Color and genomic ancestry in Brazilians.

Abstract

This work was undertaken to ascertain to what degree the physical appearance of a Brazilian individual was predictive of genomic African ancestry. Using a panel of 10 population-specific alleles, we assigned to each person an African ancestry index (AAI). The procedure was able to tell apart, with no overlaps, 20 males from northern Portugal from 20 males from São Tomé Island on the west coast of Africa. We also tested 10 Brazilian Amerindians and observed that their AAI values fell in the same range as the Europeans. Finally, we studied two different Brazilian population samples. The first consisted of 173 individuals from a rural Southeastern community, clinically classified according to their Color (white, black, or intermediate) with a multivariate evaluation based on skin pigmentation in the medial part of the arm, hair color and texture, and the shape of the nose and lips. In contrast to the clear-cut results with the African and European samples, our results showed large variances and extensive overlaps among the three Color categories. We next embarked on a study of 200 unrelated Brazilian white males who originated from cosmopolitan centers of the four major geographic regions of the country. The results showed AAI values intermediate between Europeans and Africans, even in southern Brazil, a region predominantly peopled by European immigrants. Our data suggest that in Brazil, at an individual level, color, as determined by physical evaluation, is a poor predictor of genomic African ancestry, estimated by molecular markers.

Full text

Page 1

Color and genomic ancestry in Brazilians
Flavia C. Parra*, Roberto C. Amado†, Jose ´ R. Lambertucci‡, Jorge Rocha§, Carlos M. Antunes†, and Se ´rgio D. J. Pena*¶
Departamentos de *Bioquı ´mica e Imunologia and†Parasitologia, and‡Faculdade de Medicina, Universidade Federal de Minas Gerais, 31270-901 Belo
Horizonte, Brazil; and§Instituto de Patologia e Imunologia Molecular da Universidade do Porto, 4200 Porto, Portugal
Communicated by Francisco Mauro Salzano, Federal University of Rio Grande do Sul, Porto Alegre, Brazil, October 31, 2002 (received for review
August 21, 2002)
Thisworkwasundertakentoascertaintowhatdegreethephysical
appearance of a Brazilian individual was predictive of genomic
Africanancestry.Usingapanelof10population-specificalleles,we
assigned to each person an African ancestry index (AAI). The
procedure was able to tell apart, with no overlaps, 20 males from
northern Portugal from 20 males from Sa ˜o Tome ´ Island on the west
coast of Africa. We also tested 10 Brazilian Amerindians and
observed that their AAI values fell in the same range as the
Europeans. Finally, we studied two different Brazilian population
samples. The first consisted of 173 individuals from a rural South-
eastern community, clinically classified according to their Color
(white, black, or intermediate) with a multivariate evaluation
basedonskinpigmentationinthemedialpartofthearm,haircolor
and texture, and the shape of the nose and lips. In contrast to the
clear-cutresultswiththeAfricanandEuropeansamples,ourresults
showed large variances and extensive overlaps among the three
Color categories. We next embarked on a study of 200 unrelated
Brazilianwhitemaleswhooriginatedfromcosmopolitancentersof
the four major geographic regions of the country. The results
showed AAI values intermediate between Europeans and Africans,
even in southern Brazil, a region predominantly peopled by Euro-
pean immigrants. Our data suggest that in Brazil, at an individual
level, color, as determined by physical evaluation, is a poor pre-
dictor of genomic African ancestry, estimated by molecular markers.
You see leaders today, all over the world, doing it again!
Black, white, yellow, brown, people of every color
slaughtering people of every color! Because Satan is
always the same.
