal drive
Introduction
Down syndrome (DS) corresponds to the phenotype
produced by chromosome 21 trisomy. Its incidence is
well-known to increase with strongly accelerating rate
when mothers approach menopause (Penrose 1934; Hook
1981; Morris et al. 2002, 2005). Contrary to explanations
involving accumulation of defects at recombination during
prolonged first meiotic prophase (Lamb et al. 2005), such
dynamics may be anticipated from chromosomal drive
(CD) for trisomy 21 (Day and Taylor 1998). Given strong
evidence for uterine selection against genetically disadvantaged embryos (Warburton et al. 1983; Stein et al. 1986),
relaxed filtering stringency (RFS) has been proposed as an
alternative or additional explanation for the accelerating
increase of DS rate with maternal age (Kloss and Nesse
1992; Stein et al. 1986; Forbes 1997). While RFS has been
suggested to possibly represent an adaptive maternal
strategy (Kloss and Nesse 1992; Forbes 1997), no decisive
evidence has been given yet (Krackow 1998). In this paper,
we provide the first evidence for RFS to represent an
adaptive maternal response rather than a by-product of
some non-adaptive attrition of maternal reproductive organ
function, which has important consequences for aberrant
fetal development prevention.
Genetically aberrant phenotypes are expected to have
severely impaired reproductive prospects, and discarded
embryos can be expected to be replaced by normal embryos
at the subsequent reproductive attempt. Screening for
genetic aberration and discarding aberrant embryos would
Naturwissenschaften
therefore be adaptive, from the maternal perspective, as long
as the fitness cost of delayed reproduction is outweighed by
the benefit of differentially higher fitness in normal embryos
(cf., Kozlowski and Stearns 1989). However, the fitness
differential reduces with decreased probability of further
reproduction, e.g., with approaching menopause in human
females. The stringency of filtering is therefore expected to
relax the more, the higher the risk of future infertility (Kloss
and Nesse 1992). Given that the probability function of
reproductive incapability increases with accelerating rate
with age (e.g., following a logistic function), filtering
stringency is expected to decrease with accelerating rate as
well. Accelerating rates of DS could therefore be explained
in terms of adaptive RFS as well as in terms of CD.
However, adaptive filtering involves further trade-offs
that lead to predictions not implied by non-adaptive
mechanisms (Krackow 1998). We tested two of these
predictions using data given in two previously published,
large-scale studies (Øster 1953; Hay and Barbano 1972).
We assume that expected fitness of an embryo identified as
aberrant is greater than zero, either due to some, albeit
reduced, reproductive success of aberrant offspring, or due
to the screening mechanism being error-prone, i.e., allowing
for false positives. In that case, first, at any given age of the
mother, fitness loss due to discarding a pregnancy will be
lower the larger the number of previous children on the
average. Hence, the stringency of the filter is expected to
relax less with age at larger family size because the
increment of overall fitness is lower for a DS child in large
than in small sibships, under the assumption that the age of
mother is the most decisive determinant of future fertility
rather than the previous family size.
Secondly, investment into a current offspring detracts
from investment into previous ones, which weighs higher for
offspring with impaired fitness prospects. Since competition
for parental investment decreases with age discrepancy
between siblings (Hobcraft et al. 1983; Koenig et al. 1990),
adaptive filtering is expected to be more stringent the
younger the youngest child in the family.
Materials and methods
We reviewed the literature in order to find data that may
help to support or refute the predictions mentioned above.
We used relatively old data sets (Øster 1953; Hay and
Barbano 1972) that have the advantage that any confounding impact of selective abortion after prenatal diagnosis of
malformations can be excluded, as data have been gathered
before widespread (or any) use of such measures. Note that
we do not have access to raw data but to the data published
in the tables and appendices in Øster (1953) and Hay and
Barbano (1972), respectively.
The statistical analysis was performed using SAS
version 8.2 (SAS Institute, Cary, NC, USA) and StatXact
version 6 (Cytel, Cambridge, MA, USA). The Cochran–
Armitage test with equally spaced scores was used to test
for trends in proportions given in Table 2 of Hay and
Barbano (1972, p. 273). The time among pregnancies was
given by Øster (1953, Appendix 4) in intervals. Central
categorical values from these monthly cumulated frequency
data were used to compute means and standard errors (SE)
of inter-birth intervals (IBIs) and to carry out Wilcoxon
rank sum tests (Table 2).
