Nature | www.nature.com | 1
Article
Human origins in a southern African 
palaeo-wetland and first migrations
Eva K. F. Chan1,2, Axel Timmermann3,4*, Benedetta F. Baldi1, Andy E. Moore5, Ruth J. Lyons1, 
Sun-Seon Lee3,4, Anton M. F. Kalsbeek1, Desiree C. Petersen1,11, Hannes Rautenbach6,7,12,  
Hagen E. A. Förtsch8, M. S. Riana Bornman7 & Vanessa M. Hayes1,2,7,9,10*
Anatomically modern humans originated in Africa around 200 thousand years ago 
(ka)1–4. Although some of the oldest skeletal remains suggest an eastern African origin2, 
southern Africa is home to contemporary populations that represent the earliest 
branch of human genetic phylogeny5,6. Here we generate, to our knowledge, the largest 
resource for the poorly represented and deepest-rooting maternal L0 mitochondrial 
DNA branch (198 new mitogenomes for a total of 1,217 mitogenomes) from 
contemporary southern Africans and show the geographical isolation of L0d1’2, L0k 
and L0g KhoeSan descendants south of the Zambezi river in Africa. By establishing 
mitogenomic timelines, frequencies and dispersals, we show that the L0 lineage 
emerged within the residual Makgadikgadi–Okavango palaeo-wetland of southern 
Africa7, approximately 200 ka (95% confidence interval, 240–165 ka). Genetic 
divergence points to a sustained 70,000-year-long existence of the L0 lineage before 
an out-of-homeland northeast–southwest dispersal between 130 and 110 ka. Palaeo-
climate proxy and model data suggest that increased humidity opened green 
corridors, first to the northeast then to the southwest. Subsequent drying of the 
homeland corresponds to a sustained effective population size (L0k), whereas wet–dry 
cycles and probable adaptation to marine foraging allowed the southwestern migrants 
to achieve population growth (L0d1’2), as supported by extensive south-coastal 
archaeological evidence8–10. Taken together, we propose a southern African origin of 
anatomically modern humans with sustained homeland occupation before the first 
migrations of people that appear to have been driven by regional climate changes.
Southern Africa has long been considered to be one of the regions in 
which anatomically modern humans (AMHs) originated. Home to con-
temporary populations who represent the earliest human lineages, evo-
lutionary time estimates have largely been based on mitochondrial DNA 
(mitogenomes)1,6. The maternal human phylogenetic tree consists of 
two major branches, the extensive L1’6—which includes the out-of-Africa 
ancestral L3 sub-branch (or haplogroup)—and the rare deep-rooting 
L0. The L0 lineage is predominated by southern African haplogroups: 
L0d, L0k and the recently described L0g6. By contrast, the rare L0f and 
common L0a lineages are dispersed throughout sub-Saharan Africa1,3,6. 
Through L0 pre-screening, we identified 198 southern Africans with 
poorly represented haplogroups for whom the mitogenome was 
sequenced (Supplementary Table 1), allowing for a combined analysis 
of 1,217 mitogenomes (Fig. 1a and Extended Data Table 1).
We ethno-linguistically classified study participants as KhoeSan—
southern African populations who traditionally practiced foraging 
and spoke languages containing ‘click’ consonants—or non-KhoeSan 
individuals. Non-KhoeSan who have KhoeSan-derived L0 mitogenomes 
are referred to in this study as KhoeSan ancestral, with further geo-
graphical classification (Fig. 1b and Extended Data Table 2; terminol-
ogy pertaining to southern African KhoeSan populations is complex 
and contentious, see Methods for further discussion). Contemporary 
KhoeSan include Kalahari KhoeSan (Kx’a, Tuu and central Khoe–Kwadi 
speakers) and west-coastal KhoeSan (Khoe–Kwadi Nama speakers)11. 
Peoples who speak Southern Bantu languages, who migrated down 
the east coast of Africa around 1,500 years ago, may have acquired an 
east-coastal KhoeSan heritage12. The arrival of European colonists to 
the Cape in mid-1600s gave rise to the South African Coloured and 
Namibian Baster populations (of Eurasian and indigenous descent), 
who acquired a Cape KhoeSan heritage13. Excluding the east African 
Sandawe and Hadza (whose languages also contain click consonants), 
indigenous KhoeSan populations appear to be absent northeast of the 
Zambezi river, supported by the lack of skeletal remains representing the 
KhoeSan-like hunter–forager morphology14. We classified the 198 new 
https://doi.org/10.1038/s41586-019-1714-1
Received: 30 October 2018
Accepted: 24 September 2019
Published online: Ńʼn =«ìت³ã ŃŁłŊ
1Genomics and Epigenetics Division, Garvan Institute of Medical Research, Sydney, New South Wales, Australia. 2St Vincent’s Clinical School, University of New South Wales, Sydney, New 
South Wales, Australia. 3Center for Climate Physics, Institute for Basic Science, Busan, South Korea. 4Pusan National University, Busan, South Korea. 5Department of Geology, Rhodes University, 
Grahamstown, South Africa. 6Climate Change and Variability, South African Weather Service, Pretoria, South Africa. 7School of Health Systems and Public Health, University of Pretoria, 
Pretoria, South Africa. 8Windhoek Central Hospital, Windhoek, Namibia. 9Faculty of Health Sciences, University of Limpopo, Sovenga, South Africa. 10Central Clinical School, University of 
Sydney, Sydney, New South Wales, Australia. 11Present address: The Centre for Proteomic and Genomic Research, Cape Town, South Africa. 12Present address: Akademia, Johannesburg, South 
Africa. *e-mail: timmermann@pusan.ac.kr; v.hayes@garvan.org.au
2 | Nature | www.nature.com
Article
mitogenomes as Kalahari (n = 18), west-coastal (n = 21), Cape (n = 109) and 
east-coastal (n = 29) KhoeSan, or non-KhoeSan (Bantu, n = 19), although 
two mitogenomes were classed as unknown. Using these identifiers, 
we provide a best-fit classification for all 1,217 L0 mitogenomes (Sup-
plementary Table 2).
Phylogenetic analysis confirms the major L0 haplogroups, with the 
exclusion of L0b (Extended Data Fig. 1). Using a subset of 461 mitog-
enomes, including all of the rare lineages, we establish the coalescence 
times within the L0 lineage (Fig. 2a and Supplementary Table 3) and use 
the complete dataset to reconstruct geographical dispersals (Fig. 2b). We 
redefine the emergence of the L0 lineage to 50–25 thousand years (kyr) 
before previous estimates1,6, around 200 ka (95% confidence interval, 
240–165 ka). L0d’k (n = 309; coalesced around 187 ka (the number of 
mitogenomes and the coalescence time are provided for each lineage)) 
is largely KhoeSan-specific, emerging approximately 20 kyr before 
the widely dispersed L0a’b’f’g sister branch (n = 152; around 164 ka). 
Although the exact branch resolution for L0k remains undetermined, we 
observe a preference for L0d’k (posterior probability of approximately 
0.6) over L0a’b’f’g’k (posterior probability of about 0.4). Irrespective 
of this, the L0k (n = 113) lineage appears to remain stable for around 
130 kyr before diverging into the Kalahari-specific L0k1 lineage, which 
is predominated by L0k1a (85 out of 94), and rarer L0k1b and L0k2 line-
ages distributed around the Zambezi river (Extended Data Fig. 2a). The 
L0d lineage remains stable for almost 60 kyr before splitting into the 
KhoeSan-specific L0d1’2 and rarer L0d3 lineages.
Coalescing around 113 ka, L0d2 (n = 226) emerges approximately 
15 kyr before L0d1 (n = 452). Within L0d2 (emerging about 91 ka), L0d2c 
diverged the earliest (n = 53; around 84 ka) with a broad and almost even 
KhoeSan-regional distribution (Extended Data Fig. 3 and Supplementary 
Table 4). In 2014, we derived an ancient L0d2c1c mitogenome from a 
sample of the skeleton of an approximately 2,330-year-old Cape-coastal 
marine forager (St Helena (StHe)/UCT606)15. Predating archaeologi-
cal evidence for sheep herding in the region12,16, we proposed that this 
L0d2c sub-clade represented a pre-pastoral indigenous southern Afri-
can lineage. Recently, whole-genome sequencing confirmed a unique 
southern African heritage, whereas two younger (less than 2 kyr old) 
Cape skeletons showed a genetic link to eastern Africa and the associ-
ated pastoralist migration17. Previously, an overrepresentation of the 
L0d2b (28 out of 44; around 65 ka) and L0d2a (62 out of 118; around 
60 ka) lineages within the Kalahari KhoeSan has been observed; how-
ever, by doubling the contribution of the L0d2d (6 out of 11) lineage, 
we show a broad southern African distribution (Extended Data Fig. 3 
and Supplementary Tables 5, 6). While L0d1 is also spread throughout 
the KhoeSan-regional identifier, we show notable overrepresentation 
of the L0d1b (104 out of 174; about 69 ka) and L0d1c (151 out of 184; 
approximately 59 ka) lineages within the Kalahari and of the L0d1a 
(32 out of 91; around 44 ka) lineage within the Cape (Extended Data 
Fig. 4). We contribute two new KhoeSan-ancestral L0d1d mitogenomes 
to the single published mitogenome6.
In contrast to L01’2, the L0d3 lineage is not specific to southern Africa. 
