Genomic inference of a severe human bottleneck during the Early to Middle Pleistocene transition (Hu et al., Science 2023) — FULL TEXT

Source (primary, peer-reviewed): Wangjie Hu, Ziqian Hao, Pengyuan Du, Fabio Di Vincenzo, Giorgio Manzi, Jialong Cui, Yun-Xin Fu, Yi-Hsuan Pan, Haipeng Li, "Genomic inference of a severe human bottleneck during the Early to Middle Pleistocene transition," Science 381:979–984 (1 Sept 2023). DOI: 10.1126/science.abq7487.
Provenance: full published version obtained from the open lab-copy PDF at websites.umich.edu/~zhanglab/clubPaper/09_05_2023.pdf (the Science version is otherwise paywalled). Captured: 2026-06-09, full verbatim (7 pp; PDF text extraction, figure/table layout not preserved). Provenance only; analysis on ../../topics/ancient-astronaut-theory-and-alien-dna.
What it is / why it's here: the FitCoal-method inference that the human ancestral breeding population crashed to ~1,280 individuals for ~117,000 years, ~930,000–813,000 years ago (~98.7% loss) at the Early–Middle Pleistocene transition. Cited in the base as the one real empirical datum the "engineered humanity / alien DNA" narrative attaches to — and why it does not support that reading (a population crash in a continuous line of descent, environmentally interpreted, contested by later work, and ~600,000 years before Homo sapiens). It is not evidence of manipulation.

---

HUMAN EVOLUTION
Genomic inference of a severe human bottleneck
during the Early to Middle Pleistocene transition
Wangjie Hu1,2†‡, Ziqian Hao3†, Pengyuan Du1,3, Fabio Di Vincenzo4, Giorgio Manzi5, Jialong Cui2,
Yun-Xin Fu6,7, Yi-Hsuan Pan2*, Haipeng Li1,8*
Population size history is essential for studying human evolution. However, ancient population size history
during the Pleistocene is notoriously difficult to unravel. In this study, we developed a fast infinitesimal time
coalescent process (FitCoal) to circumvent this difficulty and calculated the composite likelihood for present-
day human genomic sequences of 3154 individuals. Results showed that human ancestors went through a
severe population bottleneck with about 1280 breeding individuals between around 930,000 and 813,000 years
ago. The bottleneck lasted for about 117,000 years and brought human ancestors close to extinction. This
bottleneck is congruent with a substantial chronological gap in the available African and Eurasian fossil record.
Our results provide new insights into our ancestry and suggest a coincident speciation event.
A
lthough the lineage of humans is esti-
mated to have separated from that of
chimpanzees and bonobos more than
6 million years ago, anatomically mod-
ernhumans (Homosapiens) are estimated
to have originated around 300 thousand to
200 thousand years before the present (kyr BP)
in Africa (1–3). On the basis of present-day hu-
man genomic sequences, the recent population
size histories (i.e., the dynamics of population
size since the emergence of modern humans)
have been intensively studied, revealing the
worldwide spread of our ancestors (4–8). How-
ever, ancient population size history of the genus
Homo during the Pleistocene is still poorly
known, although it is essential for understand-
ing the origin of the human lineage. It is likely
to be very difficult or impossible to obtain an-
cient DNA from African Homo samples dated
before the emergence of H. sapiens. It would
be particularly notable if present-day human
genomic sequences could be used to robustly
infer both the recent and ancient population
size histories of humankind. Thus, a new ap-
proach is needed to improve the inference ac-
curacy of population size history.
Population size changes that occurred hun-
dreds of thousands of years ago affected the
rates of coalescence and thus have left their
signatures in the site frequency spectrum (SFS)
of genomic sequences. The SFS is the distri-
bution of allele frequencies in the sequences,
randomly collected from the present-day hu-
man population. Each SFS category contains
a certain number of mutations of the same
size. Because SFS is crucial for demographic
inference (6, 8–12) and construction of key
summary statistics (13), many efforts have been
devoted to derivingitsanalytical formulas(14–17).
However, these formulae may not achieve the
required accuracy because of propagation and
accumulation of numerical errors resulting from
their dependence on the joint probability den-
sity function of coalescent times (16–18).
In this study, to circumvent this numerical
problem, we developed the fast infinitesimal
time coalescent process (FitCoal) (Fig. 1) that
analytically derives expected branch length
for each SFS category under arbitrary demo-
graphic models. FitCoal does not need phased
haplotype data and prior information on de-
mography. The effects of sequencing errors or
hitchhiking due to positive selection can be
circumvented, largely by focusing on a subset
of SFS that are less influenced by those factors.