Longer IBIs before DS children might not only result
from RFS but, conversely, from mothers after (randomly)
longer IBIs being older, on the average, than those after
short IBIs and, therefore, bearing a greater risk of DS
because the incidence rate of DS strongly accelerates with
maternal age. To investigate whether this confounding
effect can explain the longer IBIs before DS births in
Table 2, a simulation study was performed (using SAS). In
this simulation, 10,000,000 births were simulated for each
parity, i.e., for each row of Table 2. The maternal ages were
chosen according to the distribution of maternal age for all
births (Øster 1953, Table 22). Within the given 5-year
intervals, the maternal age were simulated as uniformly
distributed. The IBI was simulated as normally distributed
with the observed means and variances, i.e., those of the
normal IBIs of the particular parity given in Table 2. The
risk of DS was simulated with the binomial proportion of
0.000627+exp(−16.2395+0.286×maternal age) (Cuckle et
al. 1987; Morris et al. 2003). With this incidence, more than
40,000 DS cases resulted for each parity. The medians and
means of these simulated DS cases are also given in
Table 2.
Overall DS incidence rates per age group were calculated
from Table 22 in Øster (1953). Incidence rates of DS for
primiparae could be estimated relative to the overall
incidence rate within the youngest age group (<30 years of
age; Fig. 1) from the number of DS children of primiparous
women per age group and the distribution of primiparae per
age group in Denmark (Table 33 in Øster, 1953). Note that
we pooled the age classes up to 29 years in order to have a
relatively large reference group. The Mantel–Haenszel test
based on the variance estimator proposed by Robins et al.
(1986) was used to compare the two groups.
Results
In accordance with the predictions, DS incidence rates
significantly decreased with family size in older mothers
(Table 1), e.g., the decreasing trend of DS rate with
previous family size (reflected by birth order) was apparent
and significant in the two highest age categories (Table 1).
Naturwissenschaften
30
25
24.1
p = 0.0003
all births
primiparae
20
shorter for all higher parities. Hence, higher DS risk after
longer IBI due to increased maternal age cannot account for
the effect of DS incidence on prior IBI.
Discussion
16.2
15
7.7
10
5
1
1.7
5.3
1.3
0
<30
30-34
35-39
40+
maternal age (years)
Fig 1 Relative DS incidence rates in different age categories when
rate in category ‘<30 years’ is set to 1 (cf., Materials and methods),
based on 94 DS cases for primiparae (<30 years: 58 DS cases; 30–34:
11; 35–39: 17; and 40+: eight DS cases)
Also, the relative incidence rate increased more steeply
in primiparae than the population average (Fig. 1). Hence,
DS incidence rate increased more strongly with age in
firstborn children than others (p=0.0003, Mantel–Haenszel
test).
Moreover, DS births were significantly spaced further
apart from the previous child than normal births (Table 2).
For each parity, normal IBI was significantly smaller than
the IBI preceding a DS birth with the exception of the
highest parity class where power was low due to small
sample size. A significant overall difference also existed for
IBI before DS child if only data were used where the DS
child does not represent the last child in the family (not
shown).
However, when comparing the observed median IBIs
before a DS child with the simulated IBIs before a DS child
without adaptive RFS, the observed IBIs are not longer for
parities 2 and 3. In contrast, the simulated IBIs are much
The results of our re-analyses of large historical data sets on
live-birth Down syndrome (DS) incidence rates corroborate
our predictions from the adaptive relaxed filtering hypothesis (RFS; cf., Introduction). First, DS incidence significantly decreased with previous family size in older mothers,
as indicated by the decreasing trend with parity in the two
oldest age categories given in Table 1. Higher DS rates with
smaller previous family size at a given age was also
apparent in our population-case comparison for primiparae
DS rates depicted in Fig. 1, e.g., DS rates are significantly
higher in older mothers when previous family size is zero
as compared to average family sizes in a given age
category. This effect is corroborated by a much smaller
(217 DS cases) case-control study that reports significantly
increased age-corrected DS incidence in primiparae (Smith
and Record 1955). More recently, Doria-Rose et al. (2003)
found a trend towards increasing risk of DS with increasing
parity. However, this study was criticized because it did not,
or only partially, account for terminations of pregnancy,
births to women for whom condition was not recorded on
their birth certificates, and lesser use of prenatal diagnosis
by women of higher parity (Chan 2003; Aliyu et al. 2005).