Although L0d3b (around 30 ka) appears to be KhoeSan-specific, the rarer 
L0d3a (about 42 ka) lineage is exclusively found north of the Zambezi 
river. Notably, three out of six L0d3a mitogenomes were derived from 
east African Sandawe individuals. Our data support previous studies 
that have suggested a genetic link between east Africa and the earliest 
southern Africans17, who last shared a common ancestor around 59 ka. 
By adding a large number of mitogenomes (27 out of 40) to the L0d3 
lineage, we observe overrepresentation of L0d3b in the Cape KhoeSan 
identifier (21 out of 34) (Extended Data Fig. 2b and Supplementary 
Table 7). Using a previously reported identifier that distinguishes mater-
nal KhoeSan ancestry for the Coloured and Baster populations13, we 
show that the L0d3b lineage is specific to the Coloured population, 
whereas the new L0d2b1a2a sub-clade is specific to the Baster popula-
tion (Extended Data Fig. 3b).
Within the L0a’b’f’g lineage, L0f is highly divergent (emerging around 
125 ka; 95% confidence interval,149–101 ka). By including a further five 
L0f mitogenomes, we were able to show that L0f1 (13 out of 27; around 
113 ka) predominates south and L0f2’3 (14 out of 27; about 121 ka) north 
of the Zambezi river (Extended Data Fig. 2c and Supplementary Table 8). 
Within L0f1, we recognize three new branches: the northeast sister clades 
L0f1c (Zambian) and L0f1b (Tanzanian), and the   South African clade 
L0f1a (n = 8). Lack of L0f representation within contemporary KhoeSan 
suggests that the presence of L0f1a within South Africa is probably 
a result of more recent east-coastal agropastoral back-migration. 
While the L0a’g lineages coalesce around 117 ka (95% confidence inter-
val, 145–94 ka), contributing 19 southern African to 347 L0a mitoge-
nomes, we concur that the L0a lineage probably diverged northeast 
of the Zambezi river (around 85 ka) and spread throughout Africa3; the 
a b
Published: 1,019
This study:  198
Total: 1,217
Archaeological caves
11
4
1
4 8
6
6
5
1
8
1
9
3
1
19
124
63
211
322
2
18
77
114
121
DRC Tanzania
Zambia
Zambezi river
Zimbabwe
Mozam-
bique
South Africa
Botswana
Namibia
Angola East-coastal
KhoeSan
Kalahari
KhoeSan
Lower Zambezi
Middle Zambezi
Upper Zambezi
Makgadikgadi
pan
Limpopo
Okavango
Kafue
Okavango delta
Non-KhoeSan
Cape KhoeSan
West-coastal
KhoeSan
St Helena skeleton
Fig. 1 | Geographical distribution of 1,217 L0 mitogenomes. a, Countries 
within (n = 1,139) or outside (n = 78) Africa from which L0 mitogenomes were 
sourced, including 198 new L0 mitogenomes (black numbers on the map). DRC, 
Democratic Republic of the Congo. b, Present-day southern Africa showing the 
geographical distribution of KhoeSan population identifiers defined as 
KhoeSan (orange), Kalahari and west-coastal; KhoeSan ancestral (green), Cape 
or east-coastal. The Zambezi river provides a geographical division between the 
KhoeSan and mostly non-KhoeSan population identifiers. Maps were generated 
in the R package ‘maps’ v.3.3.037.
Nature | www.nature.com | 3
southern representation of the L0a1b and L0a2a lineages are probably a 
result of a Bantu back-migration (Extended Data Fig. 5). First described 
in a Kx’a-speaking hunter-gatherer6, we now contribute three additional 
and reclassify five published mitogenomes as L0g (Extended Data Fig. 2d 
and Supplementary Table 9). As the L0g lineage has a broad KhoeSan 
and KhoeSan-ancestral distribution, we hypothesize that this lineage 
diverged southwest of the Zambezi river (around 69 ka), similar to the 
L0d1’2 lineage.
Our results suggest that the greater Zambezi river basin, particu-
larly the Kalahari region, had a critical role in shaping the emergence 
and prehistory of AMHs. Now a semi-desert, this region consists of salt 
pans within northern Botswana that represent desiccated vestiges of 
palaeo-lake Makgadikgadi, which at its peak in the early Pleistocene 
would have been the largest lake in Africa7,18. Contraction of the Mak-
gadikgadi palaeo-lake during the Middle Pleistocene was accompanied 
by development of the Okavango delta as a result of neotectonic rift-
ing, which—together with smaller lakes from the upper Zambezi to the 
Kafue rivers—would have created a vast residual wetland favourable for 
habitation by humans and mammals more broadly19 (Fig. 2c). Today, 
the harsh Kalahari climate and oxygen-rich salt pans are not ideal for 
fossil and pollen preservation, respectively. However, period-relevant 
lithic artefacts are documented from the Makgadikgadi pans and sur-
roundings7,20,21, while palynology suggests that this region was once a 
grassland and forest biome22. Our data further suggest that the Mak-
gadikgadi–Okavango palaeo-wetland sustained the existence of AMHs 
for around 70 kyr, supported by mitochondrial data of ancestral giraffe, 
lion and zebra23–25, before out-of-homeland migrations split the founder 
homeland populations of the L0d, L0f and L0a’g lineages.
Southwest of their homeland, the L0d1’2 lineage experienced 
episodic splits and showed a broad south-coastal occupation of the 
emerged sub-populations, whereas the ancestors of the L0g lineage were 
less successful. Bayesian skyline plot (BSP) (Fig. 2c) analysis confirms 
effective population growth for the L0d1’2 lineage (BSP L0d1’2), whereas 
extensive archaeological evidence indicates cognitively modern human 
behaviour at the southern tip of Africa8–10 between approximately 100 
and 60 ka, together with an associated increase in the density of time-
appropriate archaeological sites in coastal compared to inland regions26. 
Northeast of their homeland, the L0d3 and L0f lineages are less success-
ful, whereas the L0a lineage underwent considerable diversification, 
which post-dates the out-of-Africa migration (BSP L0a; Fig. 2c). The 
northeast migration route is further supported by the appearance of 
data-appropriate archaeological sites26. Within their homeland, the 
population carrying the L0k lineage sustained a constant effective popu-
lation size (BSP L0k), as did the Kalahari-predominant L0d2b, L0d2a and 
L0d1c lineages. Although the presence of L0k in Zambia has been sug-
gested to represent contact with an ancient pre-Bantu population27, we 
propose that these rare lineages represent an ancient out-of-homeland 
branch of the ancestral KhoeSan population.
Orbitally driven large-scale hydroclimate variations have been pro-
posed as a contributor of early human migrations28,29. In some studies, 
wetter conditions and resulting ‘green corridors’ have been proposed 
to explain the out-of-Africa migration (a ‘pull’ scenario), whereas others 
have proposed that drier conditions and resulting food shortages forced 
dispersals (a ‘push’ scenario)30. To determine whether our predicted 
homeland isolation and major dispersals may have been driven by cli-
mate shifts, we analysed four key palaeo-hydroclimate datasets29,31–33, 
along with a transient 784-kyr-long glacial–interglacial simulation con-
ducted with the LOVECLIM Earth system model28 (Fig. 3). Although 
limited by available palaeo-proxy records and a climate model of inter-
mediate complexity, we observe a considerable degree of coherence on 
orbital timescales (Extended Data Fig. 6). During the homeland period 
(200–130 ka), palaeo-data link the 21-kyr-long precession cycle, which 
St. Helena
skeleton
Archaeological
caves
Makgadikgadi–
Okavango
wetlandsL0k
L0g
L0a
L0d3 L0f
L0d1’2
L0
Southwest
migration
Northeast
migration
c
100
1,000
10,000
100,000
L0a
70 40 10 0
Date (ka)
100
1,000
10,000
100,000
L0d1’2
70 40 10 0
100
1,000
10,000
100,000
L0k
70 40 10 0
Date (ka)
Date (ka)
EP
S
EP
S
EP
S
L0f1a
(n = 8)
L0f1b
(n = 2)
L0f1c
(n = 3)
L0f1 
L0d3b
(n = 34)
L0d3a
(n = 6)
21
5
2
6
L0d3 
657
19
1
L0d1’2
L0k1a
(n = 94)
L0k2
(n = 11)
8 2
85
9 1
L0k b
0
100,000
50,000
200,000
150,000
250,000
Ye
ar
s 
be
fo
re
 p
re
se
nt
L0fL0d2b L0g
Som20 L0d2cL0d1c
L0b
L0d2a
L0k1a2 L0aL0d1b
L0k1a1
L0d3b L0d2dL0d1d
L0k1bL0d1a
L0d3a L0k2
0.9997
0.4746
0.9986
0.9746
0.2403
0.9283
0.9997
0.9973
0.6008
0.9998
0.9971
0.991
a
Fig. 2 | L0 phylogenetic tree, geographical distributions of the major 
southern African L0 haplogroup and out-of-homeland L0 dispersal routes. 
a, Phylogenetic branching and coalescence times derived from a focused subset 
of 461 L0 mitogenomes, including all rare branches, and anchored to 
Neanderthals (Homo neanderthalensis; n = 7). The Somalian-derived (Som20) 
L0d3 mitogenome3 could not be assigned. b, Geographical distribution 
(identifiers described in Fig. 1b) for all KhoeSan-specific mitogenomes (out of 
1,217): L0d3 (n = 40), L0d1’2 (n = 677, excluding one unknown), L0k (n = 105, 
excluding seven L0k1b and a single Yemen-derived L0k2), and L0f1 (n = 13). 
