2. are detected as Pvu II and Xba I restriction fragment–
length polymorphisms (RFLPs) (15,16). They are lo-
cated in intron 1, ⬃400 bp upstream of exon 2 (Figure
1). The Pvu II RFLP detects a T–C substitution at
position ⫺397 before exon 2 that is referred to as
T⫺397int1C, while the Xba I RFLP detects
G⫺351int1A. No functional effect of these sequence
variations on the expression or function of the ER␣
protein has yet been established, and they are therefore
regarded as anonymous polymorphisms. However, such
polymorphisms can be used as markers in association
analyses. Upon association, the marker alleles are sup-
posed to be in linkage with a truly functional allele
elsewhere in the gene.
Until now, only 2 studies investigating the rela-
tionship between ER␣ polymorphisms and OA have
been published. In a case–control study of 383 Japanese
women, Ushiyama et al (17) showed an association
between ER␣ gene polymorphisms and an increased
prevalence of generalized OA. Loughlin et al (18)
studied 740 subjects (371 patients who had undergone
total hip and/or knee replacement surgery for idiopathic
OA and 369 controls) and found no association between
ER␣ gene polymorphisms and OA.
The differing results of these 2 studies can be
explained by the analysis of different OA end points, the
heterogeneity of the study populations, or differences in
statistical power between studies. Furthermore, in the
aforementioned studies, genotypes were based on sepa-
rately determined Pvu II and Xba I RFLPs. Thus, some
ambiguity in the assignment of haplotypes might have
contributed to the discrepant results. However, haplo-
types constructed of the Pvu II and Xba I RFLPs could
be assigned with considerable certainty, because subjects
with a ppXX genotype were detected in neither of these
studies; therefore, the pX haplotype apparently is rare.
Nonetheless, in 4 different studies of ER␣ gene poly-
morphisms, subjects were reported to carry the pX
haplotype, albeit at low frequency (19–22).
In our study, therefore, we constructed haplotype
alleles using a direct molecular haplotyping method that
can determine haplotypes unambiguously. We studied
the association between these haplotypes and radio-
graphic OA of the knee in a large population-based
cohort of 1,483 men and women ages 55 years and older.
To refine our analysis, we also studied separate features
of radiographic knee OA—osteophytosis and joint space
narrowing.
SUBJECTS AND METHODS
Subjects. The study population comprised subjects in
the Rotterdam Study, a prospective, population-based cohort
Figure 1. Direct molecular haplotyping of the Pvu II and Xba I restriction fragment–length polymorphisms
(RFLPs) in the ER␣ gene. A, Part of intron 1 containing the Pvu II and Xba I RFLPs. B, Polymerase chain
reaction (PCR) product and possible restriction fragment combinations specific for each of the possible
haplotypes. Values are the allele frequencies (%) observed in 1,483 men and women drawn from the Rotterdam
Study.
1914 BERGINK ET AL
3. study of the incidence of, and risk factors for, chronic disabling
diseases in elderly persons (ages 55 years and older). The
rationale and study design have been described previously (23).
The focus of the Rotterdam Study is on neurogeriatric, car-
diovascular, ophthalmologic, and locomotor diseases, includ-
ing osteoarthritis and osteoporosis. All 10,275 inhabitants of
the district of Ommoord in Rotterdam, The Netherlands, were
invited to participate. The response rate was 78%, resulting in
7,983 subjects participating in the study. Of these participants,
6,450 visited a research center for a baseline examination
between 1990 and 1993. Written informed consent was ob-
tained from each subject. The Rotterdam Study was approved
by the medical ethics committee of the Erasmus Medical
Center. A total of 1,483 men and women from the Rotterdam
Study population were selected for our study, based on avail-
ability of data on radiographic OA of the knees, data on
possible confounding factors, and the availability of blood
samples suitable for DNA analysis.
Osteoarthritis. At baseline (between 1991 and 1993),
anteroposterior weight-bearing radiographs of the knees, with
the patellae in the central position, were obtained and graded
for radiographic OA on a 5-point scale (0–4) according to the
Kellgren/Lawrence (K/L) system (24), using an atlas (25). The
radiographs were scored for OA by 2 independent observers
who were blinded to all data for the participant, as described
previously (26). After each set of 150 radiographs, the scores of
the 2 readers were evaluated. Whenever the K/L score dif-
fered, the 2 readers met to read the radiographs together, and
a consensus score was determined. A subject was diagnosed as
having radiographic OA of the knee if the K/L score for 1 or
both knees was ⱖ2. The presence of osteophytes was scored
separately in both knees at 4 different sites (medial and lateral
distal femur, and medial and lateral proximal tibia) using a
4-point scale (0 ⫽ no osteophyte; 1 ⫽ minute osteophyte,
doubtful significance; 2 ⫽ definite osteophyte, moderate size;
3 ⫽ large osteophyte), in concordance with the K/L grading
scale (24). These scores were summed, and subjects whose
scores were in the sex-specific highest quartile of this sum score
were regarded as having osteophytosis. Additionally, in 1,276
persons (86%) randomly selected from the study population,
the distances between the femur and tibia at the middle of the
medial and lateral knee joint compartment were measured in
millimeters with a graduated ruler, and these measurements
were summed for both knees. Subsequently, subjects whose
scores were in the lowest sex-specific quartile of this sum score
were considered to have joint space narrowing (JSN). The
methods used for determining the separate features of radio-
graphic OA are similar to those published previously (27,28).
Other variables. Between 1990 and 1993, an extensive
baseline home interview on medical history, risk factors for
chronic diseases, and medication use was performed by trained
interviewers. Information on smoking status was obtained, and
female participants were asked about their age at and reason
for menopause (defined as cessation of menses for 12 consec-
utive months). After the home interview, subjects were invited
to the research center to undergo clinical examination and
laboratory evaluation. Height and weight were measured with
subjects wearing indoor clothing and no shoes. The body mass
index (BMI) was computed as weight in kilograms divided by
height in squared meters (kg/m2
). Bone mineral density
(BMD) measurement of the femoral neck was performed by
dual x-ray absorptiometry (DXA) (Lunar DPX-L densitome-
ter; Lunar, Madison, WI) as described previously (29).
