Menstrual cycle variability and the perimenopause

Menstrual Cycle Variability and the PerimenopauseKATHLEEN A. O’CONNOR,1,3* DARRYL J. HOLMAN,1,3 AND JAMES W. WOOD2,3

1Department of Anthropology, Center for Studies of Demography and Ecology, University ofWashington, Seattle, Washington2Department of Anthropology, Pennsylvania State University, University Park, Pennsylvania3Population Research Institute, Pennsylvania State University, University Park, Pennsylvania

ABSTRACT Menopause, the final cessation of menstrual cycling, occurs when the pool of ovarianfollicles is depleted. The one to five years just prior to the menopause are usually marked by increas-ing variability in menstrual cycle length, frequency of ovulation, and levels of reproductive hormones.Little is known about the mechanisms that account for these characteristics of ovarian cycles as themenopause approaches. Some evidence suggests that the dwindling pool of follicles itself is respon-sible for cycle characteristics during the perimenopausal transition. Another hypothesis is that theincreased variability reflects “slippage” of the hypothalamus, which loses the ability to regulatemenstrual cycles at older reproductive ages. This paper examines the underlying cause of the increas-ing variability in menstrual cycle length prior to the menopause. A model of ovarian cycles is devel-oped, based on the process of follicular growth and depletion. Under this model, the follicular phaseof each menstrual cycle is preceded by an inactive phase, a period of time when no ovarian follicleshave left the resting state and begun secreting steroids in response to gonadotropin stimulation. Themodel makes predictions about the variability in menstrual cycles across the reproductive life spanbased on the size of the surviving pool of ovarian follicles. We show that the model can explain severalcharacteristics of the perimenopause in humans and macaques and illustrate how the model can beapplied to research on the biological and cultural correlates of the timing of menopause. J. Hum. Biol.13:465–478, 2001. © 2001 Wiley-Liss, Inc.

Across most of a woman’s reproductivelife span, menstrual cycles exhibit relativelylittle variability in length around a medianof 28 days (Treloar et al., 1967). Beginningfrom one to five years before menopause(the perimenopause), menstrual cyclelength becomes increasingly variable. Inparticular, this time period is characterizedby a higher frequency of long cycles (Fig. 1).Additionally, a higher proportion of men-strual cycles are anovulatory (Metcalf,1983), and hormone levels become increas-ingly variable (Burger, 1996). Menopause it-self is the complete and irreversible cessa-tion of ovarian cycling.

The mechanism responsible for meno-pause has long been recognized to be ex-haustion or near exhaustion of the pool ofovarian follicles (Richardson et al., 1987;Gougeon et al., 1994; vom Saal et al., 1994).Less is known about the mechanisms re-sponsible for the increasingly variable na-ture of menstrual cycles across the peri-menopausal transition. At least two majorideas have been proposed. One hypothesis isthat increasingly variable cycles across theperimenopausal transition result from abreakdown in the signaling between theovary and the hypothalamus (e.g., Wise et

al., 1996; Wise 1999). The pervasive obser-vation of elevated levels of FSH in women atolder reproductive ages is cited as supportfor this hypothesis, although the mecha-nism(s) contributing to elevated FSH or hy-pothalamic faltering are unspecified (Met-calf et al., 1982; MacNaughton et al., 1992;Klein et al., 1996; Santoro et al., 1996).

An alternative hypothesis suggests thatthe increasing variability in menstrual cyclelength is almost entirely a result of the di-minished number of ovarian follicles re-maining in the ovary in the years precedingthe menopause (Richardson et al., 1987).We have proposed a mechanism and modelfor this hypothesis (O’Connor et al., 1998).The current paper presents the theoreticaldetails and implications of the model, with afocus on how it can account for many of theobserved changes in menstrual cycle char-acteristics across the perimenopause and

Contract grant sponsor: NIA; Contract grant numbers: 1 RO1AG15141, 5 T32 AG00208; Contract grant sponsor: NICHD;Contract grant numbers: F32 HD 07994, 2 P30 HD28263.