Predominant geographical representation (shaded regions), with region-
specific overflow represented by the total number of mitogenomes, including 
the country-specific representation north of the Zambezi river. c, Schematic 
map of southern Africa representing the Makgadikgadi–Okavango palaeo-
wetland sustained AMH homeland (200–130 ka), supported by archaeological 
data (represented by the trowel symbol)7 and genetic wildlife data (represented 
by the lion, zebra and giraffe symbols)23–25. The out-of-homeland migration  
(130–110 ka), results in the split of L0d with L0a’g and L0f divergence. L0d3, L0a 
and L0f migrate in a northeast direction, L0d1’2 and L0g migrate southwest, 
while L0k remains in the homeland. Insets show BSP analyses of effective 
population sizes (EPS) of major L0 haplogroups over time, predicting the 
maintenance of the homeland L0k population (orange), population growth for 
the broadly dispersed southwest L0d1’2 migrants (purple), which is supported 
by archaeological evidence (100–60 ka)8–10 and the StHe mitogenome15, while 
population growth of the northeast L0a migrants coincides with the out-of-
Africa migration (aqua). Maps were generated in the R package maps v.3.3.037.
4 | Nature | www.nature.com
Article
arises from a combination of Earth’s axis wobble and a slow rotation of 
Earth’s entire orbit around the Sun (Fig. 3a), with three wet–dry cycles 
(Fig. 3b). By contrast, the climate model simulates an extended drought, 
owing to a more pronounced eccentricity signal (Fig. 3e), suggestive of 
a wetland oasis in an otherwise vast harsh environment.
During the out-of-homeland period (130–110 ka), our model simula-
tion supports humid conditions to the northeast that facilitated the first 
dispersals, concurring with L0f coalescence (around 125 ka) (Fig. 3d). 
By contrast, the region southwest of the homeland experienced an 
approximately 15-kyr-long megadrought before an orbital shift created 
the favourable humid conditions that led to the dispersal of the L0d1’2 
lineage (around 113 ka) (Fig. 3f), which is also supported by palaeo-data 
(Fig. 3c). This is also around the time the northeast L0a and southwest 
L0g migrants last share a common ancestor (around 117 ka). During the 
last glacial period (approximately 100–11 ka), we observe a reduction in 
the amplitude of the changes in orbital-scale hydroclimate and overall 
drying within the homeland (Fig. 3b), whereas the southwest coastal 
hydroclimate was dominated by precessional variability and showed 
relatively agreeable environmental conditions (Fig. 3c, f). Notably, peri-
ods of deceleration and acceleration in the estimates of the effective 
population size of the L0d1’2 lineage coincide with regional changes in 
hydroclimate, further linking climate, population size and evolution.
We propose that the Makgadikgadi–Okavango palaeo-wetland was 
the possible homeland of AMHs. Although one cannot exclude the 
possibility of a polycentric origin34, this deltaic–lacustrine ecosystem 
would have provided an ideal geographical locality for the evolution 
and 70-kyr-long sustained existence of the deepest-branching maternal 
founder population of AMHs. Increased humid conditions, supported by 
palaeo-lake system reconstructions35, between 130 and 110 ka would have 
opened green corridors for successful northeast–southwest migrations, 
supporting a pull scenario. Drying within the homeland following the 
out-of-homeland period, supported by hydroclimate data (110–100 ka) 
and a model simulation (100–80 ka), would have created a push scenario, 
in which a reduced carrying capacity of the land would have increased 
pressure to seek out climatically more favourable regions. We propose 
that the southwest migrants maintained a successful coastal forager 
existence, while the northeast migrants—similar to the later-branching 
population of L1’6—gave rise to ancestral pastoral and farming popula-
tions. A recent publication36 provides further mitochondrial evidence 
to support the northeast out-of-homeland migration route and expan-
sion into eastern Africa around 70–60 ka. Revealing a southern African 
homeland for the emergence and extended subsistence of the L0 lineage, 
we propose that an out-of-homeland migration event, which was prob-
ably driven by astronomically induced regional shifts in hydroclimate, 
shaped the present-day ethnic and genetic diversity of modern humans.
Online content
Any methods, additional references, Nature Research reporting summa-
ries, source data, extended data, supplementary information, acknowl-
edgements, peer review information; details of author contributions 
and competing interests; and statements of data and code availability 
are available at https://doi.org/10.1038/s41586-019-1714-1.
1. Behar, D. M. et al. The dawn of human matrilineal diversity. Am. J. Hum. Genet. 82,  
1130–1140 (2008).
time [ka]
050100150200250Au
st
ra
l s
um
m
er
 in
so
la
tio
n 
(W
 p
er
 m
2 )
400
420
440
460
480
500
520
540
time [ka]
050100150200250
–2
–1
0
1
2
3
P
op
ul
at
io
n 
si
ze
 (L
0k
)
101
102
103
104
Date (ka)
050100150200250
H
yd
ro
cl
im
at
e 
(w
es
t)
–3
–2
–1
0
1
2
3
P
op
ul
at
io
n 
si
ze
 (L
0d
1’
2)
102
103
104
time [ka]
050100150200250
LO
VE
C
LI
M
 p
re
ci
pi
ta
tio
n 
(N
E)
-3
–2
–1
0
1
2
3
time [ka]
050150200250L
O
VE
C
LI
M
 p
re
ci
pi
ta
tio
n 
(h
om
el
an
d)
–2
–1
0
1
2
3
Date (ka)
050100150200250L
O
VE
C
LI
M
 p
re
ci
pi
ta
tio
n 
(S
W
)
–3
–2
–1
0
1
2
3
W
et
D
ry
L0d1’2
L0k
L0f
Obliquity
Precession
Eccentricity
Homeland Northeast migration 
Southwest migration 
W
et
D
ry
L0, L0d’k, L0a’b’f’g
Precession
H
yd
ro
cl
im
at
e 
(c
en
tra
l a
nd
 e
as
t)
a d
b e
c f
Makgadikgadi highstands
Fig. 3 | Reconstructed and simulated climatic conditions during the out- 
of-homeland migration. a, Austral summer insolation changes (blue) at 27° S.  
b, A hydroclimate composite of eastern and central southern Africa (shading) 
was obtained by averaging the Fe/K runoff record from core CD154-1006P31 and 
the Pretoria Salt Pan rainfall reconstruction29, extended from 250 to 190 ka (grey 
line). The plot shows the effective population size for homeland L0k as analysed 
by BSP (orange dashed lines). c, Southwestern hydroclimate reconstruction 
(shading) obtained by averaging normalized leaf wax data (MD08-3167)33 and 
the aridity index from cores (MD96-2094)32, for which the aridity record 
extended from 250 to 140 ka (grey line) and the effective population size of 
L0d1’2 as analysed by BSP is shown (purple dashed lines). d, Simulated 
LOVECLIM normalized precipitation changes (shading) northeast of the 
homeland (33° E, 13° S) and coalescence time probabilities for L0f haplogroup 
(blue bell curves). e, Same as for d, but for the homeland and coalescence 
probabilities for L0, L0d’k, L0a’b’f’g (black) and L0k haplogroups (orange).  
f, Same as for d, but for the area southwest of the homeland (17° E, 30° S) and 
L0d1’2 coalescence times (purple). Blue bars indicate predicted Makgadikgadi 
high stand phases35. NE, northeast; SW, southwest.
Nature | www.nature.com | 5
2. Brown, F. H., McDougall, I. & Fleagle, J. G. Correlation of the KHS tuff of the Kibish 
Formation to volcanic ash layers at other sites, and the age of early Homo sapiens (Omo I 
and Omo II). J. Hum. Evol. 63, 577–585 (2012).
3. Rito, T. et al. The first modern human dispersals across Africa. PLoS ONE 8, e80031 (2013).
4. Stringer, C. & Galway-Witham, J. On the origin of our species. Nature 546, 212–214 (2017).
5. Henn, B. M. et al. Hunter-gatherer genomic diversity suggests a southern African origin 
for modern humans. Proc. Natl Acad. Sci. USA 108, 5154–5162 (2011).
6. Chan, E. K. F. et al. Revised timeline and distribution of the earliest diverged human 
maternal lineages in southern Africa. PLoS ONE 10, e0121223 (2015).
7. Moore, A. E., Cotterill, F. P. D. & Eckardt, F. D. The evolution and ages of Makgadikgadi 
palaeo-lakes: consilient evidence from Kalahari drainage evolution south-central Africa. 
S. Afr. J. Geol. 115, 385–413 (2012).
8. Henshilwood, C. S. et al. A 100,000-year-old ochre-processing workshop at Blombos 
Cave, South Africa. Science 334, 219–222 (2011).
9. Douze, K., Wurz, S. & Henshilwood, C. S. Techno-cultural characterization of the MIS 5  
(c. 105–90 ka) lithic industries at Blombos cave, Southern Cape, South Africa. PLoS ONE 
10, e0142151 (2015).
10. Henshilwood, C. S. et al. An abstract drawing from the 73,000-year-old levels at Blombos 
Cave, South Africa. Nature 562, 115–118 (2018).