ER␣ haplotyping. Genomic DNA was isolated from
peripheral leukocytes by standard procedures. The direct
molecular haplotyping of the Pvu II and Xba I RFLPs was
performed as shown in Figure 1. A 346-bp polymerase chain
reaction (PCR) fragment was generated by a forward primer
(5⬘-GAT-ATC-CAG-GGT-TAT-GTG-GCA-3⬘) and a reverse
primer (5⬘-AGG-TGT-TGC-CTA-TTA-TAT-TAA-CCT-
TGA-3⬘) in a 10-l reaction mixture containing 20 ng genomic
DNA, 50 mM KCl, 10 mM Tris HCl (pH 8.3), 1.5 mM MgCl2,
0.2 mM deoxynucleoside triphosphate, 2 pM of each primer,
and 0.2 units of Super Taq DNA polymerase (HT Biotechnol-
ogy, Cambridge, UK). The reactions were performed in a
384-well format on an MJ Tetrad thermocycler (MJ Research,
San Francisco, CA) with a cycling protocol of 94°C, 60°C, and
72°C for 45 seconds each, for 30 cycles. Ten microliters of PCR
product was digested by simultaneously adding 5 l of diges-
tion mixture containing 5 units of Pvu II, 7 units of Xba I
restriction enzyme (both from MBI Fermentas, Hanover,
MD), and 1.5 l of REact Buffer 2 (Life Technologies, Breda,
The Netherlands) and incubating for 90 minutes at 37°C. The
digestion products were analyzed by electrophoresis in a 3%
agarose gel in 0.5⫻ TBE (1⫻ TBE ⫽ 89 mM Tris, 89 mM boric
acid, 2 mM Na2EDTA) for 80 minutes at 125 volts. Separa-
tion patterns were documented with a DC120 digital camera
(Kodak, Rochester, NY) under ultraviolet illumination
(302 nm).
To compare ER␣ Pvu II–Xba I haplotype frequencies
between our study population and the populations in Oxford,
UK (18) and Seta, Japan (17), inferred haplotype frequencies
for those studies were calculated to correspond with those
derived from our direct haplotyping method. In this way, the
frequencies of identical alleles in the different study popula-
tions could be compared. For the Seta study, haplotype
frequencies were derived from published genotype frequen-
cies. Haplotype combinations could be unambiguously deter-
mined for all genotypes except the PpXx genotype. This
genotype was assumed to consist of a combination of the
haplotypes px and PX, because subjects homozygous for the
alternative haplotype (pX) were not found in any of the
populations studied.
Statistical analysis. Differences in mean age at base-
line between the study group and subjects in the Rotterdam
Study were evaluated by analysis of variance (ANOVA), and
differences in the sex ratio were determined by the chi-square
test. All other differences in baseline characteristics were
compared by analysis of covariance (ANCOVA), with age and
sex as covariates to adjust for possible confounding effects.
Differences in baseline characteristics according to radiologic
knee OA status were analyzed in the same way.
Differences in baseline characteristics between the
different genotype groups of the ER␣ gene were compared as
follows: we allowed for 3 possible models to explain differences
between groups (i.e., an allele-dose effect, a dominant effect,
or a recessive effect). Allele-dose effect was defined as an
increase or decrease in value per copy of the allele. In case of
a consistent trend reflected as an allele-dose effect, we per-
formed a (multiple) linear or logistic regression analysis to
quantify the association. In case of a dominant or recessive
effect of the test allele, ANOVA and ANCOVA were per-
ER␣ GENE AND RADIOGRAPHIC KNEE OA 1915
4. formed. For dominant effects, we compared test allele carriers
versus noncarriers; for recessive effects, subjects homozygous
for the test allele were compared with heterozygous carriers
and noncarriers.
Odds ratios (ORs) with 95% confidence intervals (95%
CIs) were calculated by (multiple) logistic regression analyses
to estimate the relative risk of knee radiographic OA for
subjects with 1 or 2 copies of the risk allele, compared with
subjects not carrying the risk allele. First, we calculated crude
ORs, followed by adjustment for potentially confounding
factors (e.g., age, sex, BMI, femoral neck BMD, and smoking
status). Subsequently, crude and adjusted ORs were calculated
stratified by sex. For women, the ORs were additionally
adjusted for age at menopause. We used SPSS version 10.1.0
(SPSS, Chicago, IL) for all of our analyses.
RESULTS
The 3 different alleles we found using the direct
haplotyping method and their frequencies are presented
in Tables 1 and 2. We did not observe the fourth possible
haplotype (pX; ⫺397int1T and ⫺351int1G) in our study
population. To compare this distribution with those
found in other study populations, we also present the
haplotype allele frequencies for the control subjects in
the Oxford study (i.e., persons who had not undergone
total knee and/or hip replacement surgery) and in the
Seta study (i.e., women without generalized OA)
(17,18). The haplotype frequencies among subjects in
the Oxford and Seta studies were inferred, as explained
in Subjects and Methods.
The frequencies of alleles in our study population
were similar to those observed in the UK study popula-
tion. In both populations, the frequencies of alleles PX
and Px differed between men and women, with an
overrepresentation of allele PX in women and of allele
Px in men (P ⫽ 0.017, by chi-square test). In contrast,
the frequency of haplotype allele PX in Japanese sub-
jects was 18%, compared with 34% among Caucasian
subjects in the other 2 studies, while the frequency of
haplotype Px was higher in Japanese individuals (23%)
than in Caucasians (11%). The 3 alleles combined to
form 6 different genotypes, the distribution of which did
not deviate from Hardy-Weinberg equilibrium, also
when stratified by sex. The genotype frequencies in our
study population corresponded to those in the UK
population but differed from those among Japanese
subjects.
Table 3 shows the frequencies of the 3 different
haplotype alleles in our study population according to
the K/L score, the presence of osteophytosis, and JSN of
the knee. For reasons pertaining to statistical power, we
combined K/L scores 3 and 4. When allele frequencies
were calculated in subjects grouped by the K/L score of
the knee, the percentage of allele PX increased signifi-
cantly with each point increase in the K/L score. The
frequency of allele px decreased as the K/L score
increased. There were no consistent differences in the
frequencies of allele Px according to the K/L score. The
trends observed were the same in both men and women.
