Trisomic pregnancy and elevated FSH - Semantic Scholar

Human Reproduction, Vol.26, No.6 pp. 1537– 1550, 2011 Advanced Access publication on April 5, 2011 doi:10.1093/humrep/der091

ORIGINAL ARTICLE Reproductive epidemiology

Trisomic pregnancy and elevated FSH: implications for the oocyte pool hypothesis J.K. Kline 1,2,3,*, A.M. Kinney 2,4, B. Levin 2, A.C. Kelly 5, M. Ferin 5, and D. Warburton 6 1

Imprints Center, New York State Psychiatric Institute, New York, NY 10032, USA 2Joseph L. Mailman School of Public Health, Columbia University, New York, NY 10032, USA 3Gertrude H. Sergievsky Center, Columbia University, New York, NY 10032, USA 4Graduate School of Arts and Sciences, Columbia University, New York, NY 10032, USA 5Department of Obstetrics and Gynecology, Columbia University, New York, NY 10032, USA 6Department of Genetics and Development, Columbia University, New York, NY 10032, USA *Correspondence address. Epidemiology, 722 West 168th Street, Room 1607, New York, NY 10032, USA. Tel: +1-212-305-9110; Fax: +1-212-305-4653; E-mail: [email protected]

Submitted on December 9, 2010; resubmitted on February 23, 2011; accepted on March 3, 2011

background: Some studies, but not all, support the hypothesis that trisomy frequency is related to the size of the oocyte pool, with the risk increased for women with fewer oocytes (older ovarian age). We tested this hypothesis by comparing hormonal indicators of ovarian age among women who had trisomic pregnancy losses with indicators among women with non-trisomic losses or chromosomally normal births. The three primary indicators of advanced ovarian age were low level of anti-Mu¨llerian hormone (AMH), high level of follicle-stimulating hormone (FSH) and low level of inhibin B. methods: The analysis drew on data from two hospital-based case –control studies. Data were analyzed separately and the evidence from the two sites was combined. We compared 159 women with trisomic pregnancy losses to three comparison groups: 60 women with other chromosomally abnormal losses, 79 women with chromosomally normal losses and 344 women with live births (LBs) agematched to women with losses. We analyzed the hormone measures as continuous and as categorical variables. All analyses adjust for age in single years, day of blood draw, interval in storage and site. results: AMH and inhibin B did not differ between women with trisomic losses and any of the three comparison groups. Mean ln(FSH) was 0.137 units (95% confidence interval (CI): 0.055, 0.219) higher for trisomy cases compared with LB controls; it was also higher, though not significantly so, for trisomy cases compared with women with other chromosomally abnormal losses or chromosomally normal losses. The adjusted odds ratio in relation to high FSH (≥10 mIU/ml) was significantly increased for trisomy cases versus LB controls (adjusted odds ratio (OR): 3.8, 95% CI: 1.6, 8.9).

conclusions: The association of trisomy with elevated FSH is compatible with the oocyte pool hypothesis, whereas the absence of an association with AMH is not. Alternative interpretations are considered, including the possibility that elevated FSH may disrupt meiotic processes or allow recruitment of abnormal follicles. Key words: epidemiology / aneuploidy / FSH / Mu¨llerian inhibiting substance

Introduction Some studies, but not all, support the hypothesis (Warburton, 1989; Kline and Levin, 1992) that trisomy frequency is related to the size of the oocyte pool, with the risk increased for women with fewer oocytes (older ovarian age). We tested this hypothesis by comparing three hormonal indicators of ovarian age among women who had trisomic pregnancy losses with those among women with non-trisomic losses or chromosomally normal births.

The risk of trisomy increases with maternal age, with a dramatic rise after the mid-30s. The oocyte pool, which is formed during foetal development, is largest around 4– 5 months of gestation (Baker, 1963) and decreases with chronologic age, either as an exponential function (Thomford et al., 1987; Kline and Levin, 1992) or as a power function (Hansen et al., 2008). Despite this overall decline, oocyte count varies among women of the same chronologic age. We hypothesize that at any given chronologic age, the trisomy risk is related to the size of the oocyte pool.

& The Author 2011. Published by Oxford University Press on behalf of the European Society of Human Reproduction and Embryology. All rights reserved. For Permissions, please email: [email protected]

1538 Potential mechanisms linking the size of the oocyte pool to trisomy include changes in intra- or extra-follicular ovarian factors. We examined three hormonal indicators of ovarian age: anti-Mu¨llerian hormone (AMH), follicular-stimulating hormone (FSH) and inhibin B. AMH, which is expressed by the granulosa cells, is detectable in some primary and secondary follicles (2– 6 granulosa cell layers) and in essentially all pre-antral and small (,6 mm) antral follicles (Weenen et al., 2004; Stubbs et al., 2005). In a sample of 42 ovaries obtained by oophorectomy (Hansen et al., 2010), the age-adjusted correlation between serum AMH and ln(number of primordial follicles) was 0.48, supporting the view that AMH reflects the size of the oocyte pool. FSH, a gonadotrophin under negative feedback of two ovarian hormones, inhibin B and estradiol (Burger, 1994, 2000), reflects the quantity or quality of the antral follicles; it may also provide an indirect measure of characteristics of the underlying oocyte pool (Goldenberg et al., 1973; Klein et al., 2002). Inhibin B, produced by the granulosa cells of the antral follicles (Groome et al., 1996; Welt et al., 1999; Burger, 2000; Luisi et al., 2005), reflects the quality and, perhaps, the quantity of antral follicles, which reflect the size of the underlying oocyte pool. We also examined associations with estradiol because some reports (Klein et al., 1996, 2000) suggest that, among older regularly cycling women, the dominant follicle emerges early, with a consequent early rise in estradiol. Several observations support the idea that the size of the oocyte pool influences the trisomy risk. (i) One early experiment (Brook et al., 1984) showed higher rates of aneuploidy among mice with unilateral oophorectomies than among sham-operated mice. In women, however, observations relating oophorectomy to trisomy are inconsistent (Freeman et al., 2000; Warburton and Kline, 2001); inferences are further limited by the small number of women studied. (ii) Age at menopause is 1 year earlier among women with trisomic spontaneous abortions (SAs) than among women with chromosomally normal pregnancies (Kline et al., 2000). Since menopause occurs when the oocyte pool falls below a critical threshold, this observation is consistent with the oocyte pool hypothesis. (iii) Elevated FSH is more common in women after a Down syndrome birth than in controls (van Montfrans et al., 1999, 2001). (iv) In samples consisting mainly of women pregnant by assisted reproduction, the evidence is mixed: one study (Nasseri et al., 1999) suggests a possible association of elevated FSH with aneuploidy, the majority trisomic, among pregnancy losses, whereas another (Massie et al., 2008) does not. A third study (Haadsma et al., 2009) shows fewer retrieved oocytes, a possible indicator of diminished ovarian reserve, among women with trisomic pregnancies. (We exclude from consideration studies that used fluorescent in situ hybridization (FISH) to assess aneuploidy in blastomeres because recent results using genomic microarray (Treff et al., 2010) suggest that, in this circumstance, FISH does not provide accurate results.) In an earlier sample (the New York study), we assessed ovarian age with three indicators: antral follicle count, FSH and inhibin B. The three indicators, defined continuously, did not differ between women with trisomic losses and women of the same age with chromosomally normal births or women with non-trisomic losses, failing to confirm the hypothesis. On the other hand, when we dichotomized each measure to define ‘old’ ovarian age, trisomy appeared to be positively associated with FSH ≥10 mIU/ml and inhibin B ≤20 pg/ml (the limit

Kline et al.

of detection of the assay), although the confidence intervals for both associations included unity (Kline et al., 2004). The present paper draws on a second, larger sample (the New Jersey study) and the New York sample to compare hormonal indicators of ovarian age between women who had trisomic pregnancy losses and women who had other losses (non-trisomic chromosomally abnormal, chromosomally normal) or who had chromosomally normal births.

Materials and Methods The design and protocol for the New Jersey and New York studies are similar. The New Jersey study, described in full by Warburton et al. (2009), was designed to examine the relation of highly skewed X chromosome inactivation to trisomy. We saved sera in anticipation of the current analyses. We describe the protocol related to the hormone component in full here. The New York study (Kline et al., 2004), designed to test the oocyte pool hypothesis, is described briefly here. For the New York sample, this paper adds the measure of AMH and a new measure of inhibin B.