S. L. Carter (1)
T
not exist (2, 3). Yet, races do exist as social constructs that are
mutable over time and across social contexts and are sustained
by a racial ideology (4). The physical traits of an individual,
especially skin pigmentation, hair color, hair texture, and the
shape of the lips and nose, are constantly used for racial
categorization and thus play an extremely influential role in
human social relationships. However, these oligogenic traits are
believed to have emerged as adaptations to geographical selec-
tive factors such as solar radiation and heat (5, 6) and their
utilization as markers of ‘‘race’’ is erroneous. Still, regardless of
its biological meaning, skin color matters a great deal. In a 2001
essay, then President William J. Clinton initially agreed with
W. E. B. Dubois in that ‘‘the problem of the 20th century is the
problem of the color line’’ (7). He then went on to condemn
racial profiling and to hope that in the 21st century color
differences could be transformed from a problem into a promise
for America (7). Of course, the social importance of skin color
is not by any means an exclusively American phenomenon; racial
discrimination and conflict occur worldwide.
In Brazil, not withstanding relatively large levels of genetic
admixture and a myth of ‘‘racial democracy,’’ there exists a
widespread social prejudice that seems to be particularly con-
nected to the physical appearance of the individual (8). Color (in
Portuguese, cor) denotes the Brazilian equivalent of the English
term race (rac ¸a) and is based on a complex phenotypic evalu-
here is wide agreement among anthropologists and human
geneticiststhat,fromabiologicalstandpoint,humanracesdo
ation that takes into account, besides skin pigmentation, hair
type, nose shape, and lip shape (4, 9). The reason the word Color
(capitalized to call attention to this particular meaning) is
preferred to race in Brazil is probably because it captures the
continuous aspects of phenotypes (4). In contrast with the
situation in the United States, there appears to be no racial
descent rule operational in Brazil and it is possible for two
siblings differing in Color to belong to completely diverse racial
categories (8). The Brazilian emphasis on physical appearance
rather than ancestry is demonstrated by the fact that in a large
survey when asked about their origins (the question admitted
multiple responses) ?10% of Brazilian black individuals gave
Africa as one of their answers (10).
WewantedtoascertaintowhatdegreetheColorofaBrazilian
individual was predictive of the degree of genomic African
ancestry. To evaluate the latter we typed each person with a
panel of 10 population-specific alleles (PSAs), i.e., alleles with
wide frequency differentials (?48%) between Africa and Eu-
rope, developed by Parra et al. (11). With these data we could
calculate a personal likelihood-ratio estimator of African ances-
try (12) for individuals belonging to different Brazilian Color
categories. The results demonstrated that at the individual level
there was significant dissociation of Color, determined by phys-
ical evaluation, and African ancestry, estimated by molecular
markers.
Materials and Methods
Populations Studied. We estimated the extent of African ancestry
in two Brazilian population samples. The first was collected as
part of an epidemiological study of schistosomiasis and included
173 individuals from the rural community of Queixadinha (17.12
south; 41.42 west) in the Vale do Jequitinhonha region of the
state of Minas Gerais in Brazil (Fig. 1). They were all inhabitants
of the same small village, which consisted of virtually a one-class
community. Each individual was examined clinically and Color
was established by using a multivariate evaluation based on skin
color in the medial part of the arm, hair color and texture, and
the shape of the nose and lips (Table 1). The format of the nose
and lips were determined by direct comparison with depicted
models from a human morphologic atlas (13). Two health
workersevaluatedeachindividual:thosewithwhiteskin,straight
orwavyhair,prominentorupturnednosetip,thinormedianlips,
independently of hair or eye pigmentation, were called white,
whereas those with black skin, black curly hair, black eyes, and
thick lips were classified as black. All other patients, including
those for whom there was any doubt or disagreement between
the two examiners, were classed as intermediate. Twenty-nine
individuals were classified as white, 30 as black, and 114 as
intermediate. Informed consent was obtained from all individ-
uals studied.
We also studied 200 unrelated Brazilian males self-defined as
whites, randomly drawn from paternity casework. This is the
same sample described in our previous article (14). The indi-
viduals were from the following Brazilian regions (Fig. 1): 49
Abbreviations: AAI, African ancestry index; PSA, population-specific allele.