Because, in general, allele PX was associated with an
increased risk for radiographic OA, we defined this as a
“risk allele” in our further analyses. When considering the
2 separate features of radiographic OA that we measured,
an underrepresentation of allele px and an overrepresen-
tation of allele PX were observed in subjects with osteo-
Table 1. Frequencies of the ER␣ haplotype alleles in 3 study
populations*
Allele
Nucleotide
position Rotterdam Oxford
Seta,
women
⫺397 ⫺351 All Men Women All Men Women
px T A 54 54 54 55 56 53 59
PX C G 34 32 35 34 31 39 18
Px C A 12 14 11 11 13 8 23
* Values are the percentage.
Table 2. Frequencies of the ER␣ genotypes in 3 study populations*
Genotype
Rotterdam Oxford
Seta,
female controls
(n ⫽ 318)
All subjects
(n ⫽ 1,483)
Men
(n ⫽ 611)
Women
(n ⫽ 872)
All controls
(n ⫽ 369)
Men
(n ⫽ 221)
Women
(n ⫽ 148)
pxpx 434 (29) 179 (29) 255 (29) 110 (30) 67 (30) 43 (29) 115 (36)
pxPX 550 (37) 215 (35) 335 (38) 138 (37) 79 (36) 59 (40) 65 (20)
pxPx 187 (13) 87 (14) 100 (11) 49 (13) 36 (16) 13 (9) 81 (25)
PXPX 158 (11) 57 (9) 101 (12) 43 (12) 21 (10) 22 (15) 9 (3)
PXPx 132 (9) 64 (10) 68 (8) 27 (7) 16 (7) 11 (7) 33 (10)
PxPx 22 (1) 9 (1) 13 (1) 2 (0) 2 (1) 0 (0) 15 (5)
P† 0.57 0.37 0.80 0.60 0.66 0.71 0.46
* Values are the number (percentage).
† Determined by Hardy-Weinberg equilibrium test.
1916 BERGINK ET AL
5. phytosis, while no consistent differences in allele frequen-
cies were seen in subjects with JSN of the knee.
The baseline characteristics of the study subjects
according to radiographic knee OA status and number
of copies of ER␣ allele PX are shown in Table 4. The
overall prevalence of radiographic OA in our population
was 24.3% (16.4% in men, 29.9% in women). Radio-
graphic knee OA, defined as a K/L score ⱖ2 in 1 or both
knees, was more frequent in women (72%) than in men
(55%). In addition, subjects with radiographic knee OA
were, on average, nearly 3 years older than those without
radiographic knee OA, and had a 6% higher adjusted
BMI and a 5% increased age- and sex-adjusted femoral
neck BMD. Current smokers were more common
among subjects in whom radiographic OA was absent
than among those in whom it was present (27% versus
21%). In our study population, no significant difference
in the mean age at menopause was found between
women with and those without radiographic knee OA.
Analysis according to number of copies of hap-
lotype allele PX revealed no significant differences in
the baseline characteristics of subjects, except for cur-
rent smoking and age at natural menopause. The age at
menopause in women homozygous for allele PX was ⬃1
year earlier than that in women carrying 1 or no copy of
the allele. Additional adjustment for smoking status did
not change the values for age at menopause for the
different genotypes.
Table 3. Allele frequencies of the ER␣ haplotype alleles by K/L score, osteophytosis, and JSN of the knee*
Overall
All subjects ER␣ haplotypes Men ER␣ haplotypes Women ER␣ haplotypes
No. px PX Px No. px PX Px No. px PX Px
K/L 1,483 54 34 12 611 54 32 14 872 54 35 11
0 878 55 32 13 417 55 31 14 461 56 32 12
1 244 54 34 12 94 54 32 14 150 54 35 11
2 320 52 38 10 88 51 39 11 232 52 38 10
3–4 41 48 40 12 12 50 25 25 29 47 47 7
P trend – 0.04 ⬍0.01 0.25 – 0.27 0.17 0.79 – 0.08 ⬍0.01 0.20
Osteophytosis
Absent 1,111 55 32 13 477 55 31 14 634 56 33 11
Present 372 51 38 11 134 52 35 13 238 50 40 10
P – 0.02 ⬍0.01 0.36 – 0.35 0.22 0.74 – 0.03 ⬍0.01 0.50
JSN
Absent 974 55 34 12 392 56 31 13 582 54 35 11
Present 302 53 33 14 133 51 33 16 169 55 33 12
P – 0.58 0.78 0.21 – 0.15 0.50 0.23 – 0.62 0.39 0.60
* Except where indicated otherwise, values are the percentage. Joint space narrowing (JSN) scores were based on 1,276 subjects (86% of study
population). K/L ⫽ Kellgren/Lawrence; JSN ⫽ joint space narrowing.
Table 4. Baseline characteristics of subjects*
Characteristic Total
ROA (K/L score ⱖ2)
P
No. of copies of ER␣ haplotype PX
P
Absent Present 0 1 2
All subjects
No. 1,483 1,122 361 – 643 682 158 –
Sex, no. (%) women 872 (59) 611 (55) 261 (72) ⬍0.01 386 (57) 403 (59) 101 (64) 0.15†
Age, years 68.8 ⫾ 7.5 68.1 ⫾ 7.2 71.3 ⫾ 7.7 ⬍0.01 68.8 ⫾ 7.6 68.8 ⫾ 7.4 69.1 ⫾ 7.5 0.71‡
Body mass index, kg/m2
26.1 ⫾ 3.5 25.7 ⫾ 3.4 27.3 ⫾ 3.5 ⬍0.01 26.2 ⫾ 3.5 25.9 ⫾ 3.5 26.0 ⫾ 3.6 0.16§
Femoral neck BMD, gm/cm2
0.833 ⫾ 0.14 0.818 ⫾ 0.13 0.859 ⫾ 0.15 ⬍0.01 0.826 ⫾ 0.14 0.829 ⫾ 0.14 0.828 ⫾ 0.14 0.72§
Current smoker, no. (%) 356 (24) 303 (27) 76 (21) 0.02 148 (23) 184 (27) 44 (28) 0.04§
Women only
No. 872 611 261 – 368 403 101 –
Age at menopause, years 48.6 ⫾ 4.9 48.7 ⫾ 5.1 48.6 ⫾ 4.4 0.78 48.8 ⫾ 4.6 48.7 ⫾ 5.1 47.7 ⫾ 4.8 0.03‡
* Except where indicated otherwise, values are the mean ⫾ SD. Values for age, sex, and age at menopause are crude values, with P estimated by
analysis of variance or by chi-square test. All other values were adjusted for age and sex (values for bone mineral density [BMD] of the femoral neck
were additionally adjusted for body mass index), with P estimated by analysis of covariance. ROA ⫽ radiographic osteoarthritis; K/L ⫽
Kellgren/Lawrence.