The New Jersey study the protocol From 25 February 2003 to 18 November 2005, we identified women age 18 or older with singleton SAs of developmental age ,18 weeks whose products of conception were submitted to the Pathology Department of a hospital in New Jersey. We asked permission to karyotype the abortus. If a woman’s abortus was successfully karyotyped, we asked her to: (i) complete a short telephone interview to determine whether or not she was eligible for hormone measures in addition to measures of X chromosome inactivation; (ii) complete a more extensive telephone interview regarding demographic characteristics, obstetric and medical histories and common exposures; (iii) make one visit to the hospital for a blood draw and a brief update interview about recent exposures. Women who were eligible for hormone measures recorded their menstrual periods and had blood drawn on Days 2 –4 of the second or later menstrual period observed during follow-up. For each woman with a SA who completed the study, we selected a woman with a chromosomally normal live birth (LB) ≥1800 g, without a major anatomic malformation, at the study hospital 6 – 12 months prior to her selection. The purpose of the LB control group was to provide expected levels of hormones among women at the same chronologic age as women with SAs. LBs were matched to SAs for projected age (+6 months) at the blood draw. The LB served to identify a fertile sample; delivery at the study hospital served to ensure demographic comparability. Eligible LBs had no SA after the index pregnancy and no known prior chromosomally abnormal pregnancy. Candidates were selected from a roster of women who delivered between 1 April 2003 and 31 May 2006 and consented to contact about the research. Hospital staff asked 6505 women for permission to list them as candidates; 5346 (82%) agreed. If an LB did not complete the study, we replaced her to obtain a comparison group as similar as possible in age to SAs who completed the study. If an SA was eligible for hormone measures, but her matched LB was not, we used the same procedure to select and enroll a second LB who was eligible for hormone measures. Control recruitment began after SA recruitment because (i) we required that an SA complete the protocol before we selected her matched LB (so that we could match for age at blood draw) and (ii) if an SA was eligible for hormone measures, we required that the first selected LB complete her intake interview so that we could determine whether or not a second

1539

Trisomic pregnancy and elevated FSH

LB was needed. The protocol for LBs was identical to the protocol for SAs. The interviewer knew the outcome of the index pregnancy, but she did not know the karyotype of the SA (except in the ,4% of instances when a participant revealed it). Hormones were measured without knowledge of any participant characteristics, including the outcome of the index pregnancy. To obtain valid ovarian age measures, we required: no pituitary or hormonal disorder related to ovarian function, no oophorectomy, no hormonal medication, no pregnancy at the time of blood draw and no breastfeeding or breastfeeding no more than once per day during the menstrual cycle preceding the blood draw. The study reproductive endocrinologist (A.C.K.) reviewed the interview data to determine whether or not a participant currently had a condition associated with altered hormone levels. We required that samples for hormone measures be collected no later than the menstrual cycle that began 6 months after the SA or, for LBs, after mailing the letter requesting participation. Field work ended in January 2007. The study was approved by the Institutional Review Boards at our university and the study hospital. All participants gave informed consent.

Comparison of trisomy cases and controls For this analysis, women with trisomic SAs constitute the case group (Table I). They are compared with LB controls and two SA control groups: women with non-trisomic chromosomally abnormal losses and women with chromosomally normal losses. As expected, trisomy cases are older than the other SAs and the LBs. Adjusting for age, the mean number of pregnancies ending in LB is higher and the mean number of SAs is lower among LBs than among trisomy cases or other SAs. In addition, the three SA groups and LBs differ in the mean number of induced abortions, although no two-group comparison was significant at a ¼ 0.05. The three SA groups and LBs do not differ in education or ethnicity. Three technical variables (day of blood draw, number of menstrual periods between index event and blood draw, interval between blood draw and assay) vary significantly between the three SA groups and LBs (Table I, see footnotes). Day of blood draw and storage interval are also related to one or more hormones: inhibin B is higher on Days 3–4 than on Day 2; FSH decreases and estradiol increases with increasing storage interval. We therefore adjust for ‘day’ and ‘storage’ in the analyses.

The New York study protocol and sample Women with SAs We identified 855 SAs. Karyotype results were obtained on 498/517 (96%) specimens in which chromosome analysis was attempted, by either culture or FISH (Supplementary data, Table SI). Among the 498 women with karyotyped losses, 211 (42%) provided samples for hormone measures. The principal reasons for not completing the hormone protocol were: refusal or withdrawal (22%) or ineligibility (34%), primarily due to pregnancy soon after the index SA (13% of the 498). Women who completed the hormone component did not differ from those who did not in mean age. Among the 424 women who completed the first interview (which determined eligibility for hormone measures), adjusting for age, the odds of completing the hormone protocol did not differ with the number of prior births, number of pregnancy losses or ethnicity; the odds of completing the protocol were greater for women with prior induced abortions and for women with more education. Analyses exclude (i) women for whom we consider the karyotype of the abortus uncertain (Supplementary data, Table SI, footnote c) and (ii) repeat study pregnancies (to maintain independence of observations). For the four women with a trisomic SA and another pregnancy outcome, we retained the trisomic SA.

LB controls LBs were sampled with replacement. In total, we selected 892 LBs, of which 279 (31%) completed the hormone component (Supplementary data, Table SI). (The number of LBs exceeds the number of SAs primarily because 61 LBs who completed the hormone component were matched to SAs who were not eligible for hormone measures.) The principal reasons for not completing the hormone protocol were: refusal or withdrawal (13%) or ineligibility (51%), primarily due to use of hormonal contraceptives or breastfeeding (21 and 15% of the 892, respectively). Women who completed the hormone component were, on average, 1.1 years older than women who did not. Among the 765 women who completed the first interview, adjusting for age, the odds of completing the hormone protocol did not differ with education or number of LBs, SAs or induced abortions. The odds of completing were higher for nonHispanic white women than for other women.

The New York study (September 1998 – April 2001) was similar in design to the New Jersey study. It differed in the following ways: (i) it included women with singleton pre-fetal losses (developmental age ,9 weeks rather than 18 weeks); (ii) women with births were selected for trisomy cases only (rather than for all women with losses); (iii) we drew blood on Days 1 – 4 (rather than Days 2 – 4). The analytic sample includes 54 women with trisomic SAs, 24 with nontrisomic chromosomally abnormal SAs, 22 with chromosomally normal SAs and 65 with LBs (see Kline et al. (2004) for a detailed description). This report draws on assays for FSH and estradiol carried out in February of 2002 and assays for AMH and inhibin B carried out in February –March 2008. Inhibin B was decreased in samples drawn on Day 1 and estradiol was increased in samples drawn on Days 3 – 4; interval in storage was not related to any hormone.

Hormonal indicators of ovarian age Blood samples were processed in a refrigerated centrifuge and, after separation, sera were frozen at 2208C (New Jersey) or 2258C (New York) at the study hospital. They were delivered to New York City and stored at – 208C. AMH and inhibin B were measured by enzyme-linked immunosorbent assays (Diagnostics Systems Laboratories, Inc., Webster, TX, USA). FSH and estradiol were measured by solid-phase chemiluminescent enzyme immunoassays (Immulite; Diagnostic Products Co, Los Angeles, CA, USA). For AMH, sensitivity was 0.05 ng/ml; intra- and inter-assay coefficients of variation were 2.3 and 8.9%, respectively. For inhibin B, sensitivity was 15 pg/ml; intra- and inter-assay coefficients of variation were 3.9 and 6.1%, respectively. For FSH, sensitivity was 0.1 mIU/ml; intraand inter-assay coefficients of variation were 1.9 and 5.0%, respectively. For estradiol, sensitivity was 20 pg/ml; intra- and inter-assay coefficients of variation were 9.3 and 10.5%, respectively. Table II shows summary statistics for hormone levels and their correlations with chronological age in the two studies. Chronological age was related to AMH and FSH in the expected directions, but unrelated to inhibin B in the New Jersey sample, contrary to our expectation when we conceived this analysis (Kline et al., 2005); inhibin B was modestly associated with age in the New York sample. For estradiol, the modest positive correlation is compatible with observations that levels may be elevated during the menopausal transition (Klein et al., 1996, 2000).

1540

Kline et al.

Table I Selected characteristics of women who completed the protocol classified by the outcome of the index pregnancy: the New Jersey sample. Losses

................................................................................. Trisomy

Non-trisomy abnormal

Chromosomally normal

Live births

............................................................................................................................................................................................. Number of women

36

57

Age (years) at blood drawa, Mean (SD)

105 36.9 (4.7)

34.2 (3.8)

33.1 (4.8)

279

Length (days) of gestationb, Mean (SD)

63.7 (11.2)

67.0 (12.1)

67.3 (21.3)

Live births

0.9 (0.9)

1.0 (0.9)

0.9 (0.9)

1.8 (0.8)

Spontaneous abortions

1.4 (0.7)

1.6 (1.0)

1.5 (0.8)

0.3 (0.7)

Induced abortions

0.4 (0.7)

0.1 (0.3)

0.3 (0.6)

0.2 (0.6)

35.5 (4.7) dna

Pregnancies at blood drawc, Mean (SD)

d

Education (%) No college degree

27.6

16.7

25.0

27.6

College degree

43.8

58.3

48.2

48.7

Postgraduate degree

28.6

25.0

26.8

23.7

86.7

80.6

78.9

89.2

Day 2

36.2

16.7

36.8

26.2

Day 3

26.7

50.0

26.3

42.3

Day 4

37.1

33.3

36.8

31.5

2

55.8

54.3

59.6

3.6

3

29.5

25.7

26.3

5.1

4– 16

14.7

20.0

14.0

91.3

e

Ethnicity (%) White, non-Hispanic Day of blood drawf (%)

Number of menstrual periods between the index pregnancy and the blood drawg (%)

Interval (years) between blood draw and assayh [Mean (SD)]

4.3 (0.8)

4.3 (0.8)

4.1 (0.9)

3.4 (0.9)

a

Mean age varies (P , 0.0001) with the outcome of the index pregnancy. As expected, women with trisomy losses are significantly older than women with non-trisomy losses. Because live birth controls were age-matched to women with losses, women with chromosomally normal births are significantly older than women with chromosomally normal losses. b Mean gestation does not vary significantly among the three spontaneous abortion groups. c Adjusted for age, the mean number of live births (P , 0.0001), mean number of spontaneous abortions ,20 weeks gestation (P , 0.0001) and number of induced abortions (1+ versus 0) (P ¼ 0.02) vary with the outcome of the index pregnancy. d Adjusted for age, education does not vary significantly with the outcome of the index pregnancy. e Adjusted for age, ethnicity does not vary significantly with the outcome of the index pregnancy. f Day of the blood draw varies (P ¼ 0.02) with the outcome of the index pregnancy. g The number of menstrual periods varies (P , 0.0001) with the outcome of the index pregnancy. As expected, women with births had more menstrual periods between the index pregnancy and the blood draw than women with losses. The analysis excludes 14 women with unknown number of menstrual periods. h The interval between the blood draw and the assay varies (P , 0.0001) with the outcome of the index pregnancy. Because recruitment of women with births lagged, necessarily, behind identification of women with losses, the interval is shorter for women with births.