11. Güldemann, T. in Beyond ‘Khoisan’: historical relations in the Kalahari Basin (Current Issues 
in Linguistic Theory 330) (eds Güldemann, T. & Fehn, A.-M.) 330, 1–40 (John Benjamins, 
2014).
12. Lander, F. & Russell, T. The archaeological evidence for the appearance of pastoralism 
and farming in southern Africa. PLoS ONE 13, e0198941 (2018).
13. Petersen, D. C. et al. Complex patterns of genomic admixture within southern Africa. 
PLoS Genet. 9, e1003309 (2013).
14. Morris, A. G. Isolation and the origin of the Khoisan: late Pleistocene and early Holocene 
human evolution at the southern end of Africa. Hum. Evol. 17, 231–240 (2002).
15. Morris, A. G., Heinze, A., Chan, E. K. F., Smith, A. B. & Hayes, V. M. First ancient 
mitochondrial human genome from a prepastoralist southern African. Genome Biol. Evol. 
6, 2647–2653 (2014).
16. Pleurdeau, D. et al. “Of sheep and men”: earliest direct evidence of caprine domestication 
in southern Africa at Leopard Cave (Erongo, Namibia). PLoS ONE 7, e40340 (2012).
17. Skoglund, P. et al. Reconstructing prehistoric African population structure. Cell 171, 59–71 
(2017).
18. Eckardt, F. D. et al. Mapping the surface geomorphology of the Makgadikgadi Rift Zone 
(MRZ). Quat. Int. 404, 115–120 (2016).
19. Wrangham, R. W. in Interpreting the Past: Essays on Humans, Primates and Mammal 
Evolution (eds Pilbeam, D. R. et al.) 231–242 (Brill Academic, 2005).
20. Robbins, L. H. et al. The advent of herding in southern Africa: early AMS dates on 
domestic livestock from the Kalahari Desert. Curr. Anthropol. 46, 671–677 (2005).
21. Mackay, A., Stewart, B. A. & Chase, B. M. Coalescence and fragmentation in the late 
Pleistocene archaeology of southernmost Africa. J. Hum. Evol. 72, 26–51 (2014).
22. Scott, L. & Neumann, F. H. Pollen-interpreted palaeoenvironments associated with the 
Middle and Late Pleistocene peopling of Southern Africa. Quat. Int. 495, 169–184 (2018).
23. Bock, F. et al. Mitochondrial sequences reveal a clear separation between Angolan and 
South African giraffe along a cryptic rift valley. BMC Evol. Biol. 14, 219 (2014).
24. Pedersen, C. T. et al. A southern African origin and cryptic structure in the highly mobile 
Plains zebra. Nat. Ecol. Evol. 2, 491–498 (2018).
25. Moore, A. E. et al. Genetic evidence for contrasting wetland and savannah habitat 
specializations in different populations of lions (Panthera leo). J. Hered. 107, 101–103 
(2016).
26. Blome, M. W., Cohen, A. S., Tryon, C. A., Brooks, A. S. & Russell, J. The environmental 
context for the origins of modern human diversity: a synthesis of regional variability in 
African climate 150,000–30,000 years ago. J. Hum. Evol. 62, 563–592 (2012).
27. Barbieri, C. et al. Ancient substructure in early mtDNA lineages of southern Africa. Am. J. 
Hum. Genet. 92, 285–292 (2013).
28. Timmermann, A. & Friedrich, T. Late Pleistocene climate drivers of early human migration. 
Nature 538, 92–95 (2016).
29. Partridge, T. C., Demenocal, P. B., Lorentz, S. A., Paiker, M. J. & Vogel, J. C. Orbital forcing 
of climate over South Africa: a 200,000-year rainfall record from the Pretoria saltpan. 
Quat. Sci. Rev. 16, 1125–1133 (1997).
30. Tierney, J. E., deMenocal, P. B. & Zander, P. D. A climatic context for the out-of-Africa 
migration. Geology 45, 1023–1026 (2017).
31. Simon, M. H. et al. Eastern South African hydroclimate over the past 270,000 years.  
Sci. Rep. 5, 18153 (2015).
32. Stuut, J.-B. W. et al. A 300-kyr record of aridity and wind strength in southwestern Africa: 
inferences from grain-size distributions of sediments on Walvis Ridge, SE Atlantic.  
Mar. Geol. 180, 221–233 (2002).
33. Collins, J. A., Schefuß, E., Govin, A., Mulitza, S. & Tiedemann, R. Insolation and glacial–
interglacial control on southwestern African hydroclimate over the past 140 000 years. 
Earth Planet. Sci. Lett. 398, 1–10 (2014).
34. Scerri, E. M. L. et al. Did our species evolve in subdivided populations across Africa, and 
why does it matter? Trends Ecol. Evol. 33, 582–594 (2018).
35. Burrough, S. L., Thomas, D. S. G. & Bailey, R. M. Mega-lake in the Kalahari: a late 
Pleistoscene record of the palaeolake Makgadikgadi system. Quat. Sci. Rev. 28, 1392–1411 
(2009).
36. Rito, T. et al. A dispersal of Homo sapiens from southern to eastern Africa immediately 
preceded the out-of-Africa migration. Sci. Rep. 9, 4728 (2019).
37. Becker, R. A. & Wilks, A. R. maps: Draw Geographical Maps. R package version 3.3.0 
https://cran.r-project.org/web/packages/maps/index.html (2018).
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in 
published maps and institutional affiliations.
© The Author(s), under exclusive licence to Springer Nature Limited 2019
Article
Methods
No statistical methods were used to predetermine sample size. The 
experiments were not randomized and investigators were not blinded 
to allocation during experiments and outcome assessment.
Statement on population identifiers
The authors acknowledge that population identifiers (or ethnic labels) 
have different meanings to different peoples across different countries 
and between and within different ethnic groups. During the apartheid 
rule, South Africans were grouped according to ethnic identities, which 
resulted in discrimination based on population identifiers such as Bantu 
or Coloured. In turn, others view the very same population identifiers 
with cultural identity and pride. In 2013, we performed a study led by 
a Coloured co-author to assess the sensitivity in self-identification as 
Coloured. Of 521 participants, 91.2% self-identified as Coloured, Cape 
Coloured or South African Coloured, while 8.8% elected against the 
use of Coloured for self-identification14. In turn, using such population 
identifiers within the context of the United States would be seen as 
derogatory and highly offensive. We have previously genetically profiled 
the Baster population of Namibia13 and again what could be to others 
a derogatory term, to the Baster community of Rehoboth in Namibia, 
the term is used with immense pride, who recognize themselves as a 
Republic with a national flag38.
In this study, the authors have used linguistics, supported by eth-
nicity, to provide population identification, with further historical, 
geographical and genetic classification for deriving maternal contribu-
tions (described in the next section). KhoeSan (or KhoeSaan) languages 
are grouped together due to their use of click consonants as a unique 
language identifier. Once spread across the entire southern African 
region, KhoeSan languages are today restricted largely to populations 
residing in Namibia and Botswana (and southern Angola), although two 
Tanzanian isolates, Sandawe and Hadza, are believed to be linguistically 
related click languages (or east African KhoeSan)39. ‘San’ literally means 
‘forager’ and Khoe means ‘person’; culturally, the KhoiSan identifier 
refers to hunter–foragers (San) or herders (Khoi). At times linguistic 
and cultural identities clash. For example, Nama and Hai$om peoples 
both speak Nama (a Khoe–Kwadi language), while culturally and histori-
cally these two populations are quite different, representing a herder 
and hunter-gatherer ancestry, respectively. Additionally, autosomal 
genetic data have been used to provide further insights into KhoeSan 
admixture and substructures, highlighting at a genetic level the histori-
cal differences between the Nama and Hai$om40. We have attempted to 
capture both ethnic and linguistic identifiers that best reflect population 
ancestry. In contrast to KhoeSan languages, most Bantu languages do 
not contain click consonants; however, exceptions exist within Southern 
African Bantu languages (for example, isiXhosa and isiZulu languages, 
which have borrowed click consonants from their KhoeSan neighbours). 
Spoken across the entire sub-Saharan Africa (up to 500 groups), the 
Guthrie classification of languages further identifies the S-zone or South-
ern Bantu (South Africa, Zimbabwe, southern Mozambique and most 
of Botswana) and the R-zone or Southwest Bantu languages (northern 
Namibia, southern Angola and northwest Botswana)41, which are of 
relevance to this study.
Ethics statement and recruitment
The study was performed in accordance with the ethical standards of 
the overseeing human research ethics committees and local govern-
ance, as per the 1964 Helsinki Declaration. The study was reviewed 
and approved by the Ministry of Health and Social Services (MoHSS) 
in Namibia (17-3-3 2008, 2014 and 2019), with additional local approv-
als from participating community leaders, the University of Preto-
ria Human Research Ethics Committee (HREC 43/2010 and HREC 
280/2017), including US Federal-wide assurance (FWA00002567 and 
IRB00002235 IORG0001762), as well as the South African National 
Blood Service (SANBS) HREC (HREC 2012/11). Participants were 
recruited within the borders of Namibia and South Africa and self-
reported ethno-linguistic population identifiers were recorded. Blood 
samples were taken after receiving written and/or recorded informed 
consent. Isolated DNA was shipped under the Republic of South Africa 
Department of Health Export Permit ( J1/2/4/2), in accordance with the 
National Health Act 2003, to the Garvan Institute of Medical Research 
in Australia. Mitogenome sequencing was performed in accordance 
with site-specific approval granted by St Vincent’s Hospital HREC in 
Australia (SVH 15/227).