† Analyzed as dose effect.
‡ Analyzed as recessive effect.
§ Analyzed as dominant effect.
ER␣ GENE AND RADIOGRAPHIC KNEE OA 1917
6. In Table 5, the associations between ER␣ geno-
types (0, 1, and 2 copies of haplotype allele PX) and
outcomes of radiographic knee OA are presented. The
percentage of participants with radiographic OA defined
by a K/L score ⱖ2 and those with osteophytosis in-
creased with each copy of allele PX. After stratifying by
sex, there seemed to be a dominant effect in men and a
recessive effect in women when radiographic OA was
defined by the K/L score. When considering the pres-
ence of osteophytosis, we observed a dose effect in both
men and women, with an increase in the percentage of
subjects with radiographic OA with each copy of the PX
risk allele. The crude OR for the presence of radio-
graphic knee OA, defined as a K/L score ⱖ2, is 1.2 (95%
CI 0.9–1.5) for heterozygous subjects and 1.9 (95% CI
1.3–2.8) for homozygous subjects (data not shown).
Adjustment for confounding factors resulted in
slightly higher point estimates. When we calculated the
allele-dose effect adjusted for confounding factors, this
proved to be significant (P ⬍ 0.01). When we stratified
by sex, we observed risk estimates similar to those of the
total study group, with a significant allele-dose effect for
both men and women (P ⫽ 0.03 and P ⬍ 0.01, respec-
tively). The population-attributable risk was 6.8% for
subjects carrying 1 ER␣ risk allele (9.0% for men, 10.7%
for women) and 24.0% for homozygous subjects (15.3%
for men, 29.5% for women). When we considered the
association of osteophytes and the different genotypes
of the risk allele, ORs similar to those for radiographic
OA according to the K/L score were found. No signifi-
cant association between ER␣ haplotypes and JSN was
observed in our study population.
In Table 6, the ER␣ haplotype allele frequencies
according to the end point of OA as measured in the 3
study populations are shown. Whereas all patients with
OA in the Oxford study had undergone total knee
and/or hip replacement surgery, patients in the Seta
study had generalized OA, defined as a K/L score ⱖ2 in
at least 3 interphalangeal joints of each hand. When a
score ⱖ3 in 1 or more interphalangeal joints was
present, OA was considered severe. In the Rotterdam
Study population, subjects were stratified according to
the presence (K/L score ⱖ2) or absence (K/L score 0 or
1) of radiographic knee OA. For subjects in the Rotter-
dam Study, we observed associations similar to those
shown in Table 3. When considering the allele frequen-
cies in the Oxford population for both men and women
combined, no difference in distribution was seen be-
tween patients who had undergone total knee and/or hip
replacement surgery and controls. Among men in the
Oxford study, allele PX was overrepresented in patients
compared with controls, whereas among women the PX
allele was underrepresented in patients; however, nei-
ther of these differences was significant. In the Seta
study, the frequency of allele PX increased as the
severity of generalized OA increased. However, this
trend failed to reach statistical significance (P ⫽ 0.09).
DISCUSSION
In a large, population-based study of elderly
Caucasian individuals living in The Netherlands, we
observed an increase in the percentage of subjects with
prevalent radiographic knee OA with each increase in
copy of the haplotype allele PX of the ER␣ gene. After
adjustment for confounding factors, women homozy-
Table 5. Association between ER␣ haplotype allele PX genotypes and different outcomes of ROA of the knee*
ROA outcome,
allele PX
All subjects Men Women
Cases/total
(%)
OR
(95% CI)
Cases/total
(%)
OR
(95% CI)
Cases/total
(%)
OR
(95% CI)
K/L score ⱖ2
0 copies 140/643 (22) 1 37/275 (13) 1 103/368 (28) 1
1 copy 166/682 (24) 1.3 (0.9–1.7) 52/279 (19) 1.6 (1.0–2.6) 114/403 (28) 1.2 (0.8–1.7)
2 copies 55/158 (35) 2.2 (1.5–3.4) 11/57 (19) 2.1 (0.9–4.5) 44/101 (44) 2.4 (1.4–3.9)
Osteophytosis
0 copies 144/643 (22) 1 55/275 (20) 1 89/368 (24) 1
1 copy 172/682 (25) 1.3 (1.0–1.7) 63/279 (23) 1.2 (0.8–1.9) 109/403 (27) 1.4 (1.0–2.0)
2 copies 56/158 (35) 2.3 (1.5–3.4) 16/57 (28) 1.8 (0.9–3.7) 40/101 (40) 2.7 (1.6–4.5)
Joint space narrowing
0 copies 131/557 (24) 1 59/243 (24) 1 72/314 (23) 1
1 copy 143/585 (24) 1.1 (0.8–1.4) 60/234 (26) 1.0 (0.6–1.5) 83/351 (24) 1.2 (0.8–1.7)
2 copies 28/134 (21) 0.9 (0.5–1.4) 14/48 (29) 1.4 (0.7–2.8) 14/86 (16) 0.6 (0.3–1.3)
* Odds ratios (ORs) for all subjects were adjusted for age, sex, body mass index (BMI), femoral neck bone mineral density (BMD), and smoking
status. ORs for men were adjusted for age, BMI, femoral neck BMD, and smoking status. ORs for women were adjusted for age, BMI, femoral neck
BMD, smoking status, and age at menopause. ROA ⫽ radiographic osteoarthritis; 95% CI ⫽ 95% confidence interval; K/L ⫽ Kellgren/Lawrence.