The statistical analysis Primary analysis We first examined associations with each of the ovarian age indicators separately. We used age-stratified multiple linear regression models (with indicator variables for age in single years) to test the null hypothesis that, at any maternal age, there is no difference between trisomy cases and any of the three comparison groups for any of the hormone measures. All analyses adjusted also for day of blood draw (an indicator variable) or duration of time in storage (continuous). We estimated the magnitude of differences between trisomy cases and each comparison group and set 95% confidence intervals around the estimates. We used a logarithmic transformation for each hormone to meet the normal error assumption of least squares regression. Thus, differences correspond approximately

to percentage changes in the hormone level. The estimated percentage shift in hormone level is equal to 1 – exp(difference). We checked the results with a complementary analysis using conditional (matched-sample) logistic regression (Breslow and Day, 1980; Levin, 1987, 1990). All analyses adjusted by stratification for age (single years) and by model parameters for day of blood draw and duration of time in storage. For these analyses, we analyzed hormone levels categorically. For AMH and inhibin B, we dichotomized as close as possible to the fifth percentile among LBs to define older ovarian age (≤0.186 ng/ml for AMH, ≤15 pg/ml for inhibin B). For FSH, we dichotomized at ≥10 mIU/ml, which corresponds approximately to the top five per cent for LBs. A level of 10 mIU from the Immulite assay has been used to define diminished ovarian reserve in several studies of pregnancy loss and aneuploidy (Nasseri et al., 1999; Trout and Seifer, 2000; Gurbuz

1541


Table II Summary statistics and correlations with chronological age for AMH, FSH, inhibin B and estradiol for (1) 477 women in the New Jersey sample and (2) 165 women in the New York sample. AMH (ng/ml)

FSH (mIU/ml)

Inhibin B (pg/ml)

Estradiol (pg/ml)

............................................................................................................................................................................................. (1) New Jersey sample Mean (SD)

1.848 (2.489)

6.0 (2.6)

39.3 (20.7)

46 (32)

Geometric meana

0.984

5.5

35.1

41

Median

0.970

5.7

35.8

41

Rangeb

0.050–15.000

0.8 –22.6

15.0–238.6

20–391

Percent above or below assay detection level

0.8

NA

4.6

11.1

Correlationc with chronological age

20.41

0.25

0.005

0.19

Correlationc with chronological age among women with live births

20.41

0.19

20.002

0.20

(2) New York sample Mean (SD)

2.147 (2.188)

4.9 (2.6)

38.2 (19.2)

40 (24)

Geometric meana

1.220

4.4

34.2

37

Median

1.520

4.4

35.0

35

Range

0.090–11.740

0.7 –16.3

15.0–124.0

20–206

Percent below assay detection level

0.0

NA

9.1

6.1

Correlationc with chronological age

20.52

0.37

20.19

0.13

Correlationc with chronological age among women with live births

20.39

0.17

20.15

0.22

a

To meet the normality assumptions of ordinary least squares regression, we used natural logarithmic transformations for each hormone. Values are reported in the original scale using the inverse transformation, i.e. exp(mean). For AMH, values ,0.05, the limit of detection, were analyzed as 0.05; values .15.000 were analyzed as 15.000. For inhibin B, values ,15.000, the limit of detection, were analyzed as 15.000. c We report Pearson correlation coefficients of age with ln(hormone level). b

et al., 2004). For estradiol, where either high or low levels might indicate older ovarian age, we used three categories: ≤20 pg/ml, the limit of detection (corresponding approximately to the 11th percentile), ≥85 pg/ml (the upper fifth percentile) and, as the reference group, levels between 20 and 85 pg/ml. To address the possibility that associations with trisomy might be attenuated by the presence of women with prior unkaryotyped trisomies (an estimated 7 – 9% of LB controls), we repeated the primary least squares regression analyses using only the largest comparison group, LB controls, excluding from this group women with prior losses. We carried out all analyses for each of the two samples separately. In the least squares regressions, we combined the evidence from the two sites, weighting the estimated regression coefficients by the inverse of their squared standard errors. In the logistic regressions, we combined the evidence in an analysis that stratified by site as well as age. Our analysis of the New York study differs slightly from the analysis reported previously. In our previous report, we retained the trisomy – LB matches and allocated the two control SA groups to the closest age stratum. Here, for comparability with the analysis of the New Jersey study, we adjust for single year of age rather than maintaining the matches. For FSH and estradiol, the two hormones previously analyzed, results are essentially the same.

Secondary analyses Drawing on data from both studies, adjusting for site, age, day of blood draw and time in storage, we explored whether the strength of associations between trisomy and ovarian age indicators vary with chronologic age (,35 years, 35+ years) or with trisomy type (trisomy 16, other nonacrocentric trisomies, acrocentric trisomies, double and triple trisomies). We do not include the double and triple trisomies in the regression

analyses because of their small number. We analyzed the hormones as both continuous and categorical variables. In these secondary analyses, we draw on the LB control group only because the two pregnancy loss comparison groups are too small to be informative. We do not report data on variations with chronologic age because none of the associations varied significantly with age.

Results Table III shows mean hormone levels for trisomy cases and the three comparison groups. Table IV shows adjusted differences in mean ln(hormone) between trisomy cases and each comparison group for each sample and for the samples combined. Table V shows the results of the complementary categorical analyses comparing the odds of hormone levels indicative of old ovarian age for trisomy cases and LB controls. Table VI shows adjusted mean ln(hormone) and the proportion with hormone levels indicative of old ovarian age for each trisomy type. Table VII shows differences in mean ln(hormone) and odds of hormone levels indicative of old ovarian age for each trisomy type.

AMH AMH, analyzed as a continuous variable (Tables III and IV), did not differ between trisomy cases and any of the comparison groups for either sample or for the samples combined. Results were similar when the LB control group was limited to women without prior pregnancy losses (not shown). In the categorical analysis (Table V), the

1542

Kline et al.

Table III Observed means for AMH, FSH, inhibin B and estradiol among women with trisomic losses (cases) compared with women with non-trisomy chromosomally abnormal losses, with chromosomally normal losses and with live births in (1) the New Jersey sample and (2) the New York sample. Losses

..................................................................................................... Trisomy

Non-trisomy abnormal

Live births

Chromosomally normal

............................................................................................................................................................................................. (1) New Jersey sample Number of women

105

36

57

279

Mean ln(hormone) (SD)a AMH (ng/ml)

20.263 (1.042)

20.255 (1.084)

0.181 (1.108)

0.067 (1.150)

FSH (mIU/ml)

1.809 (0.428)

1.655 (0.352)

1.642 (0.368)

1.693 (0.403)

Inhibin B (pg/ml)

3.561 (0.481)

3.539 (0.527)

3.562 (0.467)

3.559 (0.457)

Estradiol (pg/ml)

3.777 (0.469)

3.876 (0.300)

3.721 (0.402)

3.659 (0.500) 1.070

Geometric mean

b

AMH (ng/ml)

0.769

0.775

1.198

FSH (mIU/ml)

6.1

5.2

5.2

5.4

Inhibin B (pg/ml)

35.2

34.4

35.2

35.1

Estradiol (pg/ml)

44

48

41

39

54

24

22

65

(2) New York sample Number of women Mean ln(hormone) (SD)a AMH (ng/ml)

0.176 (1.250)

0.524 (1.137)

0.783 (1.063)

20.100 (1.033)

FSH (mIU/ml)

1.476 (0.559)

1.441 (0.433)

1.371 (0.524)

1.535 (0.366)

Inhibin B (pg/ml)

3.460 (0.497)

3.522 (0.503)

3.684 (0.428)

3.545 (0.447)

Estradiol (pg/ml)

3.565 (0.412)

3.444 (0.345)

3.745 (0.372)

3.645 (0.451) 0.905

Geometric mean

b

AMH (ng/ml)

1.193

1.689

2.189

FSH (mIU/ml)

4.4

4.2

3.9

4.6

Inhibin B (pg/ml)

31.8

33.8

39.8

34.6

Estradiol (pg/ml)

35

31

42

38

a

To meet the normality assumption of least squares regression, we used natural logarithmic transformations for each hormone. Values are reported in the original scale using the inverse transformation, i.e. exp(mean).

b

adjusted odds ratio relating low AMH (≤0.186 ng/ml) to trisomy versus LB controls was 1.9 (95% confidence interval (CI): 0.8, 4.2). AMH and the proportion with low AMH did not differ between any of the three single trisomy groups and the LB controls (Tables VI and VII).