¶To whom correspondence should be addressed. E-mail: spena@dcc.ufmg.br.
www.pnas.org?cgi?doi?10.1073?pnas.0126614100PNAS ?
January 7, 2003 ?
vol. 100 ?
no. 1 ?
177–182
GENETICS

Page 2

from the north (states of Amazonas, Acre, Rondo ˆnia, and Para ´),
49 from the northeast (state of Pernambuco), 50 from the
southeast (state of Minas Gerais), and 52 from the south (states
of Rio Grande do Sul, Santa Catarina, and Parana ´). Discrep-
ancies in numbers were caused by geographical reclassification
of two samples.
Additionally, we typed three samples of populations con-
sidered representative of the major founding groups of Brazil.
The first, representing Europe, encompassed 20 individuals
randomly drawn from the previously described (14) sample of
93 unrelated Portuguese men from the Porto District in
northern Portugal (41.11 north; 8.36 west). The second, rep-
resenting Africa, comprised 20 individuals from the town of
Santana in Sa ˜o Tome ´ Island on the west coast of Africa (1.00
north; 7.00 east). Because this island was a Portuguese en-
trepo ˆt for assembly of slaves before shipment to Brazil, its
population had geographical origin in several regions of Africa
(15). The population-based estimate of European genetic
contribution to the Santana population was only 0.074 ? 0.015
(15). Finally, we studied a small Amerindian sample consisting
of 10 individuals from three Amazonian tribes (Karitiana,
Suruı ´, and Ticuna), all kindly provided by Judith Kidd (Yale
University, New Haven, CT).
DNA Analysis. Blood was drawn, DNA was extracted, and each
patient was then independently typed with the 10 PSAs defined
by Parra et al. (11) as follows: APO, AT3–I?D, GC*1S, GC*1F,
FY–null, LPL, OCA 2, RB 2300, Sb19.3, and ICAM1. With the
exceptionoftwoAluinsertionspolymorphism(APOandSb19.3)
and a polymorphic insertion of a 68-bp fragment at locus
AT3-I?D, all other markers are single nucleotide polymorphisms
and (except for ICAM1) were studied by digestion with the
appropriate restriction enzyme after PCR, as described by Parra
et al. (11). Typing at the ICAM1 polymorphism was achieved by
single-strand conformational polymorphism (16). After addition
of 2.0 vol of a denaturing solution (100% formamide with 5
mg?ml dextran blue 2000), PCR products were denatured at
95°C for 5 min, and the mixture was applied in a native 10%
polyacrylamide gel (18 ? 17 ? 0.1 cm), 0.5? Tris-borate?EDTA
buffer (TBE) with 10% glycerol, subjected to electrophoresis in
1.0? TBE at 15 W for 5 h at room temperature, and then
silver-stained according to Santos et al. (17). As standard, we
used an African DNA sample homozygous for the rare allele.
Based on these results we assigned to each subject an indi-
vidual African ancestry index (AAI) that was calculated as the
logarithm of the ratio of the likelihood of a given multilocus
genotype occurring in the African population to the likelihood
of it occurring in the European population. Thus, the AAI
represents a personal geographical ancestry estimate (12). For
the likelihood ratio calculations we used the African and Euro-
pean allele frequencies given by Parra et al. (11).
Data Analysis. Allele frequencies were estimated by the gene
counting method. Admixture proportions of each population
sample were estimated with two methods. The first used the
program ADMIX, which is based on a weighted least-square
method (18) and was kindly made available by Jeffrey C. Long,
University of Michigan, Ann Arbor. We also obtained point
estimates of admixtures by using the graphical method (also
based on difference minimization) described by Collins-
Schramm et al. (19).
For comparison of two samples we used the Mann–Whitney U
test and applied Bonferroni’s correction for multiple compari-
sons (20). For correlations we used Spearman’s rank correlation
test (20).