1918 BERGINK ET AL
7. gous for this risk allele had more than a 2-fold increased
risk for radiographic knee OA compared with subjects
without the risk allele, with a similar trend toward
increased risk in homozygous men. When we calculated
the allele-dose effect, this proved to be significant for
both men and women. The overall population-
attributable risk for OA in homozygous subjects was
24.0%.
To our knowledge, the only previous study show-
ing a positive association between the ER␣ gene and the
presence of OA is a case–control study of 383 Japanese
women with generalized OA (17). Because we calculated
Pvu II–Xba I haplotype frequencies based on the pub-
lished genotype results from that study, we could com-
pare results from the Japanese study with those from our
study. Our results are consistent with the findings of
Ushiyama et al, who reported an increased risk of
generalized OA in women with the PpXx genotype
compared with all other genotypes (17). When their
results are analyzed according to the frequency of hap-
lotype allele PX (our risk allele), we observe an increase
in frequency of the PX allele with increased severity of
generalized OA in the Japanese cohort. Thus, although
there is a different distribution of the ER␣ haplotype
alleles between Dutch and Japanese populations, allele
PX seems also to be a risk allele for OA in Japanese
women. This is consistent with the observation that OA
of the knee is associated with generalized OA (30,31).
For example, of the 65 cases of generalized OA in the
Seta study, 53 Japanese women had radiologically de-
fined OA of the knee (17).
The Oxford study on the relationship between
the ER␣ gene and OA did not demonstrate an associa-
tion when the data were analyzed by genotypes con-
structed from haplotypes (18). When the data were
analyzed by allele PX and stratified by sex, we observed
a nonsignificant increased frequency of the PX haplo-
type in men who had undergone total joint replacement
surgery. However, in women who had undergone total
joint replacement surgery, the frequency of the risk
allele was decreased. The reason for this difference in
association results is unclear, but the difference in OA
end points used in these studies might be a contributing
factor. For example, the indication for total joint re-
placement is based only partly on radiologic signs of OA,
while other important determinants, such as pain and
disability, are influenced by sex (32).
By using only genotype combinations of the Pvu
II and Xba I RFLPs, haplotypes can be estimated only
indirectly (i.e., inferred). In populations in which all
possible haplotypes exist, such as the population studied
by Albagha et al (20), it would be difficult, for example,
to differentiate a genotype consisting of PX–px from a
genotype consisting of Px–pX. To overcome this poten-
tial problem, we developed a direct molecular haplotyp-
ing protocol, which can unambiguously identify all pos-
sible haplotypes at this locus. Using this method, we
could not identify the pX haplotype in our study popu-
lation. The fact that this haplotype was not frequent in
either the Oxford study or the Seta study suggests that
this haplotype is extremely rare and/or is confined to
particular populations (19–22). The molecular haplotyp-
ing method also allowed us to analyze potential allele-
dose effects for individual haplotype alleles. Finding an
allele-dose effect increases the probability that an asso-
ciation is based on a true effect and not on a chance
Table 6. Frequencies of the ER␣ haplotype alleles in different study populations*
Study population/
patient definition No.
All subjects ER␣
haplotypes
No.
Men ER␣ haplotypes
No.
Women ER␣ haplotypes
px PX Px px PX Px px PX Px
Rotterdam/ROA knee
Absent 1,122 55 32 13 511 55 31 14 611 55 33 12
Present 361 51 38 11 100 51 37 13 261 51 39 10
P – 0.06 ⬍0.01 0.14 – 0.28 0.11 0.55 0.12 0.02 0.31
Oxford/TJR
Controls 369 55 34 11 221 56 31 13 148 53 39 8
Cases 102 53 35 11 46 52 38 10 56 54 33 13
P – 0.66 0.73 0.86 – 0.47 0.19 0.44 – 0.97 0.39 0.17
Seta/GOA
Controls – – – – – – – – 318 59 18 23
Cases – – – – – – – – 65 56 23 21
Severe cases – – – – – – – – 47 55 25 20
P trend – – – – – – – – – 0.40 0.09 0.52
* Except where otherwise indicated, values are the percentage. ROA ⫽ radiographic osteoarthritis; TJR ⫽ total joint (hip and/or knee) replacement;
GOA ⫽ generalized OA. In the Rotterdam Study, patients were defined by the presence or absence of ROA. In the Oxford study, cases were patients
who had undergone TJR. In the Seta study, patients had GOA.
ER␣ GENE AND RADIOGRAPHIC KNEE OA 1919
8. finding. This analytic approach therefore improves the
genetic resolution and strength of the analysis, similar to
what we previously demonstrated for the vitamin D
receptor gene (33,34). The distribution of ER␣ geno-
types in our study population, as defined by haplotypes
of the Pvu II and Xba I RFLPs, is similar to those
previously reported in other Caucasian populations
(20,35) and was found to be in Hardy-Weinberg equili-
brium.
We chose to examine radiographic OA of the
knee, because for this type of radiographic OA, data on
2 separate radiologic features of OA (osteophytosis and
JSN) were available apart from the K/L score. Further-
more, radiographic knee OA is associated with general-
ized OA and is therefore thought to be more strongly
influenced by genetic factors (30,31,36,37). We defined
radiographic OA as a K/L score ⱖ2 in 1 or both knees,
because this definition is used in the clinical setting.
Furthermore, by using this definition, the subjects with
radiographic OA roughly correspond with a quartile of
the total study population. To further investigate what
separate feature of radiographic OA is associated with
the risk allele, we used quartiles of a sum score for both
knees, because we hypothesized that any association
between the ER␣ gene and radiographic OA would be
systemic.
Polymorphisms in the ER␣ gene previously have
been shown to be associated with differences in BMD
(20,21), and the presence of OA is associated with
increased BMD (38–40). The latter was also observed in
our study cohort as baseline BMD differences in subjects
with and those without radiographic OA of the knee
(Table 4). Thus, it might be argued that the relationship
between OA and ER␣ polymorphisms is based on
differences in BMD. However, because the ORs did not
essentially change after accounting for BMD differ-
ences, we do not think this is a likely explanation.