Follicle-stimulating hormone FSH, analyzed as a continuous variable (Tables III and IV), differed between trisomy cases and each of the three comparison groups for the New Jersey sample but not for the New York sample. The association of FSH with trisomy in comparison with LB controls varied with site (P ¼ 0.01). In the New Jersey sample, on average, mean ln(FSH) was 0.202 higher for trisomy cases than for LB controls, with the 95% confidence interval excluding zero. Mean ln(FSH) was also higher for trisomy cases compared with other SAs, but not significantly so. In the New Jersey sample, the adjusted difference was greater than the unadjusted difference. The primary reason for the increase

was adjustment for storage interval. Storage intervals were longer for women with SAs and associated with decreased FSH. For example, for samples in storage 4.3 – ,5.9 years (the longest 1.5-year interval), the unadjusted difference in mean ln(FSH) between trisomy cases (n ¼ 58) and LB controls (n ¼ 62) was 0.19. For storage 2.8– ,4.3 years, the difference was 0.18 (44 trisomies, 125 LBs). For storage 1.3– ,2.8 years, the difference was 0.75 (three trisomies, 92 LBs). Combining evidence from the two samples, we estimated that, on average, mean ln(FSH) was 0.137 units higher for trisomy cases than for LB controls, 0.102 higher than for non-trisomic chromosomally abnormal SA controls and 0.063 higher than for chromosomally normal SA controls. The point estimate for the comparison with LB controls corresponded to the prediction that, at the median FSH value for LBs from the two samples (5.2 mIU/ml), FSH levels for trisomy cases are 0.8 mIU/ml higher than FSH levels for LB controls. Results were similar when the LB control group was limited to women without pregnancy losses (mean difference in ln(FSH) ¼ 0.173, not shown).

1543


Table IV Adjusted mean differences for AMH, FSH, inhibin B and estradiol among women with trisomic losses (cases) compared with women with non-trisomy chromosomally abnormal losses, with chromosomally normal losses and with live births: (1) the New Jersey sample, (2) the New York sample, (3) the samples combined. Adjusted differencea (95% CI) in mean ln(hormone) between trisomy cases and each comparison group

........................................................................................................................................................ Non-trisomy abnormal losses

Chromosomally normal losses

Live births

............................................................................................................................................................................................. (1) New Jersey sample AMH (ng/ml)

0.309 (20.096, 0.714)

20.092 (20.439, 0.256)

FSH (mIU/ml)

0.101 (20.048, 0.251)

0.122 (20.007, 0.250)

20.110 (20.368, 0.147)

Inhibin B (pg/ml)

0.085 (20.094, 0.263)

20.006 (20.159, 0.147)

0.068 (20.046, 0.182)

Estradiol (pg/ml)

20.108 (20.275, 0.059)

20.051 (20.195, 0.093)

20.084 (20.191, 0.023)

AMH (ng/ml)

20.069 (20.559, 0.421)

20.353 (20.867, 0.160)

0.288 (20.070, 0.646)

FSH (mIU/ml)

20.019 (20.237, 0.200)

0.039 (20.190, 0.268)

20.043 (20.202, 0.117)

Inhibin B (pg/ml)

0.077 (20.152, 0.305)

20.162 (20.401, 0.078)

20.030 (20.197, 0.137)

Estradiol (pg/ml)

0.132 (20.073, 0.338)

20.205 (20.420, 0.011)

20.057 (20.207, 0.093)

AMH (ng/ml)

0.156 (20.156, 0.468)

20.174 (20.462, 0.114)

FSH (mIU/ml)

0.063 (20.060, 0.187)

0.102 (20.010, 0.214)

0.202 (0.106, 0.297) P ¼ 0.0004

(2) New York sample

(3) Samples combinedb 0.025 (20.184, 0.234) 0.137 (0.055, 0.219) P ¼ 0.001

Inhibin B (pg/ml)

0.082 (20.059, 0.222)

20.051 (20.180, 0.078)

0.037 (20.057, 0.131)

Estradiol (pg/ml)

20.012 (20.142, 0.118)

20.098 (20.218, 0.021)

20.075 (20.162, 0.012)

a All analyses adjust for age in single years (with indicator variables), day of blood draw and interval between blood draw and assay (continuous). For the New Jersey sample, day of blood draw is defined as Day 2 (reference), Days 3 –4. For the New York sample, day of blood draw is defined as Day 1, Day 2 (reference), Days 3 –4. For each hormone, differences between the trisomy case group and the comparison groups were obtained from a single analysis in which trisomy cases were the reference group. We reverse the sign of the point estimates to report the difference for each hormone between the mean for the trisomy cases and the corresponding mean for each comparison group. b The adjusted differences for the sites combined are a weighted average of the differences, weighting the estimated regression coefficients by the inverse of their squared standard errors.

The proportion of women with FSH ≥10 mIU/ml was increased among trisomy cases in both samples (Table V): evidence from the sites combined showed an adjusted odds ratio (OR) of 3.8 (95% CI: 1.6, 8.9), with no significant difference between sites. Classification of trisomies by type (Tables VI and VII) showed higher mean ln(FSH) for acrocentric trisomies and non-acrocentric trisomies other than trisomy 16. The categorical analysis was consistent with this result; there were no trisomy 16 cases with high FSH.

Inhibin B Inhibin B, whether analyzed continuously (Tables III and IV) or categorically (Table IV), did not differ between trisomy cases and any of the comparison groups at either site or at the sites combined. Results were similar when the LB control group was limited to women without prior pregnancy losses. Classification of trisomies by type (Tables VI and VII) showed significantly higher inhibin B (younger ovarian age) among trisomy 16 cases than among LB controls. The categorical analysis was consistent with this result, although the 95% CI for the adjusted odds ratio included unity.

Estradiol Estradiol, analyzed continuously (Tables III and IV), did not differ between trisomy cases and any of the comparison groups at either site or at the sites combined. When the LB control group was limited to women without prior pregnancy losses, using evidence from the two sites combined, the magnitude of the difference

between trisomy cases and LB controls increased (difference: –0.108 units, 95% CI: –0.213, –0.004). In the complementary categorical analysis, the proportion of women with low estradiol (≤20 pg/ ml; Table V) was increased among trisomy cases compared with LB controls; in data from the two sites combined, the adjusted odds ratio was 3.2 (95% CI: 1.4, 7.3). High estradiol (≥85 pg/ml) is unrelated to trisomy. Classification of trisomies by type showed significantly lower estradiol in acrocentric trisomy cases compared with LB controls (Tables VI and VII). The categorical analysis was consistent with this result.

Discussion The oocyte pool hypothesis led us to predict that women with trisomic pregnancies have lower AMH, higher FSH and lower inhibin B than expected, given their chronologic age. Our data provide only partial support for this prediction, leading us to suggest two alternative hypotheses, one related to the quality of the antral follicle cohort or the dominant follicle that emerges from the cohort and the other related to a possible direct effect of elevated FSH on meiotic error or the likelihood that an abnormal follicle is recruited. FSH, analyzed continuously, was significantly increased in trisomy cases compared with LB controls. Moreover, ln(FSH) was higher for trisomy cases than for women with losses of other types. While the latter associations are not statistically significant, they support the view that the association is specific to trisomy rather than a reflection

1544

Kline et al.

Table V Proportions with low AMH, high FSH, low inhibin B, low estradiol and high estradiol among women with trisomic losses (cases) and women with live births, and adjusted odds ratios relating trisomy to hormone levels: (1) the New Jersey sample, (2) the New York sample, (3) the samples combined. Trisomy cases (%)

Live birth controls (%)

Odds ratio (95% CI) adjusted for agea

Odds ratio (95% CI) adjusted for age, day of blood draw, storage intervalb

............................................................................................................................................................................................. (1) New Jersey sample (N)

105

279

Low AMH (≤0.186 ng/ml)

10.5

5.0

1.8 (0.8, 4.1)

1.6 (0.6, 4.0)

High FSH (≥10 mIU/ml)

12.4

6.1

2.0 (0.9, 4.5)

3.4 (1.3, 8.7) P ¼ 0.014

Low inhibin B (≤15 pg/ml)

3.8

5.7

0.6 (0.2, 1.9)

0.4 (0.1, 1.4)

Low estradiol (≤20 pg/ml)

9.5

12.9

0.8 (0.4, 1.8)

3.0 (1.2, 8.0) P ¼ 0.014

High estradiol (≥85 pg/ml)

6.7

5.4

1.2 (0.5, 3.1)

1.2 (0.4, 3.5)

54

65

11.1

4.6

3.2 (0.6, 18.0)

3.1 (0.5, 19.1)

(2) New York sample (N ) Low AMH (≤0.186 ng/ml) High FSH (≥10 mIU/ml)

9.3

3.1

3.6 (0.6, 22.3)

7.3 (0.5, 109.6)


18.5

6.2

3.8 (0.9, 16.0)

3.2 (0.7, 15.9)


7.4

4.6

1.9 (0.4, 9.5)

3.1 (0.5, 19.5)


5.6

4.6

2.3 (0.3, 15.4)

2.4 (0.3, 18.3)

Low AMH (≤0.186 ng/ml)

NA

NA

2.0 (0.97, 4.2)

1.9 (0.8, 4.2)


NA

NA

2.2 (1.1, 4.6) P ¼ 0.028

3.8 (1.6, 8.9) P ¼ 0.002


NA

NA

1.2 (0.6, 2.7)