*Correspondence to: Kathleen A. O’Connor, Department ofAnthropology, University of Washington, Box 353100, Seattle,WA 98195. E-mail: [email protected]

Received 12 May 2000; Revision received 8 September 2000;Accepted 2 February 2001

AMERICAN JOURNAL OF HUMAN BIOLOGY 13:465–478 (2001)

© 2001 Wiley-Liss, Inc.

PROD #M20037R

into the menopause. Under this model theincreased proportion of long menstrualcycles during the perimenopause are a di-rect result of periods of ovarian inactivitythat increase in length as the pool of ovarianfollicles diminishes. These so-called “inac-tive” phases of the menstrual cycle have adistinct endocrine signature, and we pro-vide evidence of inactive phases from endo-crine studies in humans and macaques.

FOLLICULAR DEVELOPMENTAND DEPLETION

The story of menopause begins withevents occurring in utero. Early during em-bryonic development of the female, a life-time stock of primary oocytes is formed atthe gonadal ridge by rounds of mitosis froma pool of oogonia. Each oocyte then under-goes meiosis I through prophase. The finalpool of oocytes is formed by the second tri-mester of pregnancy, and each oocyte is sur-

rounded by a layer of granulosa cells, calleda follicle, as the ovaries form (Gougeon,1996). The follicle is the structure thathouses and nourishes the oocyte, and it isthe follicle that receives and sends hor-monal signals to the other parts of the re-productive axis to maintain regular ovariancycles.

Since this pool of ovarian follicles and oo-cytes is non-replenishable, the maximumnumber of ovarian follicles peaks in utero atabout 7 million (Austin and Short, 1982).From this point on, the pool of follicles isdepleted through a process called follicularatresia (Fig. 2). By the time of birth, the poolof follicles has depleted to roughly one halfmillion per ovary (Block, 1953). By menar-che, the pool has been reduced to about100,000 follicles per ovary. The perimeno-pausal transition, which may last up to fiveor six years, begins when approximately 100to 1,000 follicles are left in each ovary (Ri-

Fig. 1. Variation (by percentiles) in menstrual cycle length by age and perimenopausal status. Data are from along-term prospective study of American women. Redrawn from Treloar et al. (1967:90).

466 K.A. O’CONNOR ET AL.

chardson et al., 1987; Gougeon, 1996).Menopause itself is the ultimate exhaustionof the follicle stock.

Mathematically, this can be expressed asan exponential (or log–linear) decline in thenumbers of surviving follicles with age (Fig.2). This same pattern of exponential deple-tion of follicles throughout life is found inother primates and mammals (Austin andShort, 1982; Miller et al., 1999). Data fromautopsy and oophorectomy sources are sug-gestive of an increase in the rate of folliculardepletion in the years prior to menopause inwomen (Faddy and Gosden, 1995), althoughthe mechanism responsible for this accel-eration is unclear.

The process by which follicular depletionoccurs is not well understood. A resting fol-licle (one that has not begun to grow) ini-tiates growth by a process that is poorly un-derstood. Definitive endocrine, autocrine, orparacrine factors have not been identified,but gonadotropin hormones are not believedto be necessary to trigger the initial growthof the follicle (Gougeon, 1996; McGee andHsueh, 2000). Each and every follicle ap-pears to have a small and nearly constantprobability of entering into this initial grow-ing state at any point in time across the lifespan.

Growing follicles usually survive for onlya short time before undergoing atresia. Themechanisms that determine when and whyatresia occurs to particular follicles are stillunknown. The vast majority of follicles un-dergo atresia, which is a form of apoptosis

(Hsueh et al., 1994; Kaipia and Hsueh,1997). The few exceptions are those folliclesthat continue to grow and become the dom-inant follicle during a menstrual cycle.These follicles are the ones that secrete ste-roid hormones and participate in ovulationand in luteinization after ovulation.