We used FitCoal to analyze a large number of
present-day human genomic sequences from 10
African and 40 non-African populations. Results
showed that our ancestors experienced a sev-
ere population bottleneck between about 930
and 813 kyr BP, most likely because of climatic
changes. The average number of breeding indi-
viduals was only about 1280 during the bottle-
neck period. Our findings indicate that the
severe bottleneck brought the ancestral human
population close to extinction and completely
reshaped present-day human genetic diversity.
Results
FitCoal
We developed FitCoal to determine the expected
branch lengths for an SFS (Fig. 1). During
FitCoal analysis of a sample, the time period
in which the most recent common ancestor
RESEARCH
Hu et al., Science 381, 979–984 (2023)
1 September 2023
1 of 6
1CAS Key Laboratory of Computational Biology, Shanghai
Institute of Nutrition and Health, University of Chinese Academy
of Sciences, Chinese Academy of Sciences, Shanghai, China.
2Key Laboratory of Brain Functional Genomics of Ministry of
Education, School of Life Science, East China Normal University,
Shanghai, China. 3College of Artificial Intelligence and Big Data
for Medical Sciences, Shandong First Medical University &
Shandong Academy of Medical Sciences, Jinan, China. 4Natural
History Museum, University of Florence, Florence, Italy.
5Department of Environmental Biology, Sapienza University of
Rome, Rome, Italy. 6Department of Biostatistics and Data
Science, School of Public Health, University of Texas Health
Science Center at Houston, Houston, TX, USA. 7Key Laboratory
for Conservation and Utilization of Bioresources, Yunnan
University, Kunming, China. 8Center for Excellence in Animal
Evolution and Genetics, Chinese Academy of Sciences,
Kunming, China.
*Corresponding author. Email: yxpan@sat.ecnu.edu.cn (Y.-H.P.);
lihaipeng@sinh.ac.cn (H.L.)
†These authors contributed equally to this work.
‡Present address: Department of Genetics and Genomic Sciences,
Icahn School of Medicine at Mount Sinai, New York, NY, USA.
Fig. 1. Illustration of FitCoal. (Left) The backward process in which four lineages (represented by the
four solid black circles at the bottom) coalesce into one (represented by the single solid black circle at
the top) after passing through millions of infinitesimal time intervals (Dt). The area highlighted in blue shows the
backward transformation process of different coalescent states with tiny probability changes in an infinitesimal
time interval. Thick arrows indicate high transformation probabilities, and thin arrows indicate low transformation
probabilities. The blue and purple arrows correlate to the two events in the middle pane represented by
blue- and purple-colored lines. Each state is indicated with a box, in which one circle indicates one lineage.
The boxes with solid black circles represent the states with the probability of 1. The boxes with empty
circles represent the states with the probability of 0. The probabilities between 0 and 1 are represented
by gray circles. (Middle) Hypothetical coalescent trees with branches of different states, indicating the
number of lineages. Blue branches represent a transformation from four to three lineages. Purple branches
indicate that no coalescent event occurred. (Right) The size of a theoretical population over time. The
width of shadowed area denoted as N(t) indicates the effective population size (i.e., the number of breeding
individuals) at time t. It is assumed that the effective population size remains unchanged within a Dt.
Downloaded from https://www.science.org at The Medical Library of the Chinese Pla on September 03, 2023

originated was partitioned into as many time
intervals as needed, such that each time inter-
val (Dt) was very small (e.g., 1 month or 1 year).
During each time interval, the population size
was assumed to be constant. The probabilities
of all coalescent states (i.e., all possible ances-
tral lineages) were calculated backward in
time. For each state, the branch length during
a time interval was calculated by multiplying
its probability with population size and then
transformed to determine the expected branch
lengths. Because the expected branch length
of an SFS category during a time interval was
precalculated, FitCoal could be very fast.
FitCoal demographic inference
After the expected branch lengths were de-
termined, the composite likelihood of an SFS
(6, 9, 19) was calculated. FitCoal is effective
for a wide range of sample sizes in the calcu-
lation of the composite likelihood of a given
SFS and is much more accurate than simu-
lation approaches (fig. S1). When inferring
population size history, the likelihood was
maximized in a wide range of demographic
scenarios. Moreover, both instantaneous and
long-term exponential population changes were
considered. Similar to previous studies (6, 19),
the likelihood of the SFS was first maximized
with the constant size model, followed by re-
peated maximization of the likelihood with
increased number of inference time intervals
until the best model was found.
Demographic inference from simulated data
The accuracy of FitCoal demographic infer-
ence was evaluated by simulation and com-
parison of results with those of PSMC (pairwise
sequentially Markovian coalescent), Stairway
Plot, and SMC++ methods (6–8) (Fig. 2). To
ensure fair comparisons, we tested six demo-
graphic models by simulating 200 independent
datasets for each model, as described previ-
ously (6), with the assumption that a generation
time is 24 years (6, 20) and that the mutation
rate is 1.2 × 10–8 per site per generation (6, 21).