Secondly, we found increased IBIs before the birth of a
DS child, which is in agreement with more stringent
filtering the younger the youngest child in the family (see
Introduction). These intriguing findings are inexplicable by
the alternative hypotheses discussed so far (see Introduction,
and also Erickson, 1978, p. 293), i.e., non-adaptive RFS or
genetic ovum degradation scenarios, including those based
on chromosomal drive (CD).
We would like to emphasize that potentially confounding
correlations of age of mothers with our selected covariables
Table 1 Birth-order specific Down syndrome (DS) rates (number per 100,000 live births based on 4,130 reported DS cases) for six maternal age
categories (Hay and Barbano 1972, p. 273)
Maternal age (years)
Under 20
20–24
25–29
30–34
35–39
40 and over
Cochran–Armitage testa
Birth order
1
2 and 3
4 and 5
6 and over
21.9 (205)
24.7 (288)
28.8 (88)
37.0 (32)
188.5 (69)
674.1 (60)
19.2 (57)
21.2 (339)
28.3 (312)
50.0 (226)
165.4 (295)
603.6 (233)
–
19.7 (53)
27.8 (157)
45.8 (196)
147.2 (321)
570.0 (346)
–
29.7 (7)
25.4 (39)
42.0 (104)
131.3 (278)
518.3 (425)
The numbers of reported DS cases are given in parentheses.
a
Two-sided asymptotic p value
p=0.2214
p=0.0756
p=0.5422
p=0.5406
p=0.0007
p=0.0146
Naturwissenschaften
Table 2 Inter-birth interval (IBI, months) before a birth of given parity, for either normal births or births of a Down syndrome (DS) child (Øster
1953, Appendix 4, cf., Materials and methods)
Parity
2
3
4
5
6, 7
8, 9
10, higher
Normal IBIa
IBI before DS child
Wilcoxon
rank-sum testb
N
Median
Mean
SE
N
Median
Mean
SE
250
178
129
82
124
42
14
27
22
27
27
27
22
32
34.9
31.2
33.7
29.6
30.0
25.3
33.1
1.69
1.69
2.25
2.02
1.26
1.85
3.69
102
87
55
50
39
38
19
42
42
57
47
52
37
42
58.3
57.7
61.6
63.4
54.8
47.5
48.7
5.35
5.06
5.26
6.18
5.88
5.48
6.63
p<0.0001
p<0.0001
p<0.0001
p<0.0001
p=0.0001
p<0.0001
p=0.1032
Simulated IBI before DS
child without RFS
Median
Mean
50.2
41.6
47.4
36.5
34.1
28.3
37.1
49.9
41.8
47.4
36.5
34.0
28.1
37.1
See Materials and methods for details of the simulation study.
RFS relaxed filtering stringency
a
IBI between two normal pregnancies, i.e., IBI before given pregnancy in families with a later DS child.
b
Two-sided exact p value (exact test chosen because of many ties) for difference between normal IBI and IBI before DS child
are very unlikely to have led to artificial effects. First, there
is no apparent reason to believe that mothers are systematically older within each age category, the lower the birth
order of the respective births. Rather, the opposite would be
a reasonable guess in populations unaffected by contraceptive measures, i.e., where the probability to deliver firstborns falls more steeply with age than the other parity
groups. Hence, any artifactual age-confounded effect would
oppose the trend of higher DS incidence with lower parity
found in Table 1 and Fig. 1, strengthening the significance
of our findings. Secondly, slightly longer IBIs before DS
births would also result from the fact that DS rates increase
with age, i.e., mothers deciding for longer IBI for unrelated
reasons would also experience slightly higher DS rates.
However, our simulations indicate that this effect would
be far smaller than the actually observed difference, at least
for parities of 4 and higher. In fact, by using the average
maternal age distribution at all parities, while in reality
mothers are younger the lower the parity, our simulation
maximizes this confounding effect. Hence, the finding of
shorter IBIs for normal birth in the sample for lower parities
indicates the potential presence of RFS even at lower parity.
In consequence, the effects of family size and age of
previous child on DS incidence rate give compelling reason
to imply adaptive RFS as being involved in DS rate increase;
hence, both RFS and CD could potentially account for
accelerated incidence rates with maternal age. The extent of
the involvement of CD in DS occurrence hinges on accurate
estimates of trisomy 21 incidence and differential allocation
at oogenesis. However, while the most decisive predictions
from an involvement of CD are increased rates of trisomy 21
in zygotes with maternal age at conception and the absence
of nullosomies in embryos (Day and Taylor 1998), results on
these issues are hard to validate; aborted nullisomies might
go undetected at very early embryonic stages, which could
also apply to an unknown proportion of trisomies that might
itself depend on maternal age (Stein et al. 1986).