HYPOTHALAMIC–PITUITARY–OVARIAN AXIS

The ovarian follicles communicate andwork in conjunction with the hypothalamusand pituitary to maintain regular ovariancycles. The GnRH pulse generator, a neuralcomplex located in the hypothalamus, con-trols pulsatile secretion of gonadotropin-releasing hormone (GnRH) by hypothalamicneurons. GnRH, in turn, is carried to theanterior lobe of the pituitary gland by theportal capillary system, where it stimulatessecretion of luteinizing hormone (LH) andfollicle stimulating hormone (FSH). Thesehormones stimulate follicular growth anddevelopment and are necessary for initiat-ing and maintaining follicular production ofthe ovarian steroids estradiol and proges-terone. Feedback relationships between theovarian steroids, on the one hand, and thehypothalamic and pituitary gonadotropins,on the other, maintain the highly regularpattern of ovarian cycles in the adult fe-male.

The four main hormones involved in thesefeedback relationships, and how theychange across the menstrual cycle areshown in the upper two panels of Figure 3.These data are from collection and analysisof daily urine samples across three men-strual cycles from a normally cyclingwoman. The first day of menses is labeled ascycle day one on the x axis. The top graphshows the main urinary metabolites, es-trone-3-glucuronide (E3G) and pregnane-diol-3a-glucuronide (PDG), of the ovariansteroids estradiol and progesterone. Thesemetabolites parallel serum levels of estra-diol and progesterone (Munro et al., 1991).The pituitary hormones LH and FSH areshown in the lower graph of the top panel ofFigure 3.

In a typical menstrual cycle, steroid hor-mones are low at the start of each men-strual cycle (day 1 of menses), reflecting theabsence of any follicles large enough to pro-duce measurable levels of estradiol. In theabsence of ovarian steroids, the gonadotro-pin hormones LH and especially FSH be-

Fig. 2. Surviving numbers of follicles in women byage, based on counts of follicles in ovaries removed inautopsies and oophorectomies. Data are from Block(1952, 1953) and Gougeon et al. (1994).

CYCLE VARIABILITY AND THE PERIMENOPAUSE 467

Fig. 3. Urinary steroid and gonadotropin profiles for a regularly cycling woman (top panels) and a menopausalwoman (bottom panels). Each cycle day 1 is the first day of menses. Hormone values are corrected by specificgravity. Source: Pennsylvania State University Reproductive Endocrinology Laboratory.


come elevated. The elevated gonadotropinsstimulate follicular growth and steroid pro-duction in any follicles that have matured tothe point of developing receptors for FSH.Follicles responding to FSH will continue tothe next stage of growth and begin to se-crete estradiol. This estradiol then acts in anegative feedback loop with the hypothala-mus to down-regulate gonadotropin secre-tion. Hence, as estradiol secretion begins,LH and FSH secretion declines. Becausegrowing follicles secrete estradiol in propor-tion to their size, estrogen rises slowly atfirst and then more rapidly as one or morefollicles grow to larger sizes. One follicle willeventually become the dominant folliclethrough a poorly understood process calledfollicular selection.

Once the estrogen reaches very high lev-els (around day 14 or mid-cycle), a surge ofLH (and to a lesser extent, FSH) is trig-gered. This surge causes the dominant fol-licle to rupture and release the oocytewithin it, i.e., ovulation. The part of thecycle from cycle day 1 until ovulation iscalled the follicular phase, because the mainactivity in this part of the cycle is folliculargrowth.

After ovulation, the follicle from whichthe egg erupted differentiates into the cor-pus luteum, an endocrine gland that se-cretes mostly progesterone but also some es-trogen. The second half of the ovarian cycleis called the luteal phase, after the corpusluteum. The function of progesterone andestrogen in this phase is to finish preparingthe uterus for pregnancy. The elevated lev-els of progesterone and estrogen preventsignificant secretion of LH or FSH. In theabsence of pregnancy, the corpus luteum re-gresses until levels of progesterone are toolow to maintain the uterine lining. The uter-ine lining is then shed, resulting in mensesand the start of the next menstrual cycle. Asthe steroid levels decline at the end of anovarian cycle, LH and FSH slowly elevate asthey are released from negative feedbackcontrol. The elevated level of FSH thenstimulates the next cohort of follicles to de-velop. This set of feedback relationships ofthe hypothalamic–pituitary–ovarian (HPO)axis thus serves to maintain regular ovari-an cycles throughout the reproductive lifespan.