0.9 (0.4, 2.1)

(3) Samples combinedc


NA

NA

1.0 (0.5, 1.9)

3.2 (1.4, 7.3) P ¼ 0.008


NA

NA

1.3 (0.6, 3.1)

1.3 (0.5, 3.3)

a

Adjusted odds ratios were obtained from a conditional maximum likelihood logistic regression that stratified by age. For the New Jersey sample, the age strata were 19/20, 21/22, 23, 24, 25/26, single years from 27 –43, 44/45. For the New York sample, the age strata were single years from 22 –25, single years from 27 –42, 43/44, 47/48. Analyses were adjusted by stratification for age, parametrically for day of blood draw (an indicator variable) and storage interval (continuous). Day of blood draw was defined as Days 3–4 (versus Days 1–2) because data were too sparse to adjust for Day 1. For analyses of the samples combined, the analysis was also adjusted for site by stratification. In the samples combined, day of blood draw is not significantly related to AMH, FSH or inhibin B; Days 3 –4 are positively associated with high estradiol. Storage interval is not significantly related to low AMH or low inhibin B; it is inversely related to high FSH and low estradiol. c We do not sum the samples because the ratio of trisomy cases to live birth controls differs between the sites. Results for AMH, FSH and estradiol did not vary significantly between study sites; results for inhibin B differ between sites (P ¼ 0.02). b

of non-specific differences between women with index SAs and women with index LBs. For FSH defined continuously, results differed between the sites, with only the New Jersey site showing increased FSH in trisomy cases compared with each of the three control groups. On the other hand, at both sites, the odds of FSH ≥10 mIU/ml were elevated among trisomy cases compared with LB controls (adjusted OR: 3.8). This juxtaposition of results from the continuous and categorical analyses suggests that the association is confined to FSH levels above a threshold. To explore this suggestion, we calculated predicted values of FSH using least squares regression for the two sites separately, in an urn model analysis (Levin and Robbins, 1983; Gail et al., 1988; Hatch et al., 1990) which adjusted for age, day of blood draw and storage interval, but not for karyotype. The average residuals (observed minus predicted FSH) for trisomy cases begin to increase around a predicted FSH level of 7 mIU/ml, consistent with an association above this threshold (Supplementary data, Figure S1). Neither AMH nor inhibin B, defined continuously or categorically, was significantly associated with trisomy. Our sample is sufficient (power ¼ 0.80, a ¼ 0.05, two-tailed) to detect moderate to small changes in these hormones between trisomy cases and LB controls.

For AMH, the detectable shift is 25.8%; for inhibin B, the detectable shift is 12.6%. Because the odds ratio relating trisomy to low AMH (≤0.186 ng/ ml) was elevated (adjusted OR: 1.9), albeit not significantly, we explored whether the association persisted with adjustment for FSH, defined categorically. It did not. The adjusted OR relating trisomy to low AMH was 1.1 (95% CI: 0.5, 2.8), indicating that low AMH does not have an independent association with trisomy. In contrast, the association of trisomy with high FSH was unchanged when we adjusted for low AMH (adjusted OR: 4.1, 95% CI: 1.5, 10.8). Inhibin B showed no association with chronologic age in the New Jersey study and only a weak association in the New York study, leading us to conclude that, at least in our samples, this hormone is neither an indicator of the size of the antral follicle cohort nor of the size of the primordial pool. Estradiol (a secondary indicator), defined continuously, was unrelated to trisomy. Our sample is sufficient to detect an 11.7% shift in estradiol between trisomy cases and LB controls. As an indicator of the size of the oocyte pool, we hypothesized that an association, if present, would be with elevated estradiol. We detected no association with high estradiol (≥85 pg/ml), but a significant association

1545


Table VI Observed means for AMH, FSH, inhibin B and estradiol and proportions with hormone levels indicating old ovarian age among women with trisomic losses, classified by trisomy type: (1) the New Jersey sample and (2) the New York sample. Trisomy 16

Other non-acrocentric trisomy

Acrocentric trisomy

Double or triple trisomy

............................................................................................................................................................................................. (1) New Jersey samplea Number of women

23

26

50

6

Mean ln(hormone) (SD) AMH (ng/ml)

0.335 (0.776)

20.546 (0.977)

20.218 (1.025)

21.703 (0.535)

FSH (mIU/ml)

1.698 (0.313)

1.806 (0.369)

1.777 (0.414)

2.518 (0.589)

Inhibin B (pg/ml)

3.670 (0.516)

3.591 (0.477)

3.549 (0.456)

3.111 (0.395)

Estradiol (pg/ml)

3.893 (0.356)

3.854 (0.601)

3.708 (0.439)

3.567 (0.374)

Per cent old ovarian age Low AMH (≤0.186 ng/ml)

0.0

11.5

8.0


0.0

7.7

12.0

66.7 83.3


4.4

0.0

4.0

16.7


4.4

7.7

14.0

0.0


0.0

19.2

4.0

0.0

(2) New York sampleb Number of women

18

14

20

2

Mean ln(hormone) (SD) AMH (ng/ml)

0.663 (0.974)

20.165 (1.309)

20.189 (1.261)

1.846 (0.252)

FSH (mIU/ml)

1.288 (0.514)

1.836 (0.585)

1.409 (0.509)

1.312 (0.302)

Inhibin B (pg/ml)

3.717 (0.368)

3.362 (0.579)

3.268 (0.455)

3.764 (0.467)

Estradiol (pg/ml)

3.711 (0.556)

3.515 (0.301)

3.481 (0.321)

3.433 (0.046)

Per cent old ovarian age Low AMH (≤0.186 ng/ml)

5.6

14.3

15.0

0.0


0.0

28.6

5.0

0.0


0.0

28.6

30.0

0.0


5.6

0.0

15.0

0.0


16.7

0.0

0.0

0.0

a Other non-acrocentric trisomy cases include one hypertetraploid (94,XXYY,+8,+8) and one trisomy plus inherited balanced translocation (46,XX,der(13;14)(q10;q10),+9). Acrocentric trisomy cases include one hypertriploid (70,XXY,+15) and five autosomal monosomies (three 45,XX, –21, two 45,XY, – 21). b Trisomy 16 cases include one hypertriploid (70,XXY,+16). Double trisomy cases include one hypertriploid mosaic (70,XXY,+6/71,XXY,+6,+20).

of trisomy with low estradiol (≤20 pg/ml; adjusted OR: 3.2). One possibility, particularly in light of the multiple tests performed, is that this is a chance finding. Moreover, a single measure of estradiol on Days 1–4 may not tap the phenomenon of interest, namely, the earlier emergence of the dominant follicle associated with greater chronologic age (Klein et al., 1996, 2000). We consider our analyses of trisomy type exploratory because of small sample sizes (Table VI). Our data (Table VII) suggest that trisomy 16 may not exhibit the same association with FSH as the other trisomy groups. This result is intriguing because trisomy 16 is unusual: it is more common than trisomies of other chromosomes and it increases linearly with maternal age, whereas most trisomies increase at a faster rate after the mid-30s than before (Hassold and Chiu, 1985; Risch et al., 1986). Strengths of our study include use of an unselected sample of known fertility. In the New Jersey sample, 42% of women with

karyotyped losses and 31% of potential LB controls completed the study. (See Kline et al. (2004) for a similar discussion of the New York sample.) Nevertheless, we think that our sample fairly represents the ovarian age of the population from which it was drawn. Among women who were ineligible for the hormone component (45% of women with losses, 62% of LB controls), the primary reasons for ineligibility were use of hormonal contraceptives, breastfeeding among women with births and pregnancy. We think it unlikely that hormonal contraception (common among our LB controls) or breastfeeding selectively exclude women of younger or older ovarian age. Exclusion of women who became pregnant shortly after their index pregnancy might select against especially fertile women or it may reflect childbearing patterns. We cannot distinguish between these phenomena. Most of the remaining criteria for ineligibility (e.g. use of fertility medications) limit generalizations to women with neither hormonal disorders nor the constellation of circumstances

1546

Kline et al.