Participant population identifiers
Merging with published data for a total of 1,217 L0 mitogenomes, par-
ticipants were broadly classified as KhoeSan, Bantu or Cape multi-
ethnic heritage. Indigenous KhoeSan who inhabit the inland semi-desert 
Kalahari region of Botswana and Namibia include the Kx’a ( Ju|’hoan 
or Hoan, and !Xun or !Xuun), Tuu (or Taa) and Khoe–Kwadi (Naro, 
$Ani, Khwe, Buga, G$ana, G$ui, $Xokhoe, Tshwa and Shua) speakers. 
Indigenous KhoeSan who inhabit the west-coastal region of Namibia 
speak a Khoe–Kwadi or Nama language and include the Nama, Damara, 
Topnaar (ǂAonin) and Hai$om speakers42,43. Novel mitogenomes were 
derived from 15 Kalahari KhoeSan, including Ju|’hoan (n = 9), !Xun (n = 1) 
and Naro (n = 5), and 21 west-coastal KhoeSan, including Nama (n = 7), 
Damara (n = 8) and Topnaar (ǂAonin, n = 6) from Namibia. Speakers 
of southwest Bantu (non-KhoeSan) languages (which do not contain 
click consonants) of Namibia, Botswana and southerly boarders of 
Angola, presenting with KhoeSan-predominant L0 maternal lineages, 
most likely carry a Kalahari or west-coastal KhoeSan mitogenome. As a 
result of refuge provided to the Herero by the Kalahari KhoeSan during 
the early 1900 German South West African genocide44, we speculate in 
this study a probable Kalahari KhoeSan heritage for the three Herero 
mitogenomes.
Although indigenous KhoeSan are arguably absent from the coastal 
regions of South Africa, and while recognising and honouring the north-
west inland (southern Kalahari) ǂKhomani San of South Africa (although 
not recruited within the context of this study), KhoeSan skeletal remains 
spread across the region45. Hunter-gatherer KhoeSan once inhabited a 
broad southwest to east-coastal region at the tip of Africa. These skel-
etal remains predate archaeological evidence supporting the arrival 
of sheep herders who appear to have crossed the Okavango river in 
northern Namibia around 2.2 ka, migrating along the southwest coast 
to the southern Cape12,16,20,45 by around 2 ka. Recently, Cape KhoeSan 
skeletons younger than 2 ka have been genetically linked to east Africa 
and herder migration17. Migrating herders may have acquired indig-
enous KhoeSan maternal contributions. Along the east coast, southward-
migrating Bantu farmers (Southern Bantu; who presumably did not 
speak languages containing click consonants) entered South Africa 
around 1,500 years ago, while a second wave of Bantu migrants (South-
west Bantu) crossed central Africa into Namibia around 800 years ago12. 
Maternal contributions to the South African Southern-Bantu-speaking 
populations (n = 43, this study) may therefore either be of Bantu origin 
(in this case, L0a lineages and therefore non-KhoeSan) or of east-coastal 
KhoeSan-ancestry. The arrival of European colonists and Dutch–East-
Indian slaves to the Cape in the mid-1600s, gave rise to a multi-ethnic 
(European, Asian, KhoeSan and Bantu) Cape population, the ancestors 
of the South African Coloured (n = 90, this study) and Namibian Basters 
(n = 24, this study), who historically speak a Dutch-derived language 
known as Afrikaans13,46. Emerging from a common historical background 
to the Coloured, the Baster population have since the late 1800s distin-
guished themselves as independent from the Coloured, migrating to 
the Baster nation of Rehoboth in Namibia47. Although the vast majority 
of L0 mitogenomes represented in the Baster and Coloured popula-
tions are of Cape KhoeSan heritage (100% and 94.4%, respectively), we 
observe a percentage of non-KhoeSan (Bantu) L0a lineages within the 
Coloured population.
L0 haplogroup pre-screening
Subjects were selected for whole-mitogenome sequencing based on 
pre-screening for specific L0 markers using direct amplicon-specific 
Sanger sequencing. Specifically, a 2,673-bp region (Cambridge Refer-
ence Sequence (rCRS) position 3322–5995) was amplified and initially 
screened for the L0 variant T5442C. L0 samples were further screened to 
delineate L0d (T4232C), L0d1 (G3438A), L0d1b (T3618C), L0d1c (C4197T), 
L0d1’2 (A3756G), L0d2 (A3981G, C205T, A4044G), L0d2a (A5153G), L0d2d 
(G5147A, G5231A), L0d2C (A4038G, T4937C) and L0d3 (G5460A, G5773A) 
lineages. This identified 188 samples carrying a rare L0 haplogroup: 
L0d1b (n = 21), L0d1c (n = 13), L0d2a (n = 30), L0d2b (n = 7), L0d2c (n = 15), 
L0d2d (n = 6), L0d3 (n = 29), L0a1 (n = 6), L0a2 (n = 6), L0f (n = 5) and L0k 
(n = 5); as well as 55 samples that could not be unambiguously assigned 
to a major L0 sub-lineage: L0d1a’c (n = 2), L0a’b’f’k (n = 5), L0a’b (n = 2), 
L0d2 (n = 1) and L0d1 (n = 45, assumed L0d1a) (Supplementary Table 1).
Whole-mitogenome sequencing
Mitogenomes were isolated using two overlapping amplicons as previ-
ously described6,48. Specifically, two primer pairs were used to isolate 
and amplify fragments 12,250–3,005 (7.2 kb) and 2,583–12,337 (9.7 kb) 
of the circular mitogenome. This pair of primers has been demonstrated 
to effectively capture the mitogenome with high specificity while mini-
mizing off-target capture of nuclear copies of mitochondrial-derived 
DNA. Following touchdown long-range amplification with the Platinum 
Taq DNA Polymerase High Fidelity (Invitrogen), the two amplicons were 
purified using AMPure XP beads (Agencourt) and combined in a 7:13 
ratio of short:long fragments. Sequencing was performed on the Ion 
Torrent PGM platform. In brief, 200-bp single-end sequencing libraries 
were prepared using the Ion Xpress Plus Fragment Kit and Ion Xpress 
Barcode Adaptors (ThermoFisher), and 4–16 samples (barcodes) were 
pooled and sequenced on 314v2 Ion Chips. Using the Ion Torrent suite 
v.5.0.2.1, sequencing reads were quality trimmed and aligned to the 
human mitochondrial revised rCRS (accession NC_012920.1). Consensus 
mitogenome sequences were derived by first identifying variants relative 
to rCRS, using samtools (v.1.3.1) mpileup (with parameters -d 10000 -L 
1000 -Q 7 -h 50 -o 10 -e 17 -m 4)49 and bcftools (v.1.3.1) call (with param-
eters -c -M) (http://www.htslib.org/doc/bcftools.html), then converting 
to the FASTA format using the vcfutils.pl vcf2fq program in samtools.
Publicly available data
An exhaustive search for publicly available L0 mitogenomes was per-
formed between 2015 and 2017, identifying 26 studies comprising a total 
of 6,334 mitogenomes. L0 status for all mitogenomes was deduced, either 
directly from the original publication or by downloading the nucleotide 
sequences from NCBI and evaluating their haplogroup using HaploGrep2 
(v.2.1.13)50 based on PhyloTree Build 1751. From this dataset, a subset of 
1,019 L0 mitogenomes was identified and included in this study (Extended 
Data Table 1 and Supplementary Table 2). Publicly available genomes were 
broadly classified as KhoeSan, Bantu (KhoeSan ancestral), or non-KhoeSan 
based on the reported population and/or country of origin.
Whole-mitogenome haplotyping
HaploGrep250 was used to type all 1,217 sequences against PhyloTree 
Build 1751. This resulted in the refinement and reclassification of our 198 
mitogenomes, resulting in L0d1 (n = 81, including 45 L0d1a, 21 L0d1b, 
13 L0d1c and 2 L0d1d), L0d2 (n = 58, including 30 L0d2a, 8 L0d2b, 14 
L0d2c and 6 L0d2d), L0d3 (n = 27), L0a (n = 19), L0f (n = 5), L0k (n = 5) 
and L0g (n = 3) mitogenomes (Supplementary Table 1). This refined, 
and in some cases reclassified, the haplogroups of the 1,019 publicly 
available mitogenomes (Supplementary Table 2).
Phylogenetic inference
Multiple sequence alignment was performed across all 1,217 mitog-
enomes along with 7 Neanderthal genomes (Supplementary Table 10), 
using MUSCLE v.3.8.3152 with parameters -maxiters 3 -diags1. Phylo-
genetic inference was performed using FastTree v.2.1.7 (SSE3)53 using 
the generalized time reversible (-gtr) and discrete gamma model with 
20 rate categories (-gamma). A summary of the inferred phylogenetic 
tree is shown in Extended Data Fig. 1, with the tree rerooted to the 7 
Neanderthal genomes.