ENDOCRINE PATTERNS OF THEMENOPAUSE AND PERIMENOPAUSE

In menopause, one part of the HPO sys-tem breaks down as the ovaries become de-pleted of follicles. When there are no ovari-an follicles, there is no ovarian production ofestrogen or progesterone. There is, there-fore, a loss of negative feedback at the hy-pothalamus so that pituitary secretion ofLH and FSH increases to high levels. Inpost-menopausal women, LH and FSH lev-els are consistently high, reflecting this lackof feedback from ovarian steroids.

This can be clearly seen in the lower twographs of Figure 3, which show daily con-centrations of urinary steroids and gonado-tropins across 3 months in a postmeno-pausal woman. There are negligible levels ofovarian steroids, and in the absence of ovar-ian steroid feedback, gonadotropin concen-trations are about 4-fold higher than thosefound in a cycling woman.

The transition to menopause thus in-volves a switch from a well-orchestratedsystem of feedback and control to one of un-regulated gonadotropin release in the ab-sence of ovarian production of steroids. Thefinal transition can be seen in Figure 4,which tracks a 48-year-old woman acrossapproximately 2 months as she experiencesher last-ever menstrual bleed. It is clearfrom plasma progesterone levels that thecycle preceding her final menses was anovu-latory. Following the last menses is an in-terval of about 2 weeks during which someestradiol is present, presumably reflectingsome degree of partial follicular activity. Af-ter this interval, the levels of both estradioland progesterone are very low and unchang-ing, and will remain so for the rest of thewoman’s life. Coinciding with the decline inovarian steroids is a marked elevation inLH and FSH, indicating that the hypotha-lamic–pituitary axis has been freed fromnegative feedback by ovarian steroids. Hergonadotropins are likely to be high and vari-able throughout the rest of her life (Hee etal., 1993).

The period of transition from regularovarian cycles to a complete absence of ovar-ian cycles is characterized by increasinglyirregular cycle length, and a rapidly dwin-dling pool of follicles. Increasingly variablehormone profiles also occur during thistransitional period. In fact, hormone pro-


files in the perimenopause often contain el-ements of both menopausal and cycling hor-mone profiles.

Figure 5 shows three menstrual cycles fora perimenopausal 45-year-old woman.Throughout most of the period of observa-tion, she experiences normal, ovulatorycycles. Beginning, however, at about cycleday 3 of the third cycle (boxed-in area), shegoes through about a 10-day period with nosign of ovarian steroidogenesis. At the sametime, her LH and FSH rise rapidly, presum-ably because of an absence of negative feed-back from steroids. We believe that this isevidence that there are no follicles at theappropriate developmental state to beginfurther growth under the influence of FSH.Eventually, however, one or more folliclesbegins growing, as evidenced by the risingestrogen; LH and FSH levels decline as thesteroids down-regulate the hypothalamus,and the cycle continues normally throughovulation and the luteal phase.

This period of time when no follicles aredeveloping looks, hormonally, very much

like menopause. If we had analyzed a singleurine or blood specimen during this 10-dayperiod, we might have concluded that thiswoman was menopausal. But clearly she isnot; this brief period of “false menopause” isfollowed by ovulation. One effect of the2-week period of ovarian quiescence is thatthis particular cycle is inordinately long, asis often the case during the perimenopause.

THREE-PHASE MODEL OF THEMENSTRUAL CYCLE

These transient episodes of ovarian qui-escence are called “inactive phases” of themenstrual cycle. Our hypothesis is that in-active phases are direct reflections of thevery small pool of remaining follicles. Inac-tive phases are characterized by the absenceof steroid production and correspond tothose periods of time when no follicles havebegun growing. When the stock of follicles islarge, as at young reproductive ages, inac-tive phases will be rare as there will nearlyalways be a cohort of follicles at the rightstage of development (the antral stage,

Fig. 4. Plasma concentrations of ovarian steroids (top panel) and gonadotropins (bottom panel) during themenopausal transition in a 48-year-old woman. Redrawn from van Look et al. (1977).


when follicles are able to respond to FSHand begin production of estradiol; see Mc-Gee and Hsueh, 2000, for a review of follic-ular development) at the start of any givenmenstrual cycle. At older ages, however,when the follicle reserve is low, there will,by chance, be periods of time when no fol-licles are at the antral stage of developmentat the start of a menstrual cycle. The ovaryis, in essence, “sputtering” when the folliclereserve is very low. Inactive phases are anatural statistical outcome of repeated sam-pling from a stock containing small absolutenumbers of items. Menopause is effectivelya permanent inactive phase.