Results
Europeans, Africans, and Amerindians. We examined the discrimi-
nation power of the 10-allele PSA set by initially studying a
sample of 20 males from northern Portugal and 20 males from
Sa ˜o Tome ´ Island in the Gulf of Guinea, on the west coast of
Africa. These population sources were chosen because they are
geographically related to the European and African population
groups that participated in the peopling of Brazil. The box plot
obtained from these data is shown in Fig. 2. There is no overlap
between the two groups and ?21 logs separating the two
medians (9.71 and ?11.73, respectively). As expected, the Man-
n–Whitney test was highly significant (z ? ?5.4, P ? 0.0001). A
complete individual discrimination between the European and
African genomes was obtained with an existence of 7.6 logs
between the lower African value (AAI ? 2.86) and the highest
European score (AAI ? ?4.86). It was thus clear that the
10-allele set of Parra et al. (11) was highly efficient and provided
reliable individual discrimination between European and Afri-
can genomes.
Brazilians constitute a trihybrid population with European,
African, and Amerindian roots. Thus, we had to ascertain where
the Amerindians would be positioned in reference to the AAI
scale. For that, we tested DNA samples from 10 individuals from
three Amazonian tribes (Karitiana, Suruı ´, and Ticuna) and
observed that they fell in the same range as the Europeans, with
AAIs varying from ?2.77 to ?12.57 and with a median of
Fig. 1.
north region (states of Amazonas, Acre, Rondo ˆnia, and Para ´), northeast
region (Pernambuco), southeast region (Minas Gerais), and south region
(Parana ´, Santa Catarina, and Rio Grande do Sul). In the state of Minas Gerais
(southeast) the approximate location of Queixadinha is indicated (V).
Geographic localization of Brazilian localities selected for study:
Table 1. Morphological trait evaluation for establishment
of color
Skin pigmentation in the interior part of the forearm
White Brown
Hair type
StraightWavy
Hair color
Blond Brown
Eye pigmentation
Blue Green
Nose conformation
Prominent or upturned tip
Lip conformation
Thin Median
Black
Curly
BlackRed
Brown Black
Depressed tipFlat
Thick
178 ?
www.pnas.org?cgi?doi?10.1073?pnas.0126614100Parra et al.

Page 3

?10.18 (see Fig. 2). This finding is not unexpected, because
several of the population-specific alleles in Parra’s set had
diverging frequencies in Europe and Africa because of specific
selective factors operating in the African environment (11).
Amerindians, who are well known to have a distant Asian
ancestry, did not share such environments and would thus be
expected to have frequencies similar to Europeans.
Brazilians. Our first Brazilian sample was composed of 173
individuals from a Southeastern rural community, clinically
classifiedaccordingtotheirColor(white,black,orintermediate)
with a multivariate evaluation based on skin pigmentation in the
medial part of the arm, hair color and texture, and the shape of
the nose and lips. When we compared the AAI values for these
individuals, we observed that the groups had much wider ranges
than those of Europeans and Africans (Fig. 3) and that there was
very significant overlap between them. For instance, one black
individual had the fourth lowest (least African) AAI score
(?10.48) whereas another individual, classified as intermediate,
had the highest (most African) observed AAI (11.31). However,
the comparison of whites versus blacks with the Mann–Whitney
U test still showed a modest significant value (z ? 2.62; P ? 0.01).
On the other hand, comparisons of whites versus the Portuguese
and blacks versus Sa ˜o Tome ´ islanders yielded extremely high
significance (respectively, z ? 5.08, P ? 0.0001 and z ? ?5.24,
P ? 0.0001), even after applying the Bonferroni correction for
multiple comparisons (20). Thus, the differences in AAI values
of the group of Brazilian blacks compared with Brazilian whites
are very discrete and several orders of magnitude smaller than
the differences observed between Africans and Europeans. We
used the software STRUCTURE (21) to calculate the inferred
proportion of African ancestry of each individual in our sample.
There was a highly significant correlation between the two
admixture estimates, thus further validating our AAI.