Results showed that the medians of FitCoal-
inferred population size histories were almost
identical to the true models, and the 95% con-
fidence intervals of FitCoal inference were nar-
rower than those of PSMC, Stairway Plot, and
SMC++ (Fig. 2). FitCoal inference accuracy
could be further improved by increasing sam-
ple size and lengths of sequences (fig. S2). The
proportion of the most recent changes in popu-
lation size inferred from the six models showed
that FitCoal could distinguish between instan-
taneous and exponential changes (table S1).
Overall, our results confirmed that SFSs could
be used to estimate population size histories (22).
It has been suggested that a population size
history could be inferred by using a subset of
SFS or a collapsed SFS (6, 19); the latter is an
SFS with high frequency mutations combined
into one category. Resultsof simulationsshowed
that the FitCoal could still accurately determine
a population size history even when a portion
(10 to 90%) of an SFS was truncated (i.e., ex-
cluded for analysis) (figs. S3 to S5), thus reduc-
ing the impact of confounding factors, such as
hitchhiking effect due to positive selection (fig.
S6) or sequencing errors on FitCoal analyses.
Demographic inference of African populations
To infer population size histories of African
populations, seven African populations in the
1000 Genomes Project (1000GP) (23) and three
African populations in the Human Genome
Diversity Project–Centre d’Etude du Polymor-
phisme Humain (HGDP-CEPH) panel (24) were
analyzed by the FitCoal (tables S2 to S4). Only
autosomal noncoding regions were used to
partially avoid the effect of purifying selection.
To avoid hitchhiking effect due to positive
Hu et al., Science 381, 979–984 (2023)
1 September 2023
2 of 6
Effective population size
(in thousands)
10
100
1000
1
10
100
Constant size
n = 30, L = 10 Mb
10
100
1000
10
100
Instantaneous increase
n = 30, L = 10 Mb
10
100
1000
1
10
100
Bottleneck
n = 170, L = 30 Mb
Time (thousand years ago)
Effective population size
(in thousands)
10
1
100
10,000
Exponential growth I
n = 30, L = 10 Mb
Time (thousand years ago)
10
100
1
100
10,000
Exponential growth II
n = 30, L = 10 Mb
Time (thousand years ago)
10
100
1000
10
100
500
Exponential growth III
n = 170, L = 10 Mb
A
E
D
C
B
F
True Model
FitCoal
SMC++
Stairway Plot
PSMC
Fig. 2. Population size histories inferred by FitCoal, PSMC, Stairway Plot, and
SMC++ with simulated samples. (A) Constant size model. (B) Instantaneous
increase model. (C) Bottleneck model. (D) Exponential growth I model. (E) Exponential
growth II model. (F) Exponential growth III model. In all panels, thin black lines indicate
the true models. Thick red lines indicate the medians of FitCoal-inferred population size
histories; thin red lines represent 2.5 and 97.5 percentiles of FitCoal-inferred population
size histories. Yellow, green, and blue lines indicate results obtained with PSMC, Stairway
Plot, and SMC++, respectively. The mutation rate is assumed to be 1.2 × 10–8 per
base per generation, and a generation time is assumed to be 24 years. n is the number of
simulated sequences, and L is the length in Mb of each simulated sequence.
RESEARCH
|
RESEARCH ARTICLE
Downloaded from https://www.science.org at The Medical Library of the Chinese Pla on September 03, 2023

selection (25), high-frequency mutations were
excluded from the analysis. Results showed that
all 10 African populations went through a sev-
ere bottleneck (Fig. 3 and figs. S7 and S8). The
bottleneck was estimated to persist for 117 kyr,
from 930 ± 23.52 (SEM) (range, 854 to 1042)
to 813 ± 11.02 (SEM) (range, 772 to 864) kyr BP.
The average effective population size (i.e., the
number of breeding individuals) (26) during
the bottleneck period was determined to be
1280 ± 131 (SEM) (range, 770 to 2030), which
was only 1.3% of its ancestral size (98,130 ±
8720; range, 58,600 to 135,000). To evaluate the
impact of the bottleneck on current human
genetic diversity, we analyzed the expected
pairwise nucleotide diversity. Results showed
that 65.85% of current human genetic diver-
sity was lost because of the bottleneck.
Demographic inference of non-African populations
The severe bottleneck was not directly detected
in all 19 non-African populations in 1000GP and
the 21 non-African populations in the HGDP-
CEPH panel (Fig. 3, A to D; figs. S7 and S8; and
tables S2, S5, and S6). The average ancestral
population sizes of the populations in the two
datasets were 20,260 (range, 18,850 to 22,220)
and 20,030 (range, 19,060 to 21,850), respec-
tively, similar to those determined in previous
studies (7, 8, 24). The estimated population
size started to decline around 368 (range, 175 to
756) and 367 (range, 167 to 628) kyr BP, respec-
tively, which is consistent with previous find-
ings that African and non-African divergence
occurred much earlier than the out-of-Africa
dispersal (7, 8, 23, 24). The inferred out-of-
Africa dispersal and the recent population size
expansion and reduction are consistent with
those of previous studies (5–8, 23, 24).