We have observed several examples of in-active phases in our laboratory and the pub-lished literature (Sherman et al., 1976;Shideler et al., 1989; Hee et al., 1993; San-toro et al., 1996). The characteristic endo-crine pattern is one of low and unchanginglevels of reproductive steroids, coupled withelevated concentrations of FSH and to someextent LH. This pattern is seen in Figure 5

from our laboratory and in Figure 6 fromtwo published studies of the perimenopause(Fig. 6 bottom panel from Santoro et al.,1996; top panel from Shideler et al., 1989).Note that in each example the length of thecycle containing the inactive phase is con-siderably longer than the cycles precedingor following it.

If our hypothesis that inactive phases aredirect reflections of the very small pool ofremaining follicles is correct, then the dis-tribution of inactive phases by age should bepredictable, given an underlying rate atwhich follicles initiate growth. The fre-quency and mean length of inactive phasesshould increase with age, as the follicularreserve approaches exhaustion.

We model the distribution of inactivephase lengths by decomposing the ovariancycle into three distinct phases (Fig. 7): theinactive phase, the follicular phase, and theluteal phase (O’Connor et al., 1998). Thefirst phase is the inactive phase—the periodof time in which no antral follicles have ini-

Fig. 5. Urinary steroid and gonadotropin excretion in a 45-year-old perimenopausal woman (as assessed byself-reports of irregular cycle lengths), showing an inactive phase (boxed-in area) at the beginning of the thirdmenstrual cycle. Each cycle day 1 is the first day of menses. Hormone values are corrected by specific gravity.Source: Pennsylvania State University Reproductive Endocrinology Laboratory.


tiated growth or matured to the point of pro-ducing estradiol. This phase should usuallybe vanishingly small in younger women butmay be longer in perimenopausal women.

The inactive phase is followed by the follic-ular phase, which encompasses the periodfrom the first follicular secretion of estradiolto the time of ovulation. The luteal phase

Fig. 6. Urinary endocrine profiles showing inactive phases (boxed-in areas). The top panel shows two menstrualcycles in a 45-year-old woman (redrawn from Shideler et al., 1989), and the bottom panel shows an inactive phasein a perimenopausal woman (redrawn from Santoro et al., 1996). Shaded bars are days of menses; E1 is estrone-3-glucuronide and other urinary metabolites.


remains as traditionally defined, the timefrom ovulation to menses. The entire inter-menstrual interval is the sum of these threephases (Fig. 7).

In this model, the rate at which folliclesinitiate growth is assumed to be propor-tional to the rate of follicular depletion. Allfollicles are assumed to have the same haz-ard of initiating growth, and this hazarddoes not change across the life span. We alsoassume homogeneity across women in thesize of the initial follicle pool and in the rateof atresia. Other versions of the model in-corporate unmeasured heterogeneity in therate of atresia or follicle pool size or explic-itly incorporate covariates (such as cigarettesmoking) that might influence the rate ofatresia and thus the timing of menopause(discussed below).

A probability model of inactive phaselengths is given to illustrate the implica-tions of the model for menstrual cycle lengthvariability. Assume that women begin withan initial pool of n0 follicles, that folliclesinitiate growth with a constant hazard l perfollicle (roughly corresponding to the con-stant rate of follicular atresia seen in Fig.2), all follicles eventually initiate growth,and all but a small fraction of follicles un-dergo atresia. For this version of the modelwe assume that l does not vary across fol-licles and that there is no variation in n0among women (most of these assumptionsare relaxed in more complicated versions).From these assumptions, the number of fol-licles remaining in the ovaries at age a is na4 n0exp(−la). This gives rise to the distri-bution of times for a woman age a to be inthe inactive phase as a function of the num-

ber of surviving follicles in her ovaries andthe probability that no follicles initiategrowth. An inactive phase occurs when allna surviving follicles have not begun grow-ing at the beginning of a menstrual cycle.From this it follows that the probability ofan inactive phase exceeding length t at agea is Pr(T > t|l, n0, a) 4 exp(−tln0e−la). Themean time in the inactive phase at age a isE(T|a) 4 (ln0e−la)−1 and the variance isV(T|a) 4 (ln0e−la)−2.