Table VII Adjusted mean differences for AMH, FSH, inhibin B and estradiol and adjusted odds ratios relating hormonal indicators of old ovarian age to trisomy, classified by type, compared with live births: the New Jersey and New York samples combined. Trisomy 16

Other non-acrocentric trisomy

Acrocentric trisomy

............................................................................................................................................................................................. Adjusted differencea (95% CI) in mean ln(hormone) AMH (ng/ml)

0.295 (20.078, 0.668)

20.196 (20.557,0.164)

FSH (mIU/ml)

0.044 (20.098, 0.186)

0.184 (0.048, 0.320) P ¼ 0.008

Inhibin B (pg/ml)

0.185 (0.023, 0.347) P ¼ 0.026

0.035 (20.122, 0.192)

Estradiol (pg/ml)

0.095 (20.062, 0.252)

20.030 (20.183, 0.123)

20.025 (20.312, 0.263) 0.135 (0.027, 0.243) P ¼ 0.014 20.058 (20.184, 0.068) 20.176 (20.299, 20.053) P ¼ 0.005

Adjusted odds ratiob (95% CI) for old ovarian age Low AMH (≤0.186 ng/ml)

0.9 (0.1, 8.0)

1.8 (0.6, 5.6)

1.5 (0.5, 4.4)


NAc

3.4 (1.0, 11.2) P ¼ 0.048

3.1 (1.0, 9.4) P ¼ 0.041


0.3 (0.0, 2.7)

0.7 (0.2, 2.7)

1.4 (0.5, 4.0)


2.0 (0.4, 10.9)

1.8 (0.3, 9.1)

4.8 (1.7, 13.1) P ¼ 0.003


2.2 (0.5, 10.0)

2.7 (0.8, 9.1)

0.5 (0.1, 2.4)

a

Estimates obtained from least squares regression analysis. All analyses adjust for age in single years (with indicator variables), day of blood draw and interval between blood draw and assay (continuous). For the New Jersey sample, day of blood draw is defined as Day 2 (reference), Days 3 –4. For the New York sample, day of blood draw is defined Day 1, Day 2 (reference), Days 3–4. Estimates were obtained from a single stratified regression analysis in which each trisomy group was compared with live birth controls. The adjusted differences for the sites combined are a weighted average of the differences, weighting the estimated regression coefficients by the inverse of their squared standard errors. b Estimates were obtained from a single stratified logistic regression analysis. All analyses adjust for site and age (see Table V, footnote a) by stratification, day of blood draw and interval between blood draw and assay (continuous). Trisomy cases were compared with live birth controls. Day of blood draw is defined Days 1–2 (reference), Days 3–4. We were unable to adjust for Day 1 (New York) because there were too few samples. c There are no trisomy 16 cases with high FSH.

(not all medical) leading to prescription of fertility agents. Other strengths of our study include assessment of hormonal measures without the knowledge of characteristics of the woman, the ability to examine whether associations are specific to trisomic loss and the ability to control non-parametrically for chronologic age, the only consistent risk factor for trisomy. Limitations of our study include the following: first, not all our results, including those pertaining to FSH analyzed as a continuous variable, were consistent across the two samples. One possible explanation relates to chance, particularly given the small number of trisomy cases in the New York sample. Furthermore, there was similarity between the site-specific effects in the categorical analyses. We thus consider that analyses of the combined data yield the best estimates of effect size. Secondly, for the New Jersey sample, the length of time between the blood draw and the assay is inversely related to FSH and positively related to estradiol. The result for FSH is consistent with a previous report (Scriver et al., 2010). Adjustment for storage interval strengthened the association between FSH and trisomy (although the association was also apparent without adjustment). Adjusting for age and site, the mean ln(FSH) was 0.056 higher for trisomy cases compared with LB controls (analysis not shown); adjustment for storage interval and day of blood draw increased the difference to 0.137 (equivalent to a 2.5-fold increase in the shift in median FSH for trisomy cases versus LB controls). Similarly, in the categorical analysis, adjusting for age and site, the OR for high FSH in relation to trisomy was 2.2; adjustment for storage and day increased the OR to 3.8. We consider that the estimates from the fully adjusted models best summarize the strength of associations. Thirdly, we measured hormones after the index trisomy, not at conception. Among

women with pregnancy losses, on average, the interval between the last menstrual period (LMP) preceding conception and blood draw was short (158 days, SD 36, range 103–329); 80% had blood drawn within 6 months of their preconception LMP. Evidence on cycle-to-cycle variability stems mainly from studies of infertile women. It is unclear whether these can be generalized to fertile samples of reproductive age. No study of women unselected for infertility uses the interclass correlation coefficient to evaluate variability, hindering their utility. For follicular phase FSH, two studies (Hansen et al., 1996; Streuli et al., 2008) show similar means in two menstrual cycles; two others (Brown et al., 1995; Jain et al., 2003) indicate that variability between cycles may be greater in older women than younger. For AMH, one study (Streuli et al., 2008) shows similar means in two cycles; most studies (Hehenkamp et al., 2006) focus on intra-cycle variability, showing little variability within a single cycle. For inhibin B, one study (Jain et al., 2003) suggests that variability between cycles may be greater for older than younger women. There is no reason to think that variability (measurement error) is related to the study group. The effect of non-differential measurement error is to attenuate associations. Thus, the strength of the association with elevated FSH may be stronger than we estimate. For AMH and inhibin B, where we did not detect associations, estimates may also be attenuated. Our findings are consistent with some previous studies, but not others. The most relevant study (van Montfrans et al., 1999) compared women with Down syndrome births and controls. Mean ln(FSH), based on three cycles, was 9% higher for cases than controls (our estimate) compared with ≈14.6% in our data. The OR relating elevated FSH (defined by the upper fifth percentile in controls)


to Down syndrome birth was 3.0 (95% CI: 1.1, 8.6); in our sample, the OR relating elevated FSH ≥10 mIU/ml to trisomy SA is 3.8 (95% CI: 1.6, 8.9). As in our study, inhibin B and estradiol were unrelated to trisomy. Evidence on FSH and other indictors of ovarian age from samples consisting mainly of women pregnant by assisted reproduction (Nasseri et al., 1999; Massie et al., 2008; Haadsma et al., 2009) is mixed; implications for naturally conceived pregnancies are unclear. Our results for AMH are consistent with several studies of prenatal diagnosis samples which show no association with trisomy (Seifer et al., 2007; Li et al., 2010; Plante et al., 2010). Interpretation is hindered, however, by uncertainty about how pregnancy influences AMH levels. Our finding for FSH is consistent with our observation, in a previous prospective study, that menopause occurs 1 year earlier (95% CI: 22.10, 0.18) among women with trisomic pregnancy losses than among women with chromosomally normal pregnancies (Kline et al., 2000). At the time, we inferred that women with trisomic pregnancies had smaller oocyte pools than other women of the same chronologic age, so that they were chronologically younger when their pools dropped below the threshold necessary to maintain menstruation. Below, we propose a different inference. Our data do not support the oocyte pool hypothesis. As the size of the oocyte pool declines with age, so does the size of the antral follicle cohort (Block, 1952; Hansen et al., 2010). FSH is in a negative feedback loop with inhibin B, a product of the antral follicles. Thus, FSH may be an indirect indicator of the size of the underlying oocyte pool. This interpretation is inconsistent, however, with the absence of an association of trisomy with AMH. First, AMH, a product of primary, pre-antral and small antral follicles (Weenen et al., 2004; Stubbs et al., 2005), should represent the size of the pool as well as or better than FSH. Secondly, AMH is more strongly associated with chronologic age than is FSH (Table II; de Vet et al., 2002; van Rooij et al., 2005; Dafopoulos et al., 2010). The one study that examined correlations between primordial follicle count and hormone levels in women unselected for infertility shows a strong correlation (r ¼ 0.72) between the count and AMH (Hansen et al., 2010). (Since blood draws were not timed to the menstrual cycle, the correlation with FSH in these data is not informative.) An alternate interpretation is that the association of trisomy with elevated FSH reflects impaired function of the antral follicle cohort or of the dominant follicle. We think the association between trisomy and FSH indicates that elevated FSH is a characteristic of the woman (i.e. present in most menstrual cycles). Elevated FSH in the early follicular phase suggests diminished production of inhibin B by the antral follicle cohort. Diminished inhibin B levels could indicate (i) fewer antral follicles (the usual interpretation) or (ii) a change in the quality of the individual follicles such that each requires more FSH to develop or requires more time to produce sufficient inhibin B to inhibit FSH production. We have previously shown, in the New York study, that the antral follicle count on Days 5–7 does not differ between trisomy cases and LB controls (Kline et al., 2004). Thus, we reject a mechanism related to the number of antral follicles. One measure of follicular quality is inhibin B; yet we did not observe lower levels in trisomy cases. While we do not weigh our observation heavily (discussed earlier), a similar result was obtained in the study of women with Down syndrome births (van Montfrans et al., 2001). A study of follicular fluid in the dominant follicle, aspirated after stimulation with human chorionic