Bayesian phylogenetic inferences and divergence times were calcu-
lated using BEAST2 v.2.4.2 with BEAGLE v.2.054. Owing to the compu-
tational burden of this analysis, BEAST was performed on a subset of 
461 mitogenomes, selected to include: (i) only complete mitogenomes 
(27 mitogenomes with only the coding region55,56 were excluded); (ii) all 
198 novel mitogenomes from this study; (iii) all 121 L0 mitogenomes 
from our previous studies, Chan et al.6 (n = 77), Morris et al.15 (StHe, 
defining the new haplogroup L0d2c1c), McCrow et al.48 (n = 37) and 
Schuster et al.57 (n = 6); (iv) all rare haplogroups, namely L0g (n = 9), L0f 
(n = 22), L0d3 (n = 30), L0d1d (n = 3), L0d2d (n = 11) and L0k2 (n = 12); (v) all 
mitogenomes that could not be unambiguously typed by HaploGrep250 
(n = 14; none from this study); and (vi) a random subset of mitogenomes 
for all remaining sub-lineages not already represented.
Multiple sequence alignment of the subset of 461 AMH and 7 Nean-
derthal mitogenomes was converted to NEXUS format using the con-
vert function of seqmagick v.0.6.1 (https://fhcrc.github.io/seqmagick) 
with parameter --alphabet dna-ambiguous. This provided the input 
to BEAST2. Specifically, BEAUTi v.2.4.2 was used to set up the phylo-
genetic model, assuming: (i) the gamma site model with six gamma 
categories and no invariant sites; (ii) the generalized time-reversible 
substitution model; (iii) a strict constant clock model with a normal prior 
with µ = 1.665 × 10−8 and σ = 1.479 × 10−9 based on a previously published 
study58; and (iv) a coalescent constant population. Times were calibrated 
to the seven H. neanderthalensis mitogenomes with tip dates set to their 
reported approximate archaeological dating estimates: Feldhofer 1, 
40 ka59; Vindija, 38 ka59; El Sidron, 39 ka59; Feldhofer 2, 40 ka59; Mezmais-
kaya, 65 ka59; Croatia, 38.31 ka60; Altai, 50 ka61 (Supplementary Table 10). 
No prior was set on the most recent common ancestor of this taxon set, 
and calibration was applied to the leaves instead of the most recent 
common ancestor. Further, a normal prior, N(µ = 200,000, σ = 50,000), 
was set on the coalescent time of the AMH genomes, and a tip date of 
2,330 years before present was set for the StHe genome15.
Five BEAST replicates were performed, each with 100 million Markov 
chain Monte Carlo iterations, sampling every 10,000. Tracer v.1.6 was 
used to evaluate BEAST trace files (Supplementary Table 11), ensuring all 
runs had converged. The five replicates were combined using LogCom-
biner v.2.4.2, discarding 10% of the samples as burn-in for each replicate 
and without resampling states at a lower frequency.
Sampled trees from BEAST were summarized into a single maximum 
clade credibility target tree using TreeAnnotator v.2.4.2 for each of the 
five replicates, discarding the first 10% as burn-in. To summarize across 
replicates, sampled trees from the five replicates were first combined 
using LogCombiner v.2.4.2, again discarding the first 10% as burn-in 
from each replicate, but resampling at a lower frequency of 50,000 
(five replicates of 10,000 samples). The combined, resampled trees 
were then summarized with TreeAnnotator v.2.4.2 as for the individual 
replicate BEAST results.
FigTree v.1.4.2+ (http://tree.bio.ed.ac.uk/software/figtree/) was used 
to visualize all resulting trees.
BSP analysis
BSP analyses were performed to estimate the demographic history 
of each maternal haplogroup. Although maternal haplogroups do 
not necessarily equate to population groups, it has been suggested 
that the signal associated with a haplogroup can still provide insights 
into the demographic processes in the populations who carry the 
haplogroup62,63.
For each haplogroup of interest (for example, L0a, L0d1’2 and L0k), 
a nexus file was derived using SeqMagic v.0.6.1 as described above. 
Article
BSP analyses were performed using BEAST2, using BEAUTi 2 for model 
setup as described in ‘Phylogenetic inference’, with the following key 
differences: (i) the gamma shape of the gamma site model was esti-
mated with an exponential prior with mean = 1.0 and offset = 0.0; 
(ii) the molecular clock was fixed (not estimated) at 1.665 × 10−8 based 
on a previously published study58; and (iii) the phylogenetic tree prior 
was set to coalescent Bayesian skyline, assuming 20 intervals between 
the root of the tree and the present time.
Tracer v.1.6 was used to reconstruct the Bayesian skyline from the 
sampled trees for each analysis, using a stepwise constant variant and the 
lower 95% highest posterior density of the root height as the maximum 
time. Results of this analysis are summarized in Supplementary Table 12.
Geographical history of the palaeo-wetland Makgadigadi
Initiated around 2 million years ago, palaeo-lake Makgadikgadi7 origi-
nally covered an area of around 170,000 km2 at its highest lake stand, 
bounded by a shoreline of around 995 m. A degraded sand ridge (the 
Deception ridge), was associated with the 995-m shore in the southwest 
of the lake. This lake would have covered more than twice the area of 
modern Lake Victoria, and similar to the latter, would have caused a con-
siderable climatic feedback, with locally enhanced rainfall. We previously 
proposed that this was, in turn, responsible for the initiation of the sur-
rounding (now-fossil) drainages, creating a well-watered environment 
and very favourable habitat for mammals, including hominids7. Smaller 
lakes, now represented by residual wetlands, also formed on the upper 
Zambezi river and the modern Kafue Flats on the Kafue river, resulting 
in an archipelago of palaeo-lakes in south-central Africa during the Early 
and Middle Pleistocene epoch.
Palaeo-Makgadikgadi, bounded by the 995-m shoreline, was originally 
sustained by a major drainage line, which included the Chambeshi river 
as headwaters, connected to the upper Zambezi river via the upper Kafue 
river. Severance of the original links between the Chambeshi river and 
upper Kafue river, and the latter and the upper Zambezi river resulted 
in a sequential contraction of the Makgadikgadi to a much smaller water 
body. This is reflected in a series of fossil shorelines, associated with 
breaks in slope, at progressively lower levels (945 m, 936 m and 922 m). 
The Gidikwe ridge was associated with the 945-m shoreline. However, 
contraction of the lake was accompanied by the development of the 
modern Okavango delta. Timing of the contraction of the lake and 
initiation of the Okavango delta is not tightly constrained, but by the 
time that we propose that modern humans emerge within the region, 
around 200 ka, we speculate that the formerly extensive Makgadikgadi 
palaeo-lake had contracted to a much less extensive deltaic–lacustrine 
system. Together with the lakes that developed from the upper Zambezi 
and Kafue rivers to the north and the Okavango delta to the west, the 
region would have been a vast wetland, a favourable habitat for homi-
nid occupation. It is this palaeo-wetland region that we propose as the 
homeland for the founder population of AMHs.
Climate model simulations and palaeo-climate data
To place the coalescence time estimates of the L0 branch into a cli-
matic context and to test the robustness of simulated hydroclimate 
responses in South Africa to orbital-scale conditions, we use the LOVE-
CLIM Earth system model of intermediate complexity28. It is based on 
a 3-layer atmosphere, a 20-level ocean general circulation model, a 
dynamic–thermodynamic sea–ice model and a terrestrial vegetation 
model. A transient simulation that covers several glacial–interglacial 
cycles was conducted using time-dependent boundary conditions. 
The experiment28 (covering the past 784 kyr) uses time-varying bound-
ary conditions for orbital parameters, CO2 and other greenhouse gas 
concentrations obtained from Antarctic ice cores, and an estimate of 
Northern Hemispheric ice-sheet orography and albedo changes (data 
are used in Fig. 3 and Extended Data Fig. 6). The forcings are applied with 
an acceleration factor of five: one coupled model year corresponds to 
five orbital calendar years. Our analysis focuses on the past 250 kyr in 
both simulations. The climate sensitivity of this model to CO2 variations 
was modified to capture the range of reconstructed global mean surface 
temperature changes in response to radiative forcing64. The transient 
LOVECLIM model simulations have previously been validated against 
other palaeo-climate records from around the world28,64,65. Our analysis 
here focuses on the simulated precipitation as well as changes in tree 
and grass fractions in central eastern Africa and western southern Africa 
(data used in Fig. 3 d–f and Extended Data Fig. 6b–d).
As a result of its coarse horizontal atmospheric resolution (5.6°) and 
the use of only parameterized ageostrophic wind components, LOVE-
CLIM has several deficiencies. Of particular note are the lack of realistic 
El Niño–Southern Oscillation variability and the fact that annual mean 
freshwater flux corrections have been applied to mimic the atmospheric 
moisture transport from the Atlantic to the Pacific and to stabilize the 
Atlantic Meridional Overturning Circulation.