Severe bottleneck during the Early to Middle
Pleistocene transition
The ancient severe bottleneck was directly de-
tected in each of the 10 African populations
but in none of the 40 non-African populations.
To investigate this discrepancy, we performed
simulations with three demographic models,
designated bottlenecks I, II, and III (Fig. 4, A
to C, and figs. S9 and S10). Bottleneck I sim-
ulated the population size history of African
agriculturalist populations with the ancient
severe bottleneck, and bottlenecks II and III
simulated that of non-African populations with-
out and with the ancient severe bottleneck,
respectively. Both bottlenecks I and II were
inferred precisely in all simulated data sets
(tables S7 to S9). However, no ancient severe
bottleneck was detected in bottleneck III sim-
ulations, which indicates that the out-of-Africa
dispersal hinders the chance of discovering the
ancient severe bottleneck. Furthermore, the
ancient severe bottleneck was found to cause a
discrepancy in the estimation of the population
size between the bottleneck III model and the
inferred population size history after the bottle-
neckwas relieved, which suggests a hidden effect
Hu et al., Science 381, 979–984 (2023)
1 September 2023
3 of 6
Fig. 3. Histories of human populations in 1000GP and HGPD-CEPH genomic
datasets inferred by FitCoal, SMC++, Stairway Plot, PSMC, and Relate. The
mutation rate is assumed to be 1.2 × 10–8 per base per generation, and a generation
time is assumed to be 24 years. (A) Inferred population size histories of 26
populations in 1000GP. (B) Linear-scaled estimation of sizes over time of populations
in 1000GP during the severe bottleneck period. (C) Inferred population size histories
of 24 populations in the HGPD-CEPH panel. (D) Linear-scaled estimation of sizes
over time of populations in the HGPD-CEPH panel during the severe bottleneck period.
(E and F) Comparison of population size histories of African YRI and Yoruba
populations inferred by FitCoal, SMC++, Stairway Plot, PSMC, and Relate. Only
the population size histories up to 200 kyr BP were analyzed by Stairway
Plot. In (A) to (D), colored lines indicate the following: red, African populations;
brown, Middle East populations; yellow, European populations; blue, East Asian
populations; green, Central or South Asian populations; and gray, American
populations. Black dashed circles with arrow heads represent the discrepancy in
the population size between African agriculturalist and non-African populations.
In (E) and (F), red, blue, yellow, green, and gray lines indicate results obtained
with FitCoal, SMC++, PSMC, Stairway Plot, and Relate, respectively.
RESEARCH
|
RESEARCH ARTICLE
Downloaded from https://www.science.org at The Medical Library of the Chinese Pla on September 03, 2023

of the ancient severe bottleneck on non-African
populations (Fig. 4C and figs. S9C and S10C).
After the bottleneckwas relieved, the average
population size of non-African populations in
1000GP was 20,260, and that of those in HGDP-
CEHP was 20,030. For African agriculturalist
populations, the average population size in
1000GP was 27,080, and that of those in HGDP-
CEHP was 27,440. This population size difference
of 7020 (Fig. 3, A and C) is likely due to the
hidden effect of the ancient severe bottleneck
on non-African populations. Because the out-of-
Africa dispersal existed in non-African popu-
lations but not in African populations, African
populations had more lineages remaining to
be traced back to the ancient severe bottle-
neck (fig. S11). In the analysis of the African YRI
(Yoruba in Ibadan, Nigeria) population, the
minimum sample size was three individuals
for detection of the severe bottleneck (fig. S12).
Because the signal for the existence of the
severe bottleneck was too weak to be detected
in non-African populations by FitCoal, we per-
formed an extended FitCoal analysis. To eli-
minate noise effects resulting from problems
such as sequencing or overfitting errors on the
inference of population size history, we used
the FitCoal-inferred recent population size his-
tory as a starting point for size estimation of
an ancient population. With this modifica-
tion, all 19 non-African populations in 1000GP
were found to have gone through the severe
bottleneck with approximately 1450 individ-
uals between 921 and 785 kyr BP (fig. S13 and
table S10). This result is consistent with that
obtained with African populations.