1547 gonadotrophin, suggests that the age-related changes in hormone levels in the follicular fluid do not march in lockstep with serum hormone levels (Klein et al., 2000). These data raise questions about the validity of serologic measures as indicators of the secretory activity of follicles. Thus, we cannot rule out the possibility that the elevated FSH associated with trisomy reflects an altered quality of the antral follicles. FSH may be elevated for reasons unrelated to the ovary, yet few factors have been identified. Postulated influences include genetic factors or long intervals of quiescence in the hypothalamic –pituitary ovarian axis (e.g. prior to puberty; Lambalk and de Koning, 1998). Our favored hypothesis is that elevated levels of FSH, at some time during the long resting period between the arrest of meiosis I and its resumption one to five decades later, increase the risk of meiotic error. In the ovary, proteins called cohesins maintain chromatid adhesion, facilitate segregation of the homologous chromosomes and keep sister chromosomes paired at the centromere until their separation at meiosis II at fertilization (Hodges et al., 2005; Hunt and Hassold, 2008). It has been speculated, for example, that deterioration in the protein complex over time might be relevant to the age-related rise in aneuploidy (Hunt and Hassold, 2008). Evidence from mice suggests that there is little, if any, natural turnover of cohesins during oocyte maturation (Revenkova et al., 2010; TachibanaKonwalski et al., 2010). Several recent publications support the idea that, at least in the mouse, loss of chromosome-associated cohesins in aging oocytes is associated with aneuploidy (Chiang et al., 2010; Lister et al., 2010). Thus, factors that increase aneuploidy may act through the cohesin pathway, e.g. by interacting with Shugoshin-2, which protects the centromeric cohesins from phosphorylation and release during meiosis I. Support for a direct role of FSH on meiosis is sparse. One study (McTavish et al., 2007) of transgenic-FSH mice provides evidence that rising levels of FSH accelerate reproductive aging. On the basis of their observations among super-ovulated mice, the authors also suggest that, at young and middle age, FSH may operate to recruit even suboptimal follicles (based on the morphology, not the karyotype, of oocytes or embryos.) In mouse oocytes matured in vitro (Roberts et al., 2005), exposure to elevated concentrations of FSH in culture is associated with increased aneuploidy and premature separation of sister chromatids. Another study (Xu et al., 2008), however, showed no difference between mature oocytes from natural cycles and immature oocytes, also from natural cycles, cultured in a solution of FSH and luteinizing hormone. A study in infertile women (Gras et al., 1992) compared aneuploidy rates in unstimulated cycles with aneuploidy rates in cycles stimulated with (i) human menopausal gonadotrophin (which contains FSH) and other hormones or (ii) a buserelin (GnRH agonist) flare. Although aneuploidy rates did not differ significantly among the three groups, the rate of aneuploidy with hormonal stimulation (34%), irrespective of type, was higher than the rate without (20%). We do not consider here studies which used FISH to assess aneuploidy from blastomeres because evidence casts doubt on the validity of aneuploidy diagnosis in these circumstances (Treff et al., 2010). In sum, we propose that exposure to high levels of FSH increases the risk of meiotic error. An effect might be through altering the intrafollicular environment in the cycle in which the trisomic pregnancy is conceived, including increasing the chance that an oocyte with an error is recruited, or by reducing the ability of chromosomes to remain paired during the resting phase between birth and ovulation,

1548 perhaps through a direct interaction with cohesins or their controlling proteins. In light of the timing of our measures in relation to the index pregnancy, we favor a hypothesis related to chronic exposure to elevated FSH, say 7 mIU/ml or higher. We speculate that chronically elevated levels lead to faster depletion of the oocyte pool. This speculation is consistent with our observation that age at menopause is earlier for women with trisomic pregnancies. Our hypothesis has implications for further research on a possible role of FSH in trisomy formation. Our observations suggest it may prove valuable to evaluate FSH level, in addition to chronologic age, when counseling women about their risk of trisomy.

Kline et al.

women who participated to further understanding of the causes of reproductive loss.

Funding Data collection for the New Jersey study was supported by a grant from the National Institutes on Child Health and Development (R01 HD 42725). Data collection for the New York study was supported by a grant from the National Institutes on Aging (R01 AG 15386). The work for this paper, including the hormone assays, was supported by a grant from the National Institutes on Child Health and Human Development (R01 HD 053814-01A2).

Authors’ roles J.K.K. designed the study and analysis and wrote the manuscript. A.M.K. collaborated in the design of the study and analysis, carried out the statistical programing and helped write the manuscript. B.L. collaborated in the design of the study and the analysis, carried out selected analyses and helped write the manuscript. A.C.K. collaborated in the design of the study and the interview questions related to hormonal conditions, assessed whether or not participants had hormonal conditions and reviewed the manuscript. M.F. selected the assays to measure hormones, oversaw the work of laboratory staff, assisted in the interpretation of hormone results and reviewed the manuscript. D.W. collaborated in the design of the study, oversaw the laboratory that karyotyped spontaneous abortion specimens, and collaborated in the interpretation of results and writing of the manuscript.

Supplementary data Supplementary data are available at http://humrep.oxfordjournals. org/.

Acknowledgements For the New Jersey study, we thank Dr Martin Hochberg and his colleagues for providing access to their patients. We especially thank Dr Arthur Christiano in Pathology, who facilitated our work and advised on diagnostic issues. We thank Larry Bologna, Denise Campbell, Gina Chavez, Lois Deyo, Cheryl Dulaff, Diane Gerardi, Nancy Librera, Deborah Manente, Mary Reiner, Louis Rizzo, Donna Rochette and Marriett Trentacoste, who facilitated our work at the study hospital. We thank Richard Buchsbaum, whose programing and data management expertize facilitated both the day-to-day fieldwork and the statistical analysis. We gratefully acknowledge Project Director L. Perry Brothers, Fieldworkers Melissa Bieliecki, Kathleen Carstens and Beth Fishner, and Renee Davenport, who assisted in tasks too countless to list. For the New York study, we thank Dr Grace Jorgensen and her colleagues for their help in providing access to their patients. We acknowledge Maria Bautista, Jennifer Cassin, Terry Fox, the late Kris Keough and Donna West who facilitated our work at the study hospital. We thank Megan Meldrum who carried out the fieldwork of the study, Renee Davenport who assisted in data processing and checking and Antonio Sobrino, who prepared the samples for karyotyping. Neither study would have been possible without the help of the

References Baker TG. A quantitative and cytological study of germ cells in human ovaries. Proc R Soc Lond B Biol Sci 1963;158:417 –433. Block E. Quantitative morphological investigations of the follicular system in women: variations at different ages. Acta Anat 1952;14:108– 123. Breslow NE, Day NE. The analysis of case-control studies. Statistical Methods in Cancer Research, Vol.1. IARC Scientific Publications no. 32. France: International Agency for Research on Cancer, 1980. Brook JD, Gosden RG, Chandley AC. Maternal ageing and aneuploid embryos—evidence from the mouse that biological and not chronological age is the important influence. Hum Genet 1984;66:41–45. Brown JR, Liu HC, Sewitch KF, Rosenwaks Z, Berkeley AS. Variability of day 3 follicle-stimulating hormone levels in eumenorrheic women. J Reprod Med 1995;40:620 – 624. Burger HG. Diagnostic role of follicle-stimulating hormone (FSH) measurements during the menopausal transition—an analysis of FSH, oestradiol and inhibin. Eur J Endocrinol 1994;130:38– 42. Burger HG. Inhibin and reproductive aging. Exp Gerontol 2000;35:33 – 39. Chiang T, Duncan FE, Schindler K, Schultz RM, Lampson MA. Evidence that weakened centromere cohesion is a leading cause of age-related aneuploidy in oocytes. Curr Biol 2010;20:1522 – 1528. Dafopoulos A, Dafopoulos K, Georgoulias P, Galazios G, Limberis V, Tsikouras P, Koutlaki N, Maroulis G. Smoking and AMH levels in women with normal reproductive history. Arch Gynecol Obstet 2010; 282:215– 219. de Vet A, Laven JS, de Jong FH, Themmen AP, Fauser BC. Antimullerian hormone serum levels: a putative marker for ovarian aging. Fertil Steril 2002;77:357 – 362. Freeman SB, Yang Q, Allran K, Taft LF, Sherman SL. Women with a reduced ovarian complement may have an increased risk for a child with Down syndrome. Am J Hum Genet 2000;66:1680 – 1683. Gail MH, Tan WY, Piantidosi S. Tests for no treatment effect in randomized clinical trials. Biometrika 1988;75:57– 64. Goldenberg RL, Grodin JM, Rodbard D, Ross GT. Gonadotropins in women with amenorrhea. The use of plasma follicle-stimulating hormone to differentiate women with and without ovarian follicles. Am J Obstet Gynecol 1973;116:1003 –1012. Gras L, McBain J, Trounson A, Kola I. The incidence of chromosomal aneuploidy in stimulated and unstimulated (natural) human oocytes. Hum Reprod 1992;7:1396 –1401. Groome NP, Illingworth PJ, O’Brien M, Pai R, Rodger FE, Mather JP, McNeilly AS. Measurement of dimeric inhibin B throughout the human menstrual cycle. J Clin Endocrinol Metab 1996;81:1401 – 1405. Gurbuz B, Yalti S, Ozden S, Ficicioglu C. High basal estradiol level and FSH/LH ratio in unexplained recurrent pregnancy loss. Arch Gynecol Obstet 2004;270:37 – 39.