There exist only a few long-term hydroclimate datasets from southern 
Africa that cover the past >120 kyr. Here we compare the simulated LOVE-
CLIM precipitation (normalized) in central southern Africa with a south-
ern central African hydroclimate composite, obtained by averaging the 
normalized orbitally tuned rainfall reconstruction from the Pretoria salt 
pan29 and the normalized Fe/K river run-off proxy obtained from marine 
sediment core CD154-1006P31 (Fig. 3b). The composite index emphasizes 
the joint variability in both records. We find that some of the overall fea-
tures in the observations—particularly the fact that rainfall is modulated 
by the precessional cycle of austral summer insolation66 (Fig. 3a)—are 
well-captured by the LOVECLIM model experiment. However, we also find 
some discrepancies in the central part of southern Africa, such as in the 
phase of the precessional signal and the difference in overall wet and dry 
conditions during the homeland period from 200 to 120 ka. The overall 
glacial drying in the central part of southern Africa from 100 to 20 ka is, 
however, captured in both model simulation and palaeo-proxy recon-
structions (Fig. 3b, e). Orbital-scale hydroclimate variations in southern 
Africa are clearly not spatially homogenous (Fig. 3b–f). To gain a better 
understanding of the spatial patterns of hydroclimate variability, we com-
pared the model simulation with a composite index from southwestern 
Africa, obtained by averaging a normalized aridity index reconstructed 
from sediment core MD96-209433 and the normalized δ13C isotope ratio 
data of leaf wax extracted from the South Atlantic sediment core MD08-
316732 (Fig. 3c and Extended Data Fig. 6c, d). The results show a good 
correspondence between model and reconstructions on the western side 
of southern Africa, and in particular reproduce a major drought period 
that peaked around 120 ka and a subsequent increase in rainfall towards 
the last glacial period. This gradual increase in rainfall corresponds to 
an overall increase in lineage splitting of the L0d1’2 haplogroup (Fig. 3f) 
and growth of its population (Fig. 3c). This result further highlights the 
possibility that climate shifts may have played an important part in the 
southwestward migration of L0d1’2 descendants (Fig. 2).
To further test the fidelity of LOVECLIM in reproducing interhemi-
spheric orbital rainfall shifts across Africa, we also compared the simu-
lated vegetation changes with a leaf-wax index from stable hydrogen 
isotope data extracted from a sediment core in the Gulf of Aden30, which 
is indicative of hydroclimate and vegetation changes in the northeast-
ern Horn of Africa (Extended Data Fig. 6b). The comparison shows a 
good qualitative correspondence for the precessional-scale timing of 
rainfall and vegetation maxima and minima as well as of the eccentric-
ity modulated amplitude of these changes, lending further support to 
the credibility of the simulated rainfall patterns across Africa. It should 
be noted that regional patterns of paleo-rainfall changes are in general 
difficult to simulate. In response to Last Glacial Maximum boundary 
conditions, different coupled general circulation models simulate widely 
varying responses in rainfall over Africa28.
Reporting summary
Further information on research design is available in the Nature 
Research Reporting Summary linked to this paper.
Data availability
The consensus sequences for this set of 198 mitogenomes have been 
deposited in the NCBI GenBank with accession numbers MK248274–
MK248471. Requests for materials should in the first instance be 
addressed to V.M.H.
 
38. Orizio, R. Lost White Tribes: the End of Privilege and the Last Colonials in Sri Lanka, 
Jamaica, Brazil, Haiti, Namibia, and Guadeloupe (Free, 2001).
39. Heine, B. & Nurse, D. (eds) African Languages: an Introduction (Cambridge Univ. Press, 
2000).
40. Montinaro, F. et al. Complex ancient genetic structure and cultural transitions in southern 
African populations. Genetics 205, 303–316 (2017).
41. Guthrie, M. The Classification of the Bantu Languages (Oxford Univ. Press, 1948).
42. Honken, H. & Heine, B. The Kx’a family: a new Khoisan genealogy. J. Asian Afr. Stud. 79, 
5–36 (2010).
43. Güldemann, T. & Elderkin, E. D. in Khoisan Languages and Linguistics: Proc. 1st 
International Symposium January 4–8, 2003, Riezlern/Kleinwalsertal (eds Brenzinger, M. & 
König, C.) 15–52 (Rüdiger Köppe, 2010).
44. Stockton, R. The Herero genocide: Germany’s first mass murder. All That’s Interesting 
https://allthatsinteresting.com/herero-genocide (2017).
45. Smith, A. B. Excavations at Kasteelberg and the Origins of the Khoekhoen in the Western 
Cape, South Africa (Archaeopress, 2006).
46. Patterson, N. et al. Genetic structure of a unique admixed population: implications for 
medical research. Hum. Mol. Genet. 19, 411–419 (2010).
47. van der Ross, R. E. Up from Slavery: Slaves at the Cape: their Origins, Treatment and 
Contribution (Ampersand, 2005).
48. McCrow, J. P. et al. Spectrum of mitochondrial genomic variation and associated clinical 
presentation of prostate cancer in South African men. Prostate 76, 349–358 (2016).
49. Li, H. A statistical framework for SNP calling, mutation discovery, association mapping 
and population genetical parameter estimation from sequencing data. Bioinformatics 27, 
2987–2993 (2011).
50. Weissensteiner, H. et al. HaploGrep 2: mitochondrial haplogroup classification in the era 
of high-throughput sequencing. Nucleic Acids Res. 44, W58–W63 (2016).
51. van Oven, M. & Kayser, M. Updated comprehensive phylogenetic tree of global human 
mitochondrial DNA variation. Hum. Mutat. 30, E386–E394 (2009).
52. Edgar, R. C. MUSCLE: multiple sequence alignment with high accuracy and high 
throughput. Nucleic Acids Res. 32, 1792–1797 (2004).
53. Price, M. N., Dehal, P. S. & Arkin, A. P. FastTree 2—approximately maximum-likelihood 
trees for large alignments. PLoS ONE 5, e9490 (2010).
54. Bouckaert, R. et al. BEAST 2: a software platform for Bayesian evolutionary analysis. PLOS 
Comput. Biol. 10, e1003537 (2014).
55. Kivisild, T. et al. The role of selection in the evolution of human mitochondrial genomes. 
Genetics 172, 373–387 (2006).
56. Herrnstadt, C. et al. Reduced-median-network analysis of complete mitochondrial DNA 
coding-region sequences for the major African, Asian, and European haplogroups. Am. J. 
Hum. Genet. 70, 1152–1171 (2002).
57. Schuster, S. C. et al. Complete Khoisan and Bantu genomes from southern Africa. Nature 
463, 943–947 (2010).
58. Soares, P. et al. Correcting for purifying selection: an improved human mitochondrial 
molecular clock. Am. J. Hum. Genet. 84, 740–759 (2009).
59. Briggs, A. W. et al. Targeted retrieval and analysis of five Neandertal mtDNA genomes. 
Science 325, 318–321 (2009).
60. Green, R. E. et al. The Neandertal genome and ancient DNA authenticity. EMBO J. 28, 
2494–2502 (2009).
61. Prüfer, K. et al. The complete genome sequence of a Neanderthal from the Altai 
Mountains. Nature 505, 43–49 (2014).
62. Gandini, F. et al. Mapping human dispersals into the Horn of Africa from Arabian Ice Age 
refugia using mitogenomes. Sci. Rep. 6, 25472 (2016).
63. Soares, P. et al. The expansion of mtDNA haplogroup L3 within and out of Africa. Mol. 
Biol. Evol. 29, 915–927 (2012).
64. Friedrich, T., Timmermann, A., Tigchelaar, M., Elison Timm, O. & Ganopolski, A. Nonlinear 
climate sensitivity and its implications for future greenhouse warming. Sci. Adv. 2, 
e1501923 (2016).
65. Stockhecke, M. et al. Millennial to orbital-scale variations of drought intensity in the 
Eastern Mediterranean. Quat. Sci. Rev. 133, 77–95 (2016).
66. Laskar, J. et al. A long-term numerical solution for the insolation quantities of the Earth. 
Astron. Astrophys. 428, 261–285 (2004).
67. Barbieri, C. et al. Unraveling the complex maternal history of Southern African Khoisan 
populations. Am. J. Phys. Anthropol. 153, 435–448 (2014).
68. Barbieri, C., Butthof, A., Bostoen, K. & Pakendorf, B. Genetic perspectives on the origin of 
clicks in Bantu languages from southwestern Zambia. Eur. J. Hum. Genet. 21, 430–436 
(2013).
69. Barbieri, C. et al. Contrasting maternal and paternal histories in the linguistic context of 
Burkina Faso. Mol. Biol. Evol. 29, 1213–1223 (2012).
70. Barbieri, C. et al. Migration and interaction in a contact zone: mtDNA variation among 
Bantu-speakers in Southern Africa. PLoS ONE 9, e99117 (2014).
71. Batini, C. et al. Insights into the demographic history of African Pygmies from complete 
mitochondrial genomes. Mol. Biol. Evol. 28, 1099–1110 (2011).
72. Eaaswarkhanth, M. et al. Traces of sub-Saharan and Middle Eastern lineages in Indian 
Muslim populations. Eur. J. Hum. Genet. 18, 354–363 (2010).
73. Gonder, M. K., Mortensen, H. M., Reed, F. A., de Sousa, A. & Tishkoff, S. A. Whole-mtDNA 
genome sequence analysis of ancient African lineages. Mol. Biol. Evol. 24, 757–768 
(2007).
74. Horai, S., Hayasaka, K., Kondo, R., Tsugane, K. & Takahata, N. Recent African origin of 
modern humans revealed by complete sequences of hominoid mitochondrial DNAs. 
Proc. Natl Acad. Sci. USA 92, 532–536 (1995).