To further examine the severe bottleneck, we
simulated a slow population reduction start-
ing 1.5 million years ago (Fig. 4D). The FitCoal-
inferred population size histories were different
from those observed in 1000GP and HDGP-
CEHP populations, which supports the hypoth-
esis that a sudden size reduction occurred at
the beginning of the bottleneck. Results of sim-
ulations were similar to that of the observed
cases (Fig. 3, E and F) in that they also showed
that PSMC, SMC++, and Relate methods (7, 8, 27)
underestimated the severity of the ancient bot-
tleneck (Fig. 4 and figs. S14 to S17). More-
over, the inferred population declines shown
in Fig. 4A were more severe than those inferred
from the real data because the simulated data
were generated under the assumption of a neu-
trally evolved single population and a homoge-
neous recombination rate, whereas humans
evolvedwithsubpopulationsandaheterogeneous
recombination rate (28, 29). FitCoal was not
found to overestimate such severity (Fig. 4) or
to falsely detect a bottleneck when examining
the effects of continuous and pulsed introgres-
sions among existing and ghost populations
(figs. S18 to S21). Therefore, the discovery of
the ancient severe bottleneck was not due to
overfitting the data in FitCoal analyses.
Discussion
In this study, we developed FitCoal, a model-
flexible method for demographic inference.
One key feature of FitCoal is that the expected
Hu et al., Science 381, 979–984 (2023)
1 September 2023
4 of 6
Fig. 4. Verification of the severe bottleneck. (A) Bottleneck I model,
mimicking the true population size history of African agriculturalist populations.
(B) Bottleneck II model, mimicking the inferred population size history of non-
African populations. (C) Bottleneck III model, mimicking the true population
size history of non-African populations. The gap between actual and FitCoal
estimated population size is indicated by the black dashed circle and arrowhead.
(D) Bottleneck IV model, mimicking a population with an exponential reduction
in size 1.5 million years ago. (E) Bottleneck V model, mimicking a population
with a moderate bottleneck. (F) Bottleneck VI model, mimicking a population
with a weak bottleneck. Black lines represent the true models. Thick red lines
represent the medians of FitCoal estimated population sizes over time; thin
red lines represent 2.5 and 97.5 percentiles of FitCoal estimated population
sizes over time. Blue, green, yellow, and gray lines represent the medians of
10 runs each of SMC++, Stairway Plot, PSMC, and Relate. The mutation rate is
assumed to be 1.2 × 10–8 per base per generation, and a generation time is
assumed to be 24 years. Numbers of simulated sequences are 188 in bottlenecks
I, IV, V, and VI and 194 in bottlenecks II and III. The lengths of simulated
sequences are 800 Mb each.
RESEARCH
|
RESEARCH ARTICLE
Downloaded from https://www.science.org at The Medical Library of the Chinese Pla on September 03, 2023

branch lengths can be accurately determined
for an SFS under an arbitrary demographic
model, which allows precise calculation of the
likelihood. Analyses by FitCoal are, in most
cases, less time consuming than those by other
methods such as PSMC, SMC++, Stairway Plot,
and Relate (28/32 = 87.5%) (tables S11 and S12).
By discarding rare and high-frequency muta-
tions, FitCoal can avoid the effects of sequencing
errors or hitchhiking due to positive selection
without losing its inference accuracy. Because
both instantaneous and exponential changes
are allowed within each inference time inter-
val, FitCoal can reveal the dynamic of population
size precisely. Because coalescent events be-
come rare when tracing backward in time, the
length of inference time interval is usually set
to increase progressively (6–8). Although this
strategy can capture recent population size
changes, it may miss ancient ones. Therefore,
FitCoal inference time intervals are allowed to
vary during demographic inference, and FitCoal
can make a fast and accurate inference of recent
and ancient population size histories.
The most important discovery with FitCoal
is that human ancestors went through a se-
vere bottleneck in the late Lower Pleistocene
(Fig. 5). This ancient severe bottleneck was
directly found in all 10 African populations,
but only a weak signal of the existence of such
was detected in all 40 non-African popula-
tions. This observation is consistent with the
coalescent theory and the occurrence of the
out-of-Africa dispersal. Results of our large-
scale simulations demonstrated that FitCoal
did not falsely infer the bottleneck because of
positive selections (figs. S6 and S22) or pop-
ulation structure (fig. S18 to S21). Because we
observed no overfitting cases and results ob-
tained by examining different sets of genomic
regions (Fig. 3, A and C, and fig. S23) were sim-
ilar, the existence of the ancient severe bottle-
neck was ascertained.
Our results indicate that the ancient severe
bottleneck lasted for approximately 117 kyr
(Fig. 5), and that about 98.7% of human an-
cestors were lost at the beginning of the bot-
tleneck, thus threatening our ancestors with
extinction. The estimated effective population
size during the bottleneck period was only
1280 breeding individuals, which was compa-
rable to the effective population sizes of other
endangered mammals (30, 31). This size (1280)
might have been overestimated because of hid-
den population structure (32). Naturally occur-
ring population size fluctuations might have
further increased the extinction risk for our
ancestors during the bottleneck. The bottleneck
could also have increased the inbreeding level
of our ancestors, thus contributing to the 65.85%
loss in present-day human genetic diversity.
