respectively) (fig. S6, B and C). In line with this
observation, although MIX N and BPA shared
bdnfas a common dysregulated target (Fig. 1C
and fig. S6D), no other MIX N target was mod-
ulated by BPA, showing that the dysregulation
by MIX N cannot be primarily ascribed to BPA
(table S6). In zebrafish larvae, BPA did not af-
fect the genes that were affected by MIX N at
the tested concentrations (Fig. 5G, fig. S6G,
and table S6).
Taken together, the results across species
demonstrate TH-disrupting effects of MIX N
at molecular and physiological levels. These
effects are not comparable with the effects of
BPA alone, both at the concentrations mea-
sured and at concentrations associated with
adverse outcomes in humans.
MIX N disrupts locomotion in developmental
in vivo models
Finally, to establish the organismal-level im-
pact of the mixture, we used locomotor assays.
Locomotion is a well-established target of
neurodevelopmental alterations, including,
most prominently, those caused by TH dis-
ruption, given the role of TH in the matura-
tion of the central and peripheral nervous
systems ( 77 – 79 ). We analyzed light-induced
locomotion inXenopustadpoles after 72 hours
of MIX N exposure. Mobility was significantly
(P< 0.0001 at certain time points) decreased
after exposure to MIX N at 1000X and fol-
lowed a dose-dependent decreasing pattern
(Fig. 5E and fig. S9), whereas no effects were
observed after BPA exposure (fig. S6E).
In zebrafish larvae, in which activity in-
creases in darkness and abates in light, beha-
vioral effects were assessed after 48 hours of
exposure to MIX N, revealing significantly in-
creased mobility during dark periods, which
followed a dose-dependent increasing pat-
tern (P< 0.00011), and at certain time points
during light cycles (P< 0.05) after exposure
to 100X MIX N (Fig. 5H and fig. S9). No
effects were observed after exposure to BPA
(fig. S6H). Together, these results demonstrate
that MIX N affects animal behavior at con-
centrations associated with adverse outcomes
in humans.
Integrating epidemiological and experimental
evidence for risk assessment using SMACH
Finally, we combined the epidemiological and
experimental evidence of adverse neurodeve-
lopmental effects associated with prenatal
exposure to the EDCs in a mixture-based risk
assessment setting. To this end we used SMACH,
which is a two-tiered strategy aimed at (i) as-
sessingwhatpercentageofastudypopulation
has a mixture exposure sufficiently similar
to a reference mixture ( 80 ), in this case, MIX
N that was epidemiologically defined and
experimentally tested; and (ii) calculating a
similar mixture risk indicator [SMRI ( 80 )] for
study participants that were determined to
have sufficiently similar mixtures to the refe-
rence mixture. The SMRI relates the human
exposure range to experimentally determined
points of departure [in this case a benchmark
dose (BMD) lower confidence limit (BMDL)],
similar to a hazard quotient for single chemicals.
The SMRI was further evaluated as a metric
of risk assessment with an odds ratio relating
low and high deciles of SMRI with language
delay in the children.
Using equivalence testing methodology in
SMACH, we linked experimental evidence of
adverse effects in zebrafish andXenopusthat
were attributed to MIX N with concentrations
of EDCs in pregnant women in the SELMA
study. First, we built on the analytical frame-
work by Marshallet al.( 80 ) and defined a
similarity region as follows. For the XETA,
according to OECD guidelines ( 81 ), the thresh-
old for a response is set at a 12% decrease from
the reference control; thus, a benchmark re-
sponse (BMR) was set at 88% of the reference
control. The similarity regions for the BMD
were thus defined by adding a further 12%
change in response from the BMR to define
the radius of the similarity region, represented
bythechangeintotaldoseofthemixturebe-
tween a 12 and 24% response. For the zebra-
fish, a one–standard deviation (SD) change from
control defined the BMR, and an additional
one-SD change from the BMR defined the
radius of the similarity region (fig. S7 and
table S7). Using those similarity regions, the
mixture for each SELMA pregnant woman
was tested for similarity to MIX N. Based
on the results of, respectively, of the XETA and
zebrafish studies, 96 and 91% of the pregnant
women were determined to have sufficient-
ly similar exposures to the reference mixture
(table S8).
In the second step, a SMRI was constructed
for subjects that were determined to have
been exposed to mixtures sufficiently similar
to the reference mixture, such that a higher
index indicates a higher exposure. Children
with maternal prenatal concentrations suf-
ficiently similar to MIX N with the highest
decile values of SMRI were 3.3 times more
likely (odds ratio) to have language delay at
2.5 years of age than the children with SMRI
values in the lowest decile (P= 0.043 using
zebrafish results, andP= 0.031 using XETA
results).
Further, the index is similar to a hazard
quotient for single chemicals but instead uses
the specific reference mixture in units of total
concentrations for each subject divided by the
BMDL from experimental evidence of toxicity.
The lowest BMDL, in terms of SELMA con-
centrations, was 8X in XETA compared with
44X in zebrafish (table S8). The XETA BMR
is defined by OECD guidelines, and it re-
sulted in the most sensitive test. Thus, weused it to calculate the percentage of women
with SMRI values that exceed 1 and found
that 54% women had values above levels of
concern.Discussion
The current vision for improving regulatory
decision-making, in public health and health
care alike, relies on the transforming potential
of high-throughput and high-content data to
elucidate and quantify the molecular, cellular,
and organismal responses to genetic vulner-
abilities and environmental insults ( 82 , 83 ).
This work, through complementary experi-
mental methodologies, uncovered the gene
networks that were specifically altered by an
EDC mixture that is epidemiologically asso-
ciated with language delay, identifying thy-
roid, estrogen, and PPAR pathways as major
convergent targets. This empirical evidence
allowed us to determine the proportion of a
human population with exposure ranges of
concern, thereby providing an integrative
framework for mixture-based risk assessment
strategies (Fig. 6) whose key features can be
summarized as follows, alongside the attending
limitations.
First, the chemicals profiled in this study,
selected for the strong rationale presented in
the introduction, clearly only represent a por-
tion of all chemicals that humans are exposed
to at present. In terms of exposure to short-
lived pollutants (BPA and phthalates), we
measured metabolites in urine, whose con-
centrations can vary during pregnancy and can
thus be most reliably evaluated with reference
to the specific day or week ( 84 , 85 ). Our ex-
posure assessment for these chemicals may
thus only reflect concentrations near week
10 of pregnancy, and our results may not be
necessarily comparable to studies that col-
lected samples at time points later in pre-
gnancy. However, prior research in Swedish
populations has demonstrated that phthal-
ate and BPA concentrations were correlated
with many lifestyle and household character-
istics ( 86 , 87 ), whose relative stability plausibly
results in pseudo-persistent exposure to shorter-
lived compounds ( 88 ). We also standardized
urine collection by using only first-morning
voids, which may be more reliable than spot
sampling ( 89 , 90 ). Further, exposure mis-
classification should be arguably nondifferen-
tial with respect to the outcome, so we expect
the bias to be toward the null.
In terms of statistical methods, based on the
hypotheses of interest, the range of options
included Bayesian kernel machine regression
for detection of interactions ( 91 , 92 ), WQS
regression for estimating a mixture effect,
and latent class models for profile classifica-
tions. We used WQS regression as a strategy
agnostic to chemical interactions for study-
ing the association between prenatal mixtureCaporaleet al.,Science 375 , eabe8244 (2022) 18 February 2022 9 of 15
RESEARCH | RESEARCH ARTICLE