Of course, these results could not be naively extrapolated to
all of Brazil. First, we had examined only a population of the
state of Minas Gerais, where European-African interbreeding
occurred very intensely (22). Moreover, our samples were from
a poor rural zone of the state, with negligible social stratification.
We thus undertook the study of a larger cosmopolitan-based
sample with widespread geographical origins. For that, we
studied200unrelatedwhiteBrazilianmales,originatedprimarily
from large cities in the four main geographical regions of the
country (north, northeast, southeast, and south).
The distribution of the AAI values for Brazil as a whole and
for the four regions largely reproduced the previous results,
albeit with some interesting regional differences (Fig. 4). The
samplefromthesoutheast,takenfromthestateofMinasGerais,
did not differ significantly from the rural sample of white
individuals from the same state (z ? ?1.26, P ? 0.104). This
region also had the highest AAI median (?6.41). In contrast, the
sample from south Brazil, the region with the highest levels of
recent (19th and 20th centuries) European immigration, had a
median AAI value of ?9.11, 3 orders of magnitude smaller.
However, when compared with the European (Portuguese)
values, the southern sample still showed a significantly higher
AAI values (z ? 3.04, P ? 0.002).
Admixture Estimation. From our results with the PSAs described
by Parra et al. (11) we could calculate allele frequencies and use
them to estimate the proportion of African admixture in the
several samples studied (Table 2) by using two methods (16, 17).
Although the standard errors of the estimates obtained with
Long’s method (18) are not small (Table 2) there is generally
excellent agreement between the two estimates.
Discussion
Brazilians form one of the most heterogeneous populations in
the world, which is the result of five centuries of interethnic
crosses of peoples from three continents: the European colo-
nizers, mainly represented by the Portuguese, the African slaves,
andtheautochthonousAmerindians.Wehavebeenstudyingthis
diversity systematically from a phylogeographical perspective. In
our previous studies we used DNA lineage markers to ascertain
Fig. 2.
of the northern Portuguese population (n ? 20), Africans from Sa ˜o Tome ´
Island (n ? 20), and Amerindians (n ? 10). Each group is represented as a box
whose top and bottom are drawn at the lower and upper quartiles, with a
small square at the median. Thus, the box contains the middle half of the
scores in the distribution. Vertical lines outside the box extend to the largest
and the smallest observations within 1.5 interquartile ranges from the
box (20).
Box plot graph showing the distribution of values of AAI in samples
Fig. 3.
of the AAIs from the three rural samples grouped according to their pheno-
type(30black,114intermediate,and29whiteindividuals)aswellasfromtwo
of the Brazilian parental populations (20 Africans from Sa ˜o Tome ´ and 20
EuropeansfromNorthernPortugal).EachsymbolindicatestheAAIvaluefrom
one individual.
Slot plot graphic [GraphPad (San Diego) PRISM software, version 2.01]
Parra et al. PNAS ?
January 7, 2003 ?
vol. 100 ?
no. 1 ?
179
GENETICS

Page 4

the continental ancestry of Brazilian white individuals. Our data
demonstrated that the vast majority of Y chromosomes in white
Brazilian males, regardless of their regional source, were of
European origin, with a very low frequency of sub-Saharan
African Y chromosomes and a complete absence of Amerindian
contributions (14). On the other hand, by analyzing mtDNA, we
found the surprisingly high amount of 33% Amerindian and 28%
African contribution to the total mtDNA pool of white Brazil-
ians (23). Together, our results configured a picture of strong
directional mating in Brazil involving European males and
Amerindian and African females.