Clearly, the exact age-dependent initial incidence rates as
well as the concrete stringency of the adaptive filtering (i.e.,
relation of discarded to gestated aberrant embryos for the
complete pre-natal period) would be needed to elucidate the
relative contribution of CD and RFS. Recent data on large
numbers of human oocytes at metaphase II suggest that
aneuploidies might actually exhibit an accelerating trend
with maternal age (Pellestor et al. 2005). While it might be
argued that this finding terminates the relevance of aberrant
embryo filtering for our understanding of DS incidence
dynamics, Zheng et al. (2000) argue that the accelerating
increase of aberrant ova is consistent with the view of
optimal selection from the pool of developing ova, i.e.,
adaptive RFS.
It should be noted that adaptive RFS is expected to be
the less pronounced the lower the fitness prospects of
aberrant phenotypes are, which might account for intermediate increases of trisomy 18 and lowest increases of
trisomy 13 with maternal age (Drugan et al. 1999). Also,
adaptive RFS might not evolve when incidence rates are
rather low, e.g., for translocation DS (Hook 1983). Thus,
relaxed selection for trisomy 21 is not discounted by the
lack of a maternal age effect in translocation DS, in contrast
to the argument of Simpson and Elias (2003). Simpson and
Elias (2003) also report a maternal age effect in trisomy 21
abortions. However, that effect is consistent with adaptive
RFS as long as the increase of the proportion of aberrant
offspring with age is steeper for live births than for
spontaneous abortions/miscarriages.
Previously, it was argued that relaxed selection can
explain “the apparently paradoxical finding of increased
Naturwissenschaften
maternal age in cases of trisomy 21 arising from paternal
nondisjunction” (Ferguson-Smith and Yates 1984, p. 29;
see also Juberg and Mowrey 1983). However, according to
more recent results, paternal origin accounts for only about
5% of all DS cases (Buwe et al. 2005). Thus, there might be
many misclassifications of maternal cases as paternal in the
old data. According to Buwe et al. (2005) and Sherman et
al. (2005), studies of the paternally derived DS cases led to
conflicting results. One study that reports the absence of a
maternal age effect on paternal origin DS (Yoon et al. 1996)
is based on the very low sample size of N=8.
The above shows that RFS could accommodate the
exclusivity of accelerated increases with age of DS, in
contrast to other congenital aberrations that do not or only
slightly increase with age. While mechanistical constraints
are again implausible to account for this exclusivity, CD is
also anticipated to be specific to particular genetic arrangements, implying that CD effects could be exclusive to
certain chromosomal aberrations. However, CD could not
accommodate the findings of our current study. Moreover,
RFS with maternal age is, of course, adaptive not only for
chromosomal aberrations (Kloss and Nesse 1992), i.e., RFS
could operate outside the realm of CD. For instance,
hypospadias, where affected males are known to be fertile,
has been found at very high incidence in firstborns to
mothers over 40 years old (Hay and Barbano 1972).
Clearly, identification of the relaxation control mechanism
and therapeutic restoration of a stringent screen holds
promise not only for Down syndrome.
References
Aliyu MH, Jolly PE, Ehiri JE, Salihu HM (2005) High parity and
adverse birth outcomes: exploring the maze. Birth 32:45–59
Buwe A, Guttenbach M, Schmid M (2005) Effect of paternal age on the
frequency of cytogenetic abnormalities in human spermatozoa.
Cytogenet Genome Res 111:213–228
Chan A (2003) Invited commentary: parity and the risk of Down’s
syndrome—caution in interpretation. Am J Epidemiol 158:509–511
Cuckle HS, Wald NJ, Thompson SG (1987) Estimating a woman’s
risk of having a pregnancy associated with Down’s syndrome
using her age and serum alpha-feto protein level. Br J Obstet
Gynaecol 94:387–402
Day T, Taylor PD (1998) Chromosomal drive and the evolution of
meiotic nondisjunction and trisomy in humans. Proc Natl Acad
Sci USA 95:2361–2365
Doria-Rose VP, Kim HS, Augustine ETJ, Edwards KL (2003) Parity
and the risk of Down’s syndrome. Am J Epidemiol 158:503–508
Drugan A, Yaron Y, Zamir R, Ebrahim SAD, Johnson MP, Evans MI
(1999) Differential effect of advanced maternal age on prenatal
diagnosis of trisomies 13, 18 and 21. Fetal Diagn Ther 14:181–184
Erickson D (1978) Down syndrome, paternal age, maternal age and
birth order. Ann Hum Genet 41:289–298
Ferguson-Smith MA, Yates JRW (1984) Maternal age specific rates
for chromosome aberrations and factors influencing them: report
of a collaborative European study on 52965 amniocenteses.