We hypothesize that polymorphisms in the ER␣
gene may be associated with a modulatory role on bone
metabolism during adolescence and young adulthood
(41), thus possibly leading to increased susceptibility for
OA later in life as a result of altered loading of
weight-bearing joints. Estrogen receptors are present on
both articular cartilage (12,13,42) and bone cells (14).
Therefore, hypothetically, both cartilage and bone may
be involved in the association between the ER␣ gene
and radiographic OA. In our study, the observed asso-
ciation seems to be driven by osteophytosis, for which we
found an allele-dose effect in both men and women. This
suggests that variations in the ER␣ gene are associated
with changes in (juxtaarticular) bone or its response to
external influences, rather than articular cartilage, lead-
ing to more severe OA later in life. However, the
observed lack of association with JSN could also be
attributable to the way in which we measured this
radiographic feature (43–45). Although currently more
accurate methods to measure JSN are available, the
choice for the method we used was made earlier in the
course of this large epidemiologic study.
The role of age at menopause in the relationship
between the ER␣ gene and radiographic OA of the knee
is unclear. Women homozygous for the risk allele PX
reached menopause ⬃1 year earlier than women with no
or 1 copy. This is consistent with previous findings that
carriers of the P allele have an earlier age at menopause
and are more likely to undergo hysterectomy, leading to
premature menopause (46). This could potentially con-
found the association between the ER␣ gene and radio-
graphic OA of the knee. However, in our study popula-
tion, the age at menopause in women with radiographic
OA of the knee was similar to that in women without
radiographic OA. Furthermore, adjustment of the ORs
for age at menopause resulted in similar point estimates.
Thus, in this group, the association between the ER␣
gene and radiographic knee OA is not likely to be
explained by differences in age at menopause.
Our study has several limitations. First, because
the participants had to be healthy enough to come to our
research center for knee radiography, there might be a
health selection bias in our study population. However,
except for differences in age, no major differences in
baseline characteristics between the source population
of the Rotterdam Study and our study population were
seen. Furthermore, any association study is potentially
prone to stratification bias. However, because the sam-
ple we used is large, ethnically homogeneous, and
population-based, we do not believe the results we found
are influenced by this type of bias. Second, the 2 intronic
ER␣ polymorphisms detected as Pvu II and Xba I
RFLPs are, as far as we know, anonymous. Therefore,
when association is found it is assumed that alleles of
these polymorphisms are in linkage disequilibrium with
1 or more truly functional alleles of polymorphisms
located elsewhere in the gene.
Recently, functional differences between these
genotypes resulting in enhanced transcriptional activity
of intronic sequences containing the 2 RFLPs constitut-
ing the Px and px haplotypes were observed (47).
Alternatively, the Pvu II–Xba I RFLP haplotypes can be
in linkage with a truly functional polymorphism else-
where in the ER␣ gene. In the promoter region, there is
a (TA)n variable-number tandem repeat (VNTR), and it
1920 BERGINK ET AL
9. has previously been shown that this marker is in strong
linkage disequilibrium with the Pvu II–Xba I haplotypes
(19,20,35). In view of its location, 1 kb upstream of exon
1 in the 1A promoter of the ER␣ gene, the (TA)n VNTR
might be driving the association we observe. This is
consistent with the notion that the more sensitive ER␣
genotype that is assumed to exist is explained by higher
levels of ER␣ messenger RNA and protein. Salmen et al
reported that women carrying the P allele benefit more
from the protective effect of hormone replacement
therapy on fracture risk than those with the pp genotype,
which was considered to be a relatively estrogen-
insensitive genotype (48). Possibly, the relatively greater
sensitivity of ER␣ haplotype allele PX carriers com-
pared with those without this allele leads to increased
stimulation of local bone formation (49,50), resulting in
greater susceptibility to forming osteophytes (51,52).
Sensitivity, in this respect, could be explained by higher
expression levels of the ER␣ protein of this genotype,
particularly in target cells of the estrogen endocrine system.
We realize that any mechanism by which
genotype-related differences in the ER␣ gene could
influence the development of OA is speculative. Observ-
ing these associations in men as well as in women is not
unexpected. The estrogen transduction pathway is very
pleiotropic, and its actions are not limited to women. In
this respect, it should be noted that the ER␣ gene is
situated on an autosomal chromosome (6q25). Further-
more, estrogen can play a role in male metabolism
through local conversion of testosterone to estrogen
through the enzyme aromatase (Cyp19). Further analy-
ses of the functional effects of these polymorphisms
must be performed to elucidate the underlying molecu-
lar mechanism.
In conclusion, this population-based study shows
an allele-dose effect of a haplotype allele of the ER␣
gene on radiographically defined knee OA in both men
and women. This is consistent with previous findings on
the association between ER␣ polymorphisms and gen-
eralized OA in Japanese women. The association with
osteophytosis, rather than joint space narrowing, in
particular seems to contribute to the effect. The func-
tional mechanisms behind these findings remain to be
clarified, but linkage disequilibrium with a truly func-
tional polymorphism in the promoter region might be a
likely explanation.
ACKNOWLEDGMENTS
The authors thank Drs. Omar M. E. Albagha and
Stuart H. Ralston (University of Aberdeen, UK) for sharing
unpublished sequence information on the first intron of the
ER␣ gene. We are very grateful to the DXA and radiograph
technicians, L. Buist and H. W. M. Mathot. We also thank
Dr. E. Odding and Professor H. A. Valkenburg for scoring the
baseline radiographs and F. van Rooij, E. van der Heijden,
R. Vermeeren, and L. Verwey for collection of followup data.
Moreover, we thank the participating general practitioners and
the many field workers at the research center in Ommoord.
REFERENCES
1. Cecil R, Archer B. Arthritis of the menopause. JAMA 1925;84:
75–9.
2. Spector TD, Campion GD. Generalised osteoarthritis: a hormon-
ally mediated disease. Ann Rheum Dis 1989;48:523–7.
3. Cauley JA, Kwoh CK, Egeland G, Nevitt MC, Cooperstein L,
Rohay J, et al. Serum sex hormones and severity of osteoarthritis
of the hand. J Rheumatol 1993;20:1170–5.