Because Brazil has a large territory and because different
population groups moved into diverse parts of the country, there
is considerable phylogeographical heterogeneity. When we an-
alyzed mtDNA in the four main regions of Brazil we found that
there was broad agreement with what could be expected from
historical data: a predominance of Amerindian matrilineages in
the Amazon (north) region, a preponderance of African lineages
in the northeast, an equilibrium in the southeast, and European
dominance in the south (Table 2). There was relatively good
concordance of the mitochondrial data with the estimates of
African ancestry from the present study in Brazilian whites from
the four main regions of Brazil. The most pronounced levels of
admixture were observed in the southeast and the northeast, the
former being slightly higher, although the difference is obviously
not statistically significant. This finding is compatible with
historical evidence that Minas Gerais (southeast) society was
more admixed than the coastal sugar zone (northeast) (24), but
contrasts with the higher proportion of African mtDNA lineages
observed in the northeast than in the southeast (Table 2). Such
discrepancy is probably related with the demographical pecu-
liarities of the two regions. On the other hand, the south and the
north have lower African admixture although they differ signif-
icantly in the relative proportions of European and Amerindian,
i.e., non-African, mtDNA lineages. This observation highlights
a methodological shortcoming in that the set of PSAs designed
by Parra et al. (11) cannot (and was not meant to) discriminate
between Europeans and Amerindians. Considering the trihybrid
nature of the Brazilian population, it would be desirable to have
specific markers that would allow the specific dissection of the
Amerindian contribution. Although some authors (e.g., ref. 19)
have identified markers that might be useful for this purpose,
there are serious theoretical difficulties. As pointed out by
Cavalli-Sforza et al. (25), the enormous amounts of genetic drift
existent in many South American native groups has generated
exceptional gene-frequency variation (reviewed in ref. 26) that
severely weakens or renders invalid admixture inferences. Even
if we averaged gene frequencies over extant populations, we
could have no idea how these frequencies were in the past, when
the admixture took place.
There was also excellent agreement between the two admix-
ture estimates in the Queixadinha population for the three
different Color categories. It is interesting to note that the group
of individuals classified as blacks had a very high proportion of
non-African ancestry (48%). Also, the intermediate group, with
45% African ancestry, was closer to the black group than to the
white one. For the whole Queixadinha population, regardless of
skin color, the admixture estimate was 0.44 ? 0.07.
Several attempts to estimate admixture in the Brazilian pop-
ulation were performed in the past, most of them using auto-
somal polymorphisms, especially blood groups and electro-
phoresis protein markers (reviewed in ref. 27). In different
geographical regions, estimates of the African contribution to
Fig. 4.
from Brazil (n ? 200) and from the main regions of the country (NE ?
northeast, N ? north, SE ? southeast, and S ? south). For comparison we
included the data obtained from northern Portuguese and from the rural
sample of white Brazilians already shown in Fig. 3. Each group is represented
as a box whose top and bottom are drawn at the lower and upper quartiles,
with a small square at the median. Thus, the box contains the middle half of
the scores in the distribution. Vertical lines outside the box extend to the
largest and the smallest observations within 1.5 interquartile ranges from the
box (20).
Box plot graph showing the distribution of values of AAI in samples
Table 2. Values of AAI, admixture estimates obtained by two different methods from the frequencies of 10 PSAs and the proportion
of African, European, and Amerindian mtDNA lineages in two samples of the Brazilian population
Sample
AAI
Median
Admixture estimatemtDNA ancestry*
Method 1†
Method 2‡
AfricanEuropean Amerindian
Rural population
(Queixadinha)
Whites
Intermediate
Blacks
North
Northeast
Southeast
South
?4.55
?2.86
?1.25
?7.24
?6.56
?6.41
?9.03
0.31 ? 0.09
0.45 ? 0.07
0.51 ? 0.07
0.23 ? 0.06
0.29 ? 0.09
0.32 ? 0.10
0.13 ? 0.08
0.32
0.48
0.52
0.24
0.30
0.32
0.17
NS
NS
NS
0.15
0.44
0.34
0.12
NS
NS
NS
0.31
0.34
0.31
0.66
NS
NS
NS
0.54
0.22
0.33
0.22
Regions of Brazil
NS, not studied.
*Alves-Silva et al. (23).