Prenat Diagn 4:5–44
Forbes LS (1997) The evolutionary biology of spontaneous abortion
in humans. Trends Ecol Evol 12:446–450
Hay S, Barbano H (1972) Independent effects of maternal age and
birth order on the incidence of selected congenital malformations.
Teratology 6:271–280
Hobcraft J, McDonald J, Rutstein SO (1983) Child-spacing effects on
infant and early child mortality. Popul Index 49:585–618
Hook EB (1981) Rates of chromosome abnormalities at different
maternal ages. Obstet Gynecol 58:282–285
Hook EB (1983) Down syndrome rates and relaxed selection at older
maternal ages. Am J Hum Genet 35:307–1313
Juberg RC, Mowrey PN (1983) Origin of nondisjunction in trisomy
21 syndrome: all studies compiled, parental age analysis, and
international comparisons. Am J Med Genet 16:111–116
Kloss RJ, Nesse RM (1992) Trisomy: chromosome competition or
maternal strategy? Ethol Sociobiol 13:283–287
Koenig M, Phillips JA, Campbell OM, D’Souza S (1990) Birth intervals and childhood mortality in rural Bangladesh. Demography
27:251–265
Kozlowski J, Stearns SC (1989) Hypotheses for the production of
excess zygotes: models of bet-hedging and selective abortion.
Evolution 43:1369–1377
Krackow S (1998) Human pregnancy filtering: yesterday’s news?
Trends Ecol Evol 13:156
Lamb NE, Yu K, Shaffer J, Feingold E, Sherman SL (2005)
Association between maternal age and meiotic recombination
for trisomy 21. Am J Hum Genet 76:91–99
Morris JK, Mutton DE, Alberman E (2002) Revised estimates of the
maternal age specific live birth prevalence of Down’s syndrome.
J Med Screen 9:2–6
Morris JK, Wald NJ, Mutton DE, Alberman E (2003) Comparison of
models of maternal age-specific risk for Down syndrome live
births. Prenat Diagn 23:252–258
Morris JK, de Vigan C, Mutton DE, Alberman E (2005) Risk of a
Down syndrome live birth in women 45 years of age and older.
Prenat Diagn 25:275–278
Øster J (1953) Mongolism. Danish Science, Copenhagen
Pellestor F, Anahory T, Hamamah S (2005) Effect of maternal age on
the frequency of cytogenetic abnormalities in human oocytes.
Cytogenet Genome Res 111:206–212
Penrose LS (1934) The relative aetiological importance of birth order
and maternal age in mongolism. Proc R Soc Lond B 115:431–450
Robins J, Breslow N, Greenland S (1986) Estimators of the Mantel–
Haenszel variance consistent in both sparse data and large-sample
limiting models. Biometrics 42:311–323
Sherman SL, Freeman SB, Allen EG, Lamb NE (2005) Risk factors for
nondisjunction of trisomy 21. Cytogenet Genome Res 111:273–280
Simpson JL, Elias S (2003) Genetics in obstetrics and gynecology, 3rd
edn. Saunders, Philadelphia
Smith A, Record RG (1955) Maternal age and birth order in the
aetiology of mongolism. Br J Prev Soc Med 9:51–55
Stein Z, Stein W, Susser M (1986) Attrition of trisomies as a maternal
screening device. Lancet 944–947
Warburton D, Stein Z, Kline J (1983) In utero selection against fetuses
with trisomy. Am J Hum Genet 35:1059–1064
Yoon PW, Freeman SB, Sherman SL, Taft LF, Gu YC, Pettay D,
Flanders WD, Khoury MJ, Hassold TJ (1996) Advanced
maternal age and the risk of Down syndrome characterized by
the meiotic stage of the chromosomal error: a population-based
study. Am J Hum Genet 58:628–633
Zheng CJ, Guo SW, Byers B (2000) Modeling the maternal-age
dependency of reproductive failure and genetic fitness. Evolut
Develop 2:203–207