75. Ingman, M., Kaessmann, H., Pääbo, S. & Gyllensten, U. Mitochondrial genome variation 
and the origin of modern humans. Nature 408, 708–713 (2000).
76. Just, R. S., Diegoli, T. M., Saunier, J. L., Irwin, J. A. & Parsons, T. J. Complete mitochondrial 
genome sequences for 265 African American and U.S. “Hispanic” individuals. Forensic 
Sci. Int. Genet. 2, e45–e48 (2008).
77. Kujanová, M., Pereira, L., Fernandes, V., Pereira, J. B. & Cerný, V. Near eastern Neolithic 
genetic input in a small oasis of the Egyptian Western Desert. Am. J. Phys. Anthropol. 140, 
336–346 (2009).
78. Maca-Meyer, N., González, A. M., Larruga, J. M., Flores, C. & Cabrera, V. M. Major genomic 
mitochondrial lineages delineate early human expansions. BMC Genet. 2, 13 (2001).
79. Macaulay, V. et al. Single, rapid coastal settlement of Asia revealed by analysis of 
complete mitochondrial genomes. Science 308, 1034–1036 (2005).
80. Margaryan, A. et al. Eight millennia of matrilineal genetic continuity in the South 
Caucasus. Curr. Biol. 27, 2023–2028 (2017).
81. Olivieri, A. et al. Mitogenome diversity in Sardinians: a genetic window onto an island’s 
past. Mol. Biol. Evol. 34, 1230–1239 (2017).
82. van der Walt, E. M. et al. Characterization of mtDNA variation in a cohort of South African 
paediatric patients with mitochondrial disease. Eur. J. Hum. Genet. 20, 650–656 (2012).
83. Vyas, D. N. et al. Bayesian analyses of Yemeni mitochondrial genomes suggest multiple 
migration events with Africa and Western Eurasia. Am. J. Phys. Anthropol. 159, 382–393 
(2016).
Acknowledgements We thank all of the study participants, as well as the many people who 
provided assistance during participant recruitment and recording, or provided critical 
historical, cultural and linguistic insights including; C. P. Bennett (https://evolvingpicture.
com/), R. Wilkinson, J. Sinvula, H. Money, the late C. F. Heyns, R. H. Glashoff, D. de Swart, P. 
Fernandez, P. A. Venter, S. C. Schuster, M. P. Marx, the late S. M. Kooitjie (39th leader of the 
ǂAonin clan and chairperson of the Nama Traditional Leaders Association), A. A. Collins, B. 
Kaesje, J. Kayimbi, H. Mische, F. Naque, D. Naque, H. Oosthuizen, E. Oosthuizen, A. 
Oosthuysen, E. Oosthuysen, D. Roux, C. Swau and T. Tsebe. We acknowledge the late M. 
McFarlane, who identified Deception ridge and its importance in the evolution of the 
Makgadikgadi palaeo-lake. This work was supported by an Australian Research Council 
Discovery Project grant awarded to V.M.H. (DP170103071) and sampling contributed by the 
Cancer Association of South Africa to M.S.R.B. and V.M.H. A.T. and S.-S.L. received funding 
from the Institute for Basic Science (IBS) under IBS-R028-D1. V.M.H. is supported by the 
University of Sydney Foundation in a Petre Foundation chair position. Computational resources 
were provided by the Australian Government through the National Computational 
Infrastructure, the Sydney Informatics Research Hub at the University of Sydney (Artemis HPC) 
and by the Garvan Institute of Medical Research Data Intensive Computer Engineering team.
Author contributions V.M.H. designed the study. M.S.R.B., H.E.A.F. and V.M.H. obtained and 
maintain study approvals and permits, as well as community leadership support. M.S.R.B., 
D.C.P. and V.M.H. performed recruitments, consenting, sampling and processing. R.J.L., 
A.M.F.K. and D.C.P. performed pre-screening and mitogenome data generation. E.K.F.C. 
performed the bioinformatics and phylogenetic analyses. A.E.M. performed geographical 
interpretation. S.-S.L. and A.T. performed climatological model analyses and interpretation, 
with additional local climatology interpretation provided by H.R. V.M.H. led the interpretation 
of the multiple-discipline analyses, with contributions from all of the authors. E.K.F.C., B.F.B., 
S.-S.L., A.T. and V.M.H. generated and interpreted the figures. V.M.H., E.K.F.C. and A.T. wrote the 
manuscript with contributions from all of the authors.
Competing interests The authors declare no competing interests.
Additional information
Supplementary information is available for this paper at https://doi.org/10.1038/s41586-019-
1714-1.
Correspondence and requests for materials should be addressed to A.T. or V.M.H.
Peer review information Nature thanks Victor Brovkin, Rebecca Cann and the other, 
anonymous, reviewer(s) for their contribution to the peer review of this work.
Reprints and permissions information is available at http://www.nature.com/reprints.
Article
Extended Data Fig. 1 | Phylogenetic tree of all 1,217 L0 mitogenomes. 
Phylogeny was inferred using FastTree v.2.1.746, displayed using FigTree. Tips 
belonging to the same haplogroup are collapsed and coloured as in Fig. 2a. Local 
support values for each node are indicated and branch lengths are proportional 
to the number of substitutions per site. The tree is rooted to the seven 
Neanderthal mitogenomes as indicated.
Extended Data Fig. 2 | Detailed phylogenetic branching of L0k, L0d3, L0f 
and L0g. a–d, Expanded sections of the phylogenetic tree depicted in Fig. 2a are 
shown, including 34 (out of a total of 113) L0k (a), all 40 L0d3 (b), all 27 L0f (c) and 
all 9 L0g (d) mitogenomes. Each mitogenome is represented as a tip and 
coloured based on their broad ethno-linguistic classification, if known. KhoeSan 
is shown in orange, non-KhoeSan in grey and Cape multi-ethnic (KhoeSan 
ancestral) in green. Publicly available mitogenomes for which we cannot be 
certain of their broad population identifier are labelled in black font. Proposed 
new sub-lineages for L0d3, L0f and L0g1 are indicated by red-coloured node 
labels and are further described in Supplementary Tables 7–9.
Article
Extended Data Fig. 3 | Detailed phylogenetic branching of L0d2.  
a, c, d, Expanded branches of the phylogenetic tree depicted in Fig. 2a are 
shown, including 51 (out of a total of 118) L0d2a (a), 25 (out of 53) L0d2c (c) and all 
11 L0d2d (d) mitogenomes. b, For L0d2b, an additional BEAST analysis was 
performed using an alternate subset of 441 mitogenomes that included all  
43 L0d2b samples, as opposed to the n = 461 subset (Fig. 2a) that included  
only 13 L0d2b. The same model parameters were used for both data subsets. In 
all panels, each mitogenome is represented as a tip and coloured based on their 
broad ethno-linguistic classification, as in Extended Data Fig. 2. The previously 
defined L0d2c1c haplogroup, containing the coastal KhoeSan StHe skeleton6 
and other newly proposed sub-lineages are indicated by red node labels 
(Supplementary Tables 4–6).
Extended Data Fig. 4 | Detailed phylogenetic branching of L0d1.  
a–c, Expanded branches of the phylogenetic tree depicted in Fig. 2a are shown, 
including 54 (out of a total of 91) L0d1a (a), 45 (out of 174) L0d1b (b) and 33 (out of 
184) L0d1c (c) mitogenomes. Each mitogenome is represented as tips and 
coloured based on their broad ethno-linguistic classification as in Extended 
Data Fig. 2.
Article
Extended Data Fig. 5 | Detailed phylogenetic branching of L0a. The L0a 
branch of the phylogenetic tree displayed in Fig. 2a is shown, which includes a 
subset of 114 (out of a total of 294) L0a mitogenomes. Each mitogenome is 
represented as tips and coloured based on their broad ethno-linguistic 
classification as in Extended Data Fig. 2.
Extended Data Fig. 6 | Comparison of the palaeo-data and palaeo-model.  
a, Locations of key sites that are used for the comparison of the palaeo-model 
and palaeo-data in this study are highlighted in red. The map was generated in 
Paraview v.5.6 (https://www.paraview.org/). b, Simulated tree fraction (%) at 
Horn of Africa (land grid points nearest to RC09-166) (grey, dark-blue bars) and 
stable hydrogen isotopic composition of leaf wax, corrected for ice volume 
contributions from the Gulf of Aden marine sediment core RC09-16630 (orange), 
indicating changes in hydroclimate. c, Relative precipitation changes (%) 
simulated by LOVECLIM transient model (all forcings) for 11° E, 19° S (grey, dark-
blue bars) and grain-size aridity index reconstructed from sediment core MD96-
209432 (orange). d, Grass fraction changes simulated by LOVECLIM transient 
model (all forcings) at 11° E, 14–17° S (grey, dark-blue bars) and reconstructed 
δ13C changes of n-alkanes (orange) (South Atlantic sediment core MD08-3167) 
indicative of abundance of C3 and C4 plants in the Namibian desert and further 
inland33.
Article
Extended Data Table 1 | L0 mitogenomes included in this 
study
Numbers of mitogenomes taken from previously published67–83 studies. 
*Previously published data by our group with verified population metadata. 
†Mitochondrial DNA sequences of the coding-region only. 
‡Sequence has non-canonical start position corresponding to position 577 of rCRS. 
§Coriell cell lines.
Extended Data Table 2 | KhoeSan population identifiers used in this study