†Estimate ? SE (18).
‡Collins-Schramm et al. (19).
180 ?
www.pnas.org?cgi?doi?10.1073?pnas.0126614100Parra et al.

Page 5

the urban population varied from 4% to 34% and those for
Amerindian involvement ranged from 0% to 27% (27). For
instance, in the northeast, Salzano (28) calculated for the
population as a whole, 51% European, 36% African, and 13%
Amerindian ancestries whereas in the north, Santos and Guer-
reiro (29) obtained 47% European, 12% African, and 41%
Amerindian descent, and in the southernmost state of Rio
Grande do Sul, Dornelles et al. (30) calculated 82% European,
7% African, and 11% Amerindian ancestries. Krieger et al. (31)
studied a population of Brazilian northeastern origin living in
Sa ˜o Paulo with blood groups and electrophoretic markers and
showed that whites presented 18% of African and 12% of
Amerindian genetic contribution and that blacks presented 28%
of European and 5% of Amerindian genetic contribution (31).
Of course, all of these Amerindian admixture estimates are
subject to the caveat mentioned in the previous paragraph. At
any rate, compared with these previous studies, our estimates
showed higher levels of bidirectional admixture between Afri-
cans and non-Africans.
The focus of past admixture studies has been the population as
a whole and not the individual (28). Because physical appearance
is so often used for racial categorization and prejudice, we wanted
to ascertain to what degree it was correlated with genomic African
ancestry,whichcouldbeaccomplishedonlyatapersonallevel.The
first sample studied was a group of 173 individuals from the
Queixadinha rural community, in the state of Minas Gerais (south-
east). Each individual was examined clinically and classified as
white, black, or intermediate, on the basis of a multivariate evalu-
ationthatincludedskinpigmentationinthemedialpartofthearm,
hair color and texture, and the shape of the nose and lips. This
evaluationwas,ofcourse,subjective.Neverthelesswetriedtomake
it more reliable by introducing the necessity of concordance by two
independentexaminersforclassificationinthecategorieswhiteand
black. Others have used more quantitative parameters such as skin
reflectance spectrophotometry (e.g., ref. 32), but measuring only
skinpigmentationfailstotakeintoconsiderationtheotherqualities
thatcontributetothecomplexColortrait(4,9).Oneofthepeculiar
features of social relations in Brazil is the large number of Color
categories that are subjectively distinguished in a community.
Harris and Kotak (8) elicited 40 Color terms from a sample of 100
people in a small fishing village in northeastern Brazil. Of course,
mostofthesetermsdenotedColorsinbetweenwhiteandblack,and
in our study we lumped them in the intermediate category, thus
considerably simplifying the classification procedure.
The second component of the study was the estimation of the
personal genomic origin, which was based on a battery of 10
PSAs described by Parra et al. (11) and was summarized in the
individual AAI. It might be considered risky to try to establish
genomic African ancestry on the basis of only 10 alleles. How-
ever, as demonstrated in our study of 20 Portuguese and 20
African individuals, the AAI values performed extremely well at
an individual level. The AAI values observed for the Sa ˜o Tome ´
sample ranged from 2.9 to 13.6, whereas for the Portuguese it
ranged from ?16.4 to ?4.9. On the other hand, when we
analyzed the individuals from Queixadinha, we observed large
variances and very significant overlaps among the three Color
categories (Fig. 2). Only nine of the 30 black individuals had
values within the African range and eight (27%) had AAI values
within the Portuguese range. On the other hand, only 16 of the
29 white individuals had values within the Portuguese range and
none had AAI values within the African range. Moreover, 40 of
the 114 intermediate individuals (35%) had AAI values in the
Portuguese range and 15 (13%) had values within the African
range.Inotherwords,attheindividuallevel,therewasnochance
of obtaining a reliable Color classification on the basis of the
genomic analysis of 10 PSAs.