The ancient population size reduction that
occurred around 930 kyr BP was likely driven
by climatic changes during the Early to Mid-
dle Pleistocene transition (33, 34). During this
transition period known as the “0.9 Ma event”
(Ma,millionyearsago),glaciationswerechanged
from predominantly short-term to long-term
events with more extreme thermic intensity,
especially at the peak of glaciation. This event
resulted in a decrease in marine surface tem-
perature to the lowest that occurred during the
entire transition period (33), with an inferred
long period of drought and extensive wildlife
turnover in Africa and Eurasia (35).
The existence of the ancient severe bottle-
neck could explain the extreme scarcity of the
available hominin fossil record in Africa and
Eurasia between 950 and 650 kyr BP (Fig. 5 and
fig. S24). In Africa, only a few fossil specimens
dated in this time period have been found, in-
cluding the cranial fragments from Gombore
in Ethiopia and the fossil samples from Tighenif
in Algeria (36, 37). Although the taxonomic sta-
tuses of these fossils are still not clear, they have
features resembling those of later fossils at-
tributed to Homo heidelbergensis. They are dif-
ferent from the coeval Homo antecessor from a
paleoanthropological site in Spain (Atapuerca,
Gran Dolina), and some scholars considered
H. antecessor as a possible alternative for the last
common ancestor (LCA) (38). During the same
chronological interval, the East Asian fossil
record contains specimens identified as Homo
erectus (39). It does not appear that East Asian
H. erectus is connected to the ancient severe
bottleneck because it is unlikely to have contrib-
uted to the lineage leading to modern humans
(38).Inaddition,coincidentwiththisbottleneck,
two ancestral chromosomes are believed to have
fused to form chromosome 2 in humans around
900 to 740 kyr BP (40, 41). Therefore, the ancient
severe bottleneck possibly marks a speciation
event leading to the emergence of the LCA
shared by Denisovans, Neanderthals, and mod-
ern humans, whose divergence has been dated
to about 765 to 550 kyr BP (38, 42, 43).
A rapid population recovery was detected in
all 10 African populations with a 20-fold in-
crease in size around 813 kyr BP. Control of
fire could be part of the explanation for this
population expansion, which is shown by the
early archaeological evidence found in Israel
dated about 790 kyr BP (44). Other factors, such
as climatic changes (33, 34), might also be a
driving force for this rapid population recovery.
The ancient severe bottleneck was not de-
tected in previous SFS-based analyses (6, 10, 12, 14).
This failure might be due to the use of prede-
fined demographic models. In this study, we
found that the likelihood must be accurately
calculated to detect the severe bottleneck (fig.
S1). The use of other methods such as Stairway
Plot—which may not have sufficient resolu-
tion power for estimation of ancient popula-
tion size history—is another possible reason for
the failure (6).
Our study revealed that an extremely small
human population lasted for about 117 kyr
around 930 to 813 kyr BP. Many questions
remain unanswered, such as where these indi-
viduals lived, how they overcame the cata-
strophic climate changes, and how the ancient
population remained so small for so long. Fur-
ther studies are warranted to investigate these
matters to obtain a more detailed picture of
human evolution during the Early to Middle
Pleistocene transition.
REFERENCES AND NOTES
1.
J. Galway-Witham, C. Stringer, Science 360, 1296–1298 (2018).
2.
C. M. Schlebusch et al., Science 358, 652–655 (2017).
3.
A.Meneganzin,T.Pievani,G. Manzi, Evol.Anthropol.31,199–212(2022).
4.
M. Stoneking, J. Krause, Nat. Rev. Genet. 12, 603–614 (2011).
5.
S. Ramachandran et al., Proc. Natl. Acad. Sci. U.S.A. 102,
15942–15947 (2005).
6.
X. Liu, Y. X. Fu, Nat. Genet. 47, 555–559 (2015).
7.
H. Li, R. Durbin, Nature 475, 493–496 (2011).
Hu et al., Science 381, 979–984 (2023)
1 September 2023
5 of 6
Fig. 5. Schematic diagram of
human population size history.
Both African (light green) and non-
African (light blue) populations
are presented. The width of the boxes
represents the effective population
size (i.e., the number of breeding
individuals) with naturally occurred
fluctuations. The occurrence time of
the out-of-Africa dispersal and the
divergence between African and
non-African populations are indicated.
The gray-shaded time duration
indicates the Early to Middle Pleis-
tocene transition between 1250 and
700 kyr BP. The red arrow indicates
the peak of glaciation during the
transition (i.e., the 0.9 Ma event).