Figure 8 shows the distribution of inac-tive phases by age, predicted by this simplemodel. The top panel illustrates the prob-ability of remaining in an inactive phase fora given number of days, at five differentages. Clearly, a 30-year-old woman will al-most always have extremely short, andlargely undetectable, inactive phases. Byage 40, the median length is still quite shortbut there is now some non-negligible prob-

Fig. 7. The three-phase model of the ovarian cycle.The inactive phase corresponds to periods of no detect-able ovarian steroidogenesis. The shaded box denotesmenses. See text for discussion of each phase.

Fig. 8. Hypothetical example of times spent in inac-tive phase, under a simple model that assumes an ini-tial follicle pool (at birth) of 1,000,000 follicles and adaily hazard of a follicle entering the growing state of l4 0.00065 (based on data from Block, 1952, 1953, andGougeon et al., 1994). The top panel shows the probabil-ity of no follicles initiating growth, at five different ages.The bottom panel shows the mean and variance in in-active phase length by age of a woman.


ability of an inactive phase being up to 12days long. By 50 years of age, many cyclesare expected to have long inactive phases.

The age-specific mean lengths for inactivephases are shown in the bottom panel ofFigure 8. At younger ages, the length of theinactive phase is short enough so that theinactive phase almost always adds no mea-surable time to the overall length of themenstrual cycle, and the variability is lowenough that we would rarely expect an in-active phase of any discernible length. Asmenopause approaches, both the mean andvariability in the length of the inactivephase increases greatly, showing a patternthat is strikingly similar to the increasedvariability in menstrual cycle length duringthe perimenopause (compare Fig. 1 withFig. 8 lower panel).

One prediction that follows from themodel and Figure 8 is that we should veryoccasionally see inactive phases in youngwomen. Figure 9 shows an example fromour laboratory that we believe is a short in-active phase in a healthy, normally cycling

25-year-old woman. The boxed-in area high-lights a brief period where the steroid hor-mones are quite low and constant and LHand FSH are elevated due to the absence ofnegative feedback from reproductive ste-roids. There are several missed days of col-lection following the inactive phase in thisstudy participant, but the rising estrogenand LH peak after the inactive phase indi-cate that follicular development and ovula-tion ultimately did occur.

Another prediction following from ourmodel is that we should expect to see inac-tive phases in other animals with similarreproductive biology and life history traits.If the follicular depletion system is the ulti-mate cause of the hormonal and ovariancycle features characteristic of the peri-menopause and menopause, then inactivephases should be present in those speciespossessing follicular depletion. The follicu-lar depletion system is in fact a primitivetrait that is highly conserved across mostmammals (Austad, 1997) and many verte-brates (Norris, 1997). In most mammalian

Fig. 9. Urinary steroid and gonadotropin excretion in a 25-year-old cycling woman showing a short inactivephase at the beginning of the second menstrual cycle (boxed-in area). Each cycle day 1 is the first day of menses.Hormone values are corrected by specific gravity. Source: Pennsylvania State University Reproductive Endocri-nology Laboratory.


species, the stock of follicles tends to lastlonger than the average life span, thereforeevidence of inactive phases should be ex-pected only in long-lived species that sur-vive beyond reproductive ages.

Hormonal data for large-bodied or long-lived mammals are rare. Some excellentdata are available from a study by Gilardiand colleagues (1997) that examines the en-docrinology, ovarian histology and bleedingpatterns of the perimenopause and meno-

pause of captive rhesus macaques. The toppanel of Figure 10 shows daily urinary es-trogen and progesterone metabolites for a20- to 25-year-old regularly cycling rhesusmacaque from this study. Unfortunately,there are no LH or FSH data for these ma-caques, but we can see clearly here that thisanimal is experiencing regular cycles, andthey are remarkably like those seen in hu-man females. The middle panel shows hor-mone data for another macaque from the