4. Nevitt MC, Felson DT, Williams EN, Grady D. The effect of
estrogen plus progestin on knee symptoms and related disability in
postmenopausal women: the Heart and Estrogen/Progestin Re-
placement Study, a randomized, double-blind, placebo-controlled
trial. Arthritis Rheum 2001;44:811–8.
5. Dennison EM, Arden NK, Kellingray S, Croft P, Coggon D,
Cooper C. Hormone replacement therapy, other reproductive
variables and symptomatic hip osteoarthritis in elderly white
women: a case-control study. Br J Rheumatol 1998;37:1198–202.
6. Hart DJ, Doyle DV, Spector TD. Incidence and risk factors for
radiographic knee osteoarthritis in middle-aged women: the
Chingford Study. Arthritis Rheum 1999;42:17–24.
7. Spector TD, Nandra D, Hart DJ, Doyle DV. Is hormone replace-
ment therapy protective for hand and knee osteoarthritis in
women? The Chingford Study. Ann Rheum Dis 1997;56:432–4.
8. Wluka AE, Davis SR, Bailey M, Stuckey SL, Cicuttini FM. Users
of oestrogen replacement therapy have more knee cartilage than
non-users. Ann Rheum Dis 2001;60:332–6.
9. Turner AS, Athanasiou KA, Zhu CF, Alvis MR, Bryant HU.
Biochemical effects of estrogen on articular cartilage in ovariec-
tomized sheep. Osteoarthritis Cartilage 1997;5:63–9.
10. Zhang Y, McAlindon TE, Hannan MT, Chaisson CE, Klein R,
Wilson PW, et al. Estrogen replacement therapy and worsening of
radiographic knee osteoarthritis: the Framingham Study. Arthritis
Rheum 1998;41:1867–73.
11. Tsai CL, Liu TK, Chen TJ. Estrogen and osteoarthritis: a study of
synovial estradiol and estradiol receptor binding in human osteo-
arthritic knees. Biochem Biophys Res Commun 1992;183:1287–91.
12. Richmond RS, Carlson CS, Register TC, Shanker G, Loeser RF.
Functional estrogen receptors in adult articular cartilage: estrogen
replacement therapy increases chondrocyte synthesis of proteogly-
cans and insulin-like growth factor binding protein 2. Arthritis
Rheum 2000;43:2081–90.
13. Ushiyama T, Ueyama H, Inoue K, Ohkubo I, Hukuda S. Expres-
sion of genes for estrogen receptors alpha and beta in human
articular chondrocytes. Osteoarthritis Cartilage 1999;7:560–6.
14. Ciocca DR, Roig LM. Estrogen receptors in human nontarget
tissues: biological and clinical implications. Endocr Rev 1995;16:
35–62.
15. Zuppan PJ, Hall JM, Ponglikitmongkol M, Spielman R, King MC.
Polymorphisms at the estrogen receptor (ESR) locus and linkage
relationships on chromosome 6q [abstract]. Cytogenet Cell Genet
1989;51:1116.
16. Yaich L, Dupont WD, Cavener DR, Parl FF. Analysis of the PvuII
restriction fragment-length polymorphism and exon structure of
the estrogen receptor gene in breast cancer and peripheral blood.
Cancer Res 1992;52:77–83.
17. Ushiyama T, Ueyama H, Inoue K, Nishioka J, Ohkubo I, Hukuda
ER␣ GENE AND RADIOGRAPHIC KNEE OA 1921
10. S. Estrogen receptor gene polymorphism and generalized osteo-
arthritis. J Rheumatol 1998;25:134–7.
18. Loughlin J, Sinsheimer JS, Mustafa Z, Carr AJ, Clipsham K,
Bloomfield VA, et al. Association analysis of the vitamin D
receptor gene, the type I collagen gene COL1A1, and the estrogen
receptor gene in idiopathic osteoarthritis. J Rheumatol 2000;27:
779–84.
19. Becherini L, Gennari L, Masi L, Mansani R, Massart F, Morelli A,
et al. Evidence of a linkage disequilibrium between polymorphisms
in the human estrogen receptor alpha gene and their relationship
to bone mass variation in postmenopausal Italian women. Hum
Mol Genet 2000;9:2043–50.
20. Albagha OM, McGuigan FE, Reid DM, Ralston SH. Estrogen
receptor alpha gene polymorphisms and bone mineral density:
haplotype analysis in women from the United Kingdom. J Bone
Miner Res 2001;16:128–34.
21. Kobayashi S, Inoue S, Hosoi T, Ouchi Y, Shiraki M, Orimo H.
Association of bone mineral density with polymorphism of the
estrogen receptor gene. J Bone Miner Res 1996;11:306–11.
22. Stavrou I, Zois C, Ioannidis JP, Tsatsoulis A. Association of
polymorphisms of the oestrogen receptor alpha gene with the age
of menarche. Hum Reprod 2002;17:1101–5.
23. Hofman A, Grobbee DE, de Jong PT, van den Ouweland FA.
Determinants of disease and disability in the elderly: the Rotter-
dam Elderly Study. Eur J Epidemiol 1991;7:403–22.
24. Kellgren JH, Lawrence JS. Radiological assessment of osteoarthri-
tis. Ann Rheum Dis 1957;16:494–501.
25. Kellgren JH, Jeffrey MR, Ball J. The epidemiology of chronic
rheumatism. Atlas of standard radiographs of arthritis. Oxford:
Blackwell Scientific Publications; 1963.
26. Odding E, Valkenburg HA, Algra D, Vandenouweland FA,
Grobbee DE, Hofman A. Associations of radiological osteoarthri-
tis of the hip and knee with locomotor disability in the Rotterdam
Study. Ann Rheum Dis 1998;57:203–8.
27. Uitterlinden AG, Burger H, Huang Q, Odding E, van Duijn CM,
Hofman A, et al. Vitamin D receptor genotype is associated with
radiographic osteoarthritis at the knee. J Clin Invest 1997;100:
259–63.