One of the loci tested in the PSA battery, OCA2, encodes a
melanosome transmembrane protein (P protein) and may be
responsible for part of the normal human variation in pigmen-
tation(6).Thus,itspresenceinthebatterymightintroduceabias
by inflating the AAI values of black-skinned individuals and
conversely deflating them in white individuals, independently of
genomic ancestry. Thus, we recalculated the AAI values for the
Queixadinha Color categories omitting the OCA2 results (data
not shown). Indeed, this recalculation resulted in significant
convergence of AAI values: the median for black individuals
decreased from ?0.87 to ?1.90 and the median for white
individuals increased from ?4.55 to ?2.62. After removal of
OCA2, the difference between the Queixadinha white and black
categories is no longer statistically significant in a two-tailed test
(z ? 1.8, P ? 0.07). These results strengthen our conclusion that
in Brazilians Color is poorly correlated with African ancestry.
We further confirmed our results by using a second sample
composed of 200 unrelated white Brazilian males, who came
from the four major geographic regions of the country. All
regions showed AAI values intermediate between Europeans
and Africans, even southern Brazil, a region predominantly
peopled by European immigrants. In essence, our data indicate
that, in Brazil as a whole, Color is a weak individual predictor of
African ancestry.
If we consider some peculiarities of Brazilian history and social
structure, we can construct a model to explain why Color should
indeedbeapoorpredictorofAfricanancestryatanindividuallevel.
Most Africans have black skin, genetically determined by a very
smallnumberofgenesthatwereevolutionarilyselected(5,6).Thus,
if we have a social race identification system based primarily on
phenotype, such as occurs in Brazil, we classify individuals on the
basis of the presence of certain alleles at a small number of genes
that have impact on the physical appearance, while ignoring all of
the rest of the genome. Assortative mating based on Color, which
has been shown by demographic studies to occur in Brazil (22, 33),
will produce strong associations among the individual components
of Color. Indeed, we detected the presence of such positive asso-
ciations at highly significant levels in the Queixadinha population
(data not shown). On the other hand, we expect that any initial
admixture association between Color and the PSAs will inevitably
decay over time. It is easy to see how this combination of social
forces could produce a population with distinct Color groups and
yet with similar levels of African ancestry. Let us take as an
example, the historically common Brazilian mating of a white
EuropeanmalewithablackAfricanslavewoman:thechildrenwith
more physical African features would be considered black, whereas
those with more European features would be considered white,
even though they would have exactly the same proportion of
African and European alleles. In the next generation, the light-
skinned individuals would assortatively tend to marry other whites
and conversely the darker individuals would marry blacks. The
long-term tendency would then be for this pattern to produce a
white group and a black group, which would, nonetheless, have a
similar proportion of African ancestry.
Our study makes clear the hazards of equating color or race
with geographical ancestry and using interchangeably terms such
as white, Caucasian, and European on one hand, and black,
Negro, or African on the other, as is often done in scientific and
medical literature (34, 35).
WearegratefultoDr.JudithKiddofYaleUniversityforthekinddonation
of Amerindian DNA samples, Dr. Jeffrey Long of the University of
Michigan for allowing us to use his software ADMIX, and three unknown
referees who made invaluable suggestions. We thank Eduardo Tarazona-
Santos for useful discussions and Fabı ´ola A. Cardoso for help in the
statistical analysis. Neuza A. Rodrigues and Ka ´tia Barroso provided expert
technical assistance. This work was supported by grants from the following
Brazilian agencies: Conselho Nacional de Pesquisas, Comissa ˜o de Aper-
feic ¸oamento de Pessoal de Nı ´vel Superior, and Fundac ¸a ˜o de Amparo
a ` Pesquisa do Estado de Minas Gerais, and Portugal’s Fundac ¸a ˜o para a
Cie ˆncia e a Tecnologia (FCT Project POCTI?ANT?4510?2001).
Parra et al. PNAS ?
January 7, 2003 ?
vol. 100 ?
no. 1 ?
181
GENETICS