The ancient severe bottleneck
inferred in this study is highlighted. The gap in the available African hominin fossil record and an indicative
chronology for H. erectus, the LCA, and H. sapiens are shown. The estimated time period in which two
ancestral chromosomes (chromosome, Chr.) fused to become one is also shown on the right.
RESEARCH
|
RESEARCH ARTICLE
Downloaded from https://www.science.org at The Medical Library of the Chinese Pla on September 03, 2023

J. Terhorst, J. A. Kamm, Y. S. Song, Nat. Genet. 49, 303–309 (2017).
9.
H. Li, W. Stephan, PLOS Genet. 2, e166 (2006).
10. L. Excoffier, I. Dupanloup, E. Huerta-Sánchez, V. C. Sousa,
M. Foll, PLOS Genet. 9, e1003905 (2013).
11. R. C. Griffiths, S. Tavaré, Math. Comput. Model. 23, 141–158 (1996).
12. R. N. Gutenkunst, R. D. Hernandez, S. H. Williamson,
C. D. Bustamante, PLOS Genet. 5, e1000695 (2009).
13. Y. X. Fu, Theor. Popul. Biol. 145, 95–108 (2022).
14. J. Jouganous, W. Long, A. P. Ragsdale, S. Gravel, Genetics 206,
1549–1567 (2017).
15. Y. X. Fu, Theor. Popul. Biol. 48, 172–197 (1995).
16. A. Polanski, M. Kimmel, Genetics 165, 427–436 (2003).
17. D. Zivković, T. Wiehe, Genetics 180, 341–357 (2008).
18. A. Polanski, A. Bobrowski, M. Kimmel, Theor. Popul. Biol. 63,
33–40 (2003).
19. A. Bhaskar, Y. X. R. Wang, Y. S. Song, Genome Res. 25,
268–279 (2015).
20. A. Scally, R. Durbin, Nat. Rev. Genet. 13, 745–753 (2012).
21. A. Kong et al., Nature 488, 471–475 (2012).
22. A. Bhaskar, Y. S. Song, Ann. Stat. 42, 2469–2493 (2014).
23. The 1000 Genomes Project Consortium, Nature 526, 68–74 (2015).
24. A. Bergström et al., Science 367, eaay5012 (2020).
25. J. C. Fay, C. I. Wu, Genetics 155, 1405–1413 (2000).
26. H. C. Harpending et al., Proc. Natl. Acad. Sci. U.S.A. 95,
1961–1967 (1998).
27. L. Speidel, M. Forest, S. Shi, S. R. Myers, Nat. Genet. 51,
1321–1329 (2019).
28. E. M. L. Scerri et al., Trends Ecol. Evol. 33, 582–594 (2018).
29. Z. Hao, P. Du, Y. H. Pan, H. Li, Hum. Genet. 141, 273–281 (2022).
30. H. Li et al., Proc. Natl. Acad. Sci. U.S.A. 113, 14079–14084 (2016).
31. Y. X. Chen et al., Zool. Res. 43, 523–527 (2022).
32. J. Wakeley, Genetics 153, 1863–1871 (1999).
33. M. J. Head, P. L. Gibbard, in Early-Middle Pleistocene
Transitions: The Land–Ocean Evidence, M. J. Head, P. L. Gibbard,
Eds. (Geological Society of London, 2005), pp. 1–18.
34. P. U. Clark et al., Quat. Sci. Rev. 25, 3150–3184 (2006).
35. M. J. Head, B. Pillans, S. A. Farquhar, Episodes 31, 255–259 (2008).
36. A. Profico, F. Di Vincenzo, L. Gagliardi, M. Piperno, G. Manzi,
J. Anthropol. Sci. 94, 41–63 (2016).
37. A. Mounier, F. Marchal, S. Condemi, J. Hum. Evol. 56, 219–246
(2009).
38. A. Bergström, C. Stringer, M. Hajdinjak, E. M. L. Scerri,
P. Skoglund, Nature 590, 229–237 (2021).
39. J.-J. Bahain et al., Anthropologie 121, 215–233 (2017).
40. B. Poszewiecka, K. Gogolewski, P. Stankiewicz, A. Gambin,
BMC Genomics 23, 616 (2022).
41. T. R. Dreszer, G. D. Wall, D. Haussler, K. S. Pollard, Genome Res. 17,
1420–1430 (2007).
42. K. Prüfer et al., Nature 505, 43–49 (2014).
43. D. Reich et al., Nature 468, 1053–1060 (2010).
44. N. Goren-Inbar et al., Science 304, 725–727 (2004).
45. H. Li, Genomic inference of a severe human bottleneck during
the Early to Middle Pleistocene transition, version 3, Mendeley
Data, 2023; https://doi.org/10.17632/xmf5r8nzrn.3.
46. W. Hu et al., FitCoal v1.1., Zenodo, 2023; https://doi.org/
10.5281/zenodo.7857456.