Haadsma ML, Mooij TM, Groen H, Burger CW, Lambalk CB, Broekmans FJ, van Leeuwen FE, Bouman K, Hoek A. A reduced size of the ovarian follicle pool is associated with an increased risk of a trisomic pregnancy in IVF-treated women. Hum Reprod 2009; 25:552– 558. Hansen LM, Batzer FR, Gutmann JN, Corson SL, Kelly MP, Gocial B. Evaluating ovarian reserve: follicle stimulating hormone and oestradiol variability during cycle days 2 – 5. Hum Reprod 1996;11:486 – 489. Hansen KR, Knowlton NS, Thyer AC, Charleston JS, Soules MR, Klein NA. A new model of reproductive aging: the decline in ovarian non-growing follicle number from birth to menopause. Hum Reprod 2008; 23:699– 708. Hansen KR, Hodnett GM, Knowlton N, Craig LB. Correlation of ovarian reserve tests with histologically determined primordial follicle number. Fertil Steril 2010;95:170– 175. Hassold T, Chiu D. Maternal age-specific rates of numerical chromosome abnormalities with special reference to trisomy. Hum Genet 1985; 70:11– 17. Hatch M, Kline J, Levin B, Hutzler M, Warburton D. Paternal age and trisomy among spontaneous abortions. Hum Genet 1990;85:355 – 361. Hehenkamp WJ, Looman CW, Themmen AP, de Jong FH, Te Velde ER, Broekmans FJ. Anti-Mullerian hormone levels in the spontaneous menstrual cycle do not show substantial fluctuation. J Clin Endocrinol Metab 2006;91:4057 – 4063. Hodges CA, Revenkova E, Jessberger R, Hassold TJ, Hunt PA. SMC1beta-deficient female mice provide evidence that cohesins are a missing link in age-related nondisjunction. Nat Genet 2005; 37:1351– 1355. Hunt PA, Hassold TJ. Human female meiosis: what makes a good egg go bad? Trends Genet 2008;24:86 – 93. Jain T, Klein NA, Lee DM, Sluss PM, Soules MR. Endocrine assessment of relative reproductive age in normal eumenorrheic younger and older women across multiple cycles. Am J Obstet Gynecol 2003; 189:1080– 1084. Klein NA, Battaglia DE, Miller PB, Branigan EF, Giudice LC, Soules MR. Ovarian follicular development and the follicular fluid hormones and growth factors in normal women of advanced reproductive age. J Clin Endocrinol Metab 1996;81:1946 – 1951. Klein NA, Battaglia DE, Woodruff TK, Padmanabhan V, Giudice LC, Bremner WJ, Soules MR. Ovarian follicular concentrations of activin, follistatin, inhibin, insulin-like growth factor I (IGF-I), IGF-II, IGF-binding protein-2 (IGFBP-2), IGFBP-3, and vascular endothelial growth factor in spontaneous menstrual cycles of normal women of advanced reproductive age. J Clin Endocrinol Metab 2000;85: 4520 – 4525. Klein NA, Harper AJ, Houmard BS, Sluss PM, Soules MR. Is the short follicular phase in older women secondary to advanced or accelerated dominant follicle development? J Clin Endocrinol Metab 2002; 87:5746– 5750. Kline J, Levin B. Trisomy and age at menopause: predicted associations given a link with rate of oocyte atresia. Paediatr Perinat Epidemiol 1992;6:225 – 239. Kline J, Kinney A, Levin B, Warburton D. Trisomic pregnancy and earlier age at menopause. Am J Hum Genet 2000;67:395 –404. Kline J, Kinney A, Reuss ML, Kelly A, Levin B, Ferin M, Warburton D. Trisomic pregnancy and the oocyte pool. Hum Reprod 2004; 19:1633– 1643. Kline J, Kinney A, Kelly A, Reuss ML, Levin B. Predictors of antral follicle count during the reproductive years. Hum Reprod 2005;20:2179– 2189. Lambalk CB, de Koning CH. Interpretation of elevated FSH in the regular menstrual cycle. Maturitas 1998;30:215 – 220.

1549 Levin B. Conditional likelihood analysis in stratum-matched retrospective studies with polytomous disease states. Commun Stat 1987;B:699– 718. Levin B. The saddlepoint correction in conditional logistic likelihood analysis. Biometrika 1990;77:275– 285. Levin B, Robbins H. Urn models for regression analysis with applications to employment discrimination studies. Law Contemp Probl 1983; 46:247– 267. Li HW, Hui PW, Tang MH, Lau ET, Yeung WS, Ho PC, Ng EH. Maternal serum anti-Mullerian hormone level is not superior to chronological age in predicting Down syndrome pregnancies. Prenat Diagn 2010; 30:320– 324. Lister LM, Kouznetsova A, Hyslop LA, Kalleas D, Pace SL, Barel JC, Nathan A, Floros V, Adelfalk C, Watanabe Y et al. Age-related meiotic segregation errors in Mammalian oocytes are preceded by depletion of cohesin and Sgo2. Curr Biol 2010;20:1511 – 1521. Luisi S, Florio P, Reis FM, Petraglia F. Inhibins in female and male reproductive physiology: role in gametogenesis, conception, implantation and early pregnancy. Hum Reprod Update 2005; 11:123– 135. Massie JA, Burney RO, Milki AA, Westphal LM, Lathi RB. Basal follicle-stimulating hormone as a predictor of fetal aneuploidy. Fertil Steril 2008;90:2351– 2355. McTavish KJ, Jimenez M, Walters KA, Spaliviero J, Groome NP, Themmen AP, Visser JA, Handelsman DJ, Allan CM. Rising follicle-stimulating hormone levels with age accelerate female reproductive failure. Endocrinology 2007;148:4432– 4439. Nasseri A, Mukherjee T, Grifo JA, Noyes N, Krey L, Copperman AB. Elevated day 3 serum follicle stimulating hormone and/or estradiol may predict fetal aneuploidy. Fertil Steril 1999;71:715 – 718. Plante BJ, Beamon C, Schmitt CL, Moldenhauer JS, Steiner AZ. Maternal antimullerian hormone levels do not predict fetal aneuploidy. J Assist Reprod Genet 2010;27:409– 414. Revenkova E, Herrmann K, Adelfalk C, Jessberger R. Oocyte cohesin expression restricted to predictyate stages provides full fertility and prevents aneuploidy. Curr Biol 2010;20:1529– 1533. Risch N, Stein Z, Kline J, Warburton D. The relationship between maternal age and chromosome size in autosomal trisomy. Am J Hum Genet 1986;39:68 – 78. Roberts R, Iatropoulou A, Ciantar D, Stark J, Becker DL, Franks S, Hardy K. Follicle-stimulating hormone affects metaphase I chromosome alignment and increases aneuploidy in mouse oocytes matured in vitro. Biol Reprod 2005;72:107– 118. Scriver J, Baker VL, Young SL, Behr B, Pastore LM. Inter-laboratory validation of the measurement of follicle stimulating hormone (FSH) after various lengths of frozen storage. Reprod Biol Endocrinol 2010;8:145. Seifer DB, MacLaughlin DT, Cuckle HS. Serum mullerian-inhibiting substance in Down’s syndrome pregnancies. Hum Reprod 2007; 22:1017– 1020. Streuli I, Fraisse T, Pillet C, Ibecheole V, Bischof P, de Ziegler D. Serum antimullerian hormone levels remain stable throughout the menstrual cycle and after oral or vaginal administration of synthetic sex steroids. Fertil Steril 2008;90:395– 400. Stubbs SA, Hardy K, Da Silva-Buttkus P, Stark J, Webber LJ, Flanagan AM, Themmen AP, Visser JA, Groome NP, Franks S. Anti-Mullerian hormone (AMH) protein expression is reduced during the initial stages of follicle development in human polycystic ovaries. J Clin Endocrinol Metab 2005;90:5536 – 5543. Tachibana-Konwalski K, Godwin J, van der Weyden L, Champion L, Kudo NR, Adams DJ, Nasmyth K. Rec8-containing cohesin maintains bivalents without turnover during the growing phase of mouse oocytes. Genes Dev 2010;24:2505 – 2516.

1550 Thomford PJ, Jelovsek FR, Mattison DR. Effect of oocyte number and rate of atresia on the age of menopause. Reprod Toxicol 1987;1:41 – 51. Treff NR, Levy B, Su J, Northrop LE, Tao X, Scott RT Jr. SNP microarray based 24 chromosome aneuploidy screening is significantly more consistent than FISH. Mol Hum Reprod 2010;16:583 – 589. Trout SW, Seifer DB. Do women with unexplained recurrent pregnancy loss have higher day 3 serum FSH and estradiol values? Fertil Steril 2000;74:335 – 337. van Montfrans JM, Dorland M, Oosterhuis GJ, van Vugt JM, Rekers-Mombarg LT, Lambalk CB. Increased concentrations of follicle-stimulating hormone in mothers of children with Down’s syndrome. Lancet 1999;353:1853– 1854. van Montfrans JM, Lambalk CB, van Hooff MH, van Vugt JM. Are elevated FSH concentrations in the pre-conceptional period a risk factor for Down’s syndrome pregnancies?. Hum Reprod 2001;16:1270– 1273. van Rooij IA, Broekmans FJ, Scheffer GJ, Looman CW, Habbema JD, de Jong FH, Fauser BJ, Themmen AP, te Velde ER. Serum antimullerian hormone levels best reflect the reproductive decline with age in normal women with proven fertility: a longitudinal study. Fertil Steril 2005;83:979 – 987.

Kline et al.

Warburton D. The effect of maternal age on the frequency of trisomy: change in meiosis or in utero selection? Prog Clin Biol Res 1989; 311:165– 181. Warburton D, Kline J. Maternal age and aneuploidy: A re-examination of the limited oocyte pool hypothesis. Salisbury Medical Journal 2001; 101(suppl.):51– 64. Warburton D, Kline J, Kinney A, Yu CY, Levin B, Brown S. Skewed X chromosome inactivation and trisomic spontaneous abortion: no association. Am J Hum Genet 2009;85:179– 193. Weenen C, Laven JS, Von Bergh AR, Cranfield M, Groome NP, Visser JA, Kramer P, Fauser BC, Themmen AP. Anti-Mullerian hormone expression pattern in the human ovary: potential implications for initial and cyclic follicle recruitment. Mol Hum Reprod 2004;10:77– 83. Welt CK, McNicholl DJ, Taylor AE, Hall JE. Female reproductive aging is marked by decreased secretion of dimeric inhibin. J Clin Endocrinol Metab 1999;84:105 – 111. Xu L, Wang Y, Zhou P, Cao YX, Huang TH, Chian RC. Cytogenetic analysis of in vivo and in vitro matured oocytes derived from naturally cycling and stimulated mice. Syst Biol Reprod Med 2008;54: 155 – 162.