28. Uitterlinden AG, Burger H, van Duijn CM, Huang Q, Hofman A,
Birkenhäger JC, et al. Adjacent genes, for COL2A1 and the
vitamin D receptor, are associated with separate features of
radiographic osteoarthritis of the knee. Arthritis Rheum 2000;43:
1456–64.
29. Burger H, van Daele PL, Algra D, van den Ouweland FA,
Grobbee DE, Hofman A, et al. The association between age and
bone mineral density in men and women aged 55 years and over:
the Rotterdam Study. Bone Miner 1994;25:1–13.
30. Gunther KP, Sturmer T, Sauerland S, Zeissig I, Sun Y, Kessler S,
et al. Prevalence of generalised osteoarthritis in patients with
advanced hip and knee osteoarthritis: the Ulm Osteoarthritis
Study. Ann Rheum Dis 1998;57:717–23.
31. Felson DT. Epidemiology of hip and knee osteoarthritis. Epide-
miol Rev 1988;10:1–28.
32. Keefe FJ, Lefebvre JC, Egert JR, Affleck G, Sullivan MJ, Caldwell
DS. The relationship of gender to pain, pain behavior, and
disability in osteoarthritis patients: the role of catastrophizing.
Pain 2000;87:325–34.
33. Uitterlinden AG, Pols HA, Burger H, Huangn O, van Daele, van
Duijn, et al. A large-scale population-based study of the associa-
tion of vitamin D receptor gene polymorphisms with bone mineral
density. J Bone Miner Res 1996;11:1241–8.
34. Peltekova VD, Rubin L, Uitterlinden AG, Hawker G, Vieth R,
Trang H, et al. Direct haplotyping at the vitamin D receptor locus
improves genetic resolution. J Bone Miner Res 1997;12:494–5.
35. Langdahl BL, Lokke E, Carstens M, Stenkjaer LL, Eriksen EF. A
TA repeat polymorphism in the estrogen receptor gene is associ-
ated with osteoporotic fractures but polymorphisms in the first
exon and intron are not. J Bone Miner Res 2000;15:2222–30.
36. Spector TD, Cicuttini F, Baker J, Loughlin J, Hart D. Genetic
influences on osteoarthritis in women: a twin study. BMJ 1996;
312:940–3.
37. Hirsch R, Lethbridge-Cejku M, Hanson R, Scott WW Jr, Reichle
R, Plato CC, et al. Familial aggregation of osteoarthritis: data from
the Baltimore Longitudinal Study on Aging. Arthritis Rheum
1998;41:1227–32.
38. Dequeker J. The relationship between osteoporosis and osteoar-
thritis. Clin Rheum Dis 1985;11:271–96.
39. Dequeker J, Boonen S, Aerssens J, Westhovens R. Inverse rela-
tionship osteoarthritis-osteoporosis: what is the evidence? What
are the consequences? Br J Rheumatol 1996;35:813–8.
40. Burger H, van Daele PL, Odding E, Valkenburg HA, Hofman A,
Grobbee DG, et al. Association of radiographically evident osteo-
arthritis with higher bone mineral density and increased bone loss
with age: the Rotterdam Study. Arthritis Rheum 1996;39:81–6.
41. Ongphiphadhanakul B, Rajatanavin R, Chanprasertyothin S, Pia-
seu N, Chailurkit L, Sirisriro R, et al. Estrogen receptor gene
polymorphism is associated with bone mineral density in premeno-
pausal women but not in postmenopausal women. J Endocrinol
Invest 1998;21:487–93.
42. Claassen H, Hassenpflug J, Schunke M, Sierralta W, Thole H,
Kurz B. Immunohistochemical detection of estrogen receptor
alpha in articular chondrocytes from cows, pigs and humans: in situ
and in vitro results. Ann Anat 2001;183:223–7.
43. Spector TD, Hart DJ, Byrne J, Harris PA, Dacre JE, Doyle DV.
Definition of osteoarthritis of the knee for epidemiological studies.
Ann Rheum Dis 1993;52:790–4.
44. Felson DT, McAlindon TE, Anderson JJ, Naimark A, Weissman
BW, Aliabadi P, et al. Defining radiographic osteoarthritis for the
whole knee. Osteoarthritis Cartilage 1997;5:241–50.
45. Lanyon P, O’Reilly S, Jones A, Doherty M. Radiographic assess-
ment of symptomatic knee osteoarthritis in the community: defi-
nitions and normal joint space. Ann Rheum Dis 1998;57:595–601.
46. Weel AE, Uitterlinden AG, Westendorp IC, Burger H, Schuit SC,
Hofman A, et al. Estrogen receptor polymorphism predicts the
onset of natural and surgical menopause. J Clin Endocrinol Metab
1999;84:3146–50.
47. Maruyama H, Toji H, Harrington CR, Sasaki K, Izumi Y, Ohnuma
T, et al. Lack of an association of estrogen receptor alpha gene
polymorphisms and transcriptional activity with Alzheimer dis-
ease. Arch Neurol 2000;57:236–40.
48. Salmen T, Heikkinen AM, Mahonen A, Kroger H, Komulainen M,
Saarikoski S, et al. The protective effect of hormone-replacement
therapy on fracture risk is modulated by estrogen receptor alpha
genotype in early postmenopausal women. J Bone Miner Res
2000;15:2479–86.
49. Ernst M, Parker MG, Rodan GA. Functional estrogen receptors in
osteoblastic cells demonstrated by transfection with a reporter
gene containing an estrogen response element. Mol Endocrinol
1991;5:1597–606.
50. Ernst M, Rodan GA. Estradiol regulation of insulin-like growth
factor-I expression in osteoblastic cells: evidence for transcrip-
tional control. Mol Endocrinol 1991;5:1081–9.
51. Denko CW, Boja B, Moskowitz RW. Growth promoting peptides
in osteoarthritis: insulin, insulin-like growth factor-1, growth hor-
mone. J Rheumatol 1990;17:1217–21.
52. Fernihough JK, Richmond RS, Carlson CS, Cherpes T, Holly JM,
Loeser RF. Estrogen replacement therapy modulation of the
insulin-like growth factor system in monkey knee joints. Arthritis
Rheum 1999;42:2103–11.
1922 BERGINK ET AL