ACKNOWLEDGMENTS
We thank D. Živković for sharing his codes to calculate the expected
branch lengths, X. Liu for providing simulation results, and Bio-Med
Big Data Center at Shanghai Institute of Nutrition and Health for
providing a computing facility. Funding: National Natural Science
Foundation of China grants 32270674, 91131010, and 91731304
(H.L.); National Natural Science Foundation of China grant 31100273
(Y.H.P.); National Natural Science Foundation of China grants
91631304 and 32130011 (Y.X.F.); National Natural Science
Foundation of China grant 82171801 (Z.H.); Strategic Priority Research
Program of the Chinese Academy of Sciences grant XDPB17 (H.L.);
National Key Research and Development Project grant 2022YFF1203202
(H.L.); National Institutes of Health grant R01HG009524 (Y.X.F.);
Education Bureau of Jinan and Shandong First Medical University grant
JNSX2021046 (Z.H.); Key Laboratory of Brain Functional Genomics at
East China Normal University grant 20SKBFGK2 (Y.H.P. and H.L.);
Shanghai Institute of Nutrition and Health grant JBGSRWBD-SINH-2021-
10 (H.L.); China Postdoctoral Science Foundation grant
2022M711978 (Z.H.); Shandong Provincial Natural Science
Foundation grant ZR2022QC062 (Z.H.); and Shandong Provincial
Postdoctoral Innovation Talent Support Program grant
SDBX2022012 (Z.H.). Author contributions: Conceptualization:
W.H., Z.H., F.D.V., G.M., Y.X.F., Y.H.P., and H.L.; Methodology: W.H.,
Z.H., Y.H.P., and H.L.; Software: W.H., Z.H., and H.L.; Investigation:
W.H., Z.H., P.D., F.D.V., G.M., J.C., Y.X.F., Y.H.P., and H.L.; Visualization:
W.H., Z.H., P.D., J.C., F.D.V., and G.M.; Funding acquisition: Z.H., Y.X.F.,
Y.H.P., and H.L.; Supervision: G.M., Y.H.P., and H.L.; Writing: W.H.,
Z.H., P.D., F.D.V., G.M., Y.X.F., Y.H.P., and H.L. Competing interests:
The authors declare that they have no competing interests. Data
and materials availability: Raw data are deposited at Mendeley (45),
and FitCoal is archived at Zenodo (46). The download links for
1000GP and HGDP-CEPH data are available in table S2. License
information: Copyright © 2023 the authors, some rights reserved;
exclusive licensee American Association for the Advancement of
Science. No claim to original US government works. https://www.
science.org/about/science-licenses-journal-article-reuse
SUPPLEMENTARY MATERIALS
science.org/doi/10.1126/science.abq7487
Materials and Methods
Supplementary Text
Figs. S1 to S60
Tables S1 to S23
References (47–96)
MDAR Reproducibility Checklist
Submitted 28 April 2022; resubmitted 7 February 2023
Accepted 11 July 2023
10.1126/science.abq7487
Hu et al., Science 381, 979–984 (2023)
1 September 2023
6 of 6
RESEARCH
|
RESEARCH ARTICLE
Downloaded from https://www.science.org at The Medical Library of the Chinese Pla on September 03, 2023

Use of this article is subject to the Terms of service
Science (ISSN ) is published by the American Association for the Advancement of Science. 1200 New York Avenue NW, Washington, DC
20005. The title Science is a registered trademark of AAAS.
Copyright © 2023 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim
to original U.S. Government Works
Genomic inference of a severe human bottleneck during the Early to Middle
Pleistocene transition
Wangjie Hu, Ziqian Hao, Pengyuan Du, Fabio Di Vincenzo, Giorgio Manzi, Jialong Cui, Yun-Xin Fu, Yi-Hsuan Pan, and
Haipeng Li
Science, 381 (6661), .
DOI: 10.1126/science.abq7487
Editor’s summary
Today, there are more than 8 billion human beings on the planet. We dominate Earth’s landscapes, and our activities
are driving large numbers of other species to extinction. Had a researcher looked at the world sometime between
800,000 and 900,000 years ago, however, the picture would have been quite different. Hu et al. used a newly
developed coalescent model to predict past human population sizes from more than 3000 present-day human
genomes (see the Perspective by Ashton and Stringer). The model detected a reduction in the population size of our
ancestors from about 100,000 to about 1000 individuals, which persisted for about 100,000 years. The decline appears
to have coincided with both major climate change and subsequent speciation events. —Sacha Vignieri
View the article online
https://www.science.org/doi/10.1126/science.abq7487
Permissions
https://www.science.org/help/reprints-and-permissions
Downloaded from https://www.science.org at The Medical Library of the Chinese Pla on September 03, 2023