Fig. 10. Urinary estrone conjugates (E1C) and progesterone excretion for three rhesus macaques. The top panelis from a normally cycling macaque. The stippled bars show days of menses. The middle panel is from a menopausalmacaque. The bottom panel is from a perimenopausal macaque and shows a long inactive phase followed byfollicular development, ovulation and menses. Redrawn from Gilardi et al. (1997).


same study, but this animal is 29 years oldand amenorrheic. This is the hormone pro-file of a menopausal macaque. The bottompanel shows a perimenopausal macaque inher mid-twenties. The first 40 or so days ofcollection for this animal look menopausal,but then there is a period of sustained es-trogen production indicative of follicular de-velopment. This is followed by a rise in pro-gesterone, suggesting ovulation hasoccurred, and then a menses. Without LHand FSH it is hard to be absolutely sure thatthe prolonged period of steroid inactivityhere is an inactive phase, but this profilelooks exactly as we would predict during theperimenopause.

The simple model presented above can beextended to accommodate more realistic as-sumptions. For example, we can explicitlymodel the increase in the rate at which fol-licles are depleted at later reproductive ages(Faddy and Gosden, 1995). One possiblemechanism for the increased rate of deple-tion, which has found experimental supportin the work of Flaws et al. (1997), is thatexposure to high gonadotropin levels maydirectly damage ovarian follicles. If so, thengonadotropin damage may increase at laterreproductive ages when LH and FSH con-centrations periodically increase during in-active phases. Thus, there would be twocommon routes by which follicles are de-pleted from the ovaries: atresia, which is ex-pected to remain a negative exponentialprocess, and gonadotropin damage, whichwould be some function of inactive phaselength at a given age. Together these twocauses of depletion describe a multifactorialprocess proposed by Leidy et al. (1998).

Under one such multifactorial model,some proportion f of surviving follicleswould die in proportion to the expectedlength of the inactive phase at age a. A firstorder approximation of this model (which ig-nores second-order effects when the folliclepool is very small) is dna/da 4 −[lna +fna(lna)−1], and simplifies to dna/da 4−[ lna + f/l]. The term fna(lna)−1 is theexpectation from the exponential atresiaprocess, weighted by the fraction of suscep-tible follicles (fna) that might be damagedby a long inactive phase. The survivingnumber of follicles at age a is S(a) 4exp(−la) + f[exp(−la) − 1]/(n0l2), andshows a nearly log–linear decline in folliclenumbers with an accelerated depletion atthe oldest ages, similar to Figure 2.

A second useful extension explicitly mod-els the requirement that early growingfollicles must survive until they begin pro-ducing estradiol in the presence of gonado-tropins (which typically happens at a size ofabout 2 mm; Gougeon, 1996) before they canterminate the inactive phase. One simpleway to accommodate this extension is to as-sume a proportional fraction h of earlygrowing follicles survive to the 2 mm stage.This revision gives Pr(T > t|l, n0, h) 4exp(−tlhn0e−lha). Essentially, this model isthe same as the simplest model, except thatl has been rescaled by h. More complex ver-sions would treat the survival to the 2 mmstage as a multistate process.

Some applications of the inactive phasemodel include estimating the rate of follicu-lar atresia, estimating the initial folliclepool size, and examining the independenteffects of covariates on the rate of follicularatresia. The rate of atresia is typically esti-mated by counting follicles in ovaries overthe lifespan (Block, 1952, 1953; Gougeon etal., 1994; Richardson et al., 1987). The ova-ries for this research are obtained from au-topsy and oophorectomy cases. The inactivephase model provides another way of esti-mating the rate of atresia as well as the sizeof the initial follicle pool. The data for thisapproach consist of observations of inactivephases over some period of the lifecourse,derived from daily urine (or blood) samplesassayed for reproductive steroids and go-nadotropins. A hormonal criterion deter-mines the start and end of each inactivephase. For each menstrual cycle we observean inactive phase of t days length for awoman of age a. The probability for the ithsuch observation is rnexp{−r[ai + tinexp(−rai)]}, where n is the size of the initialpool of follicles and r is the rate of atresiaunder the simple model. If r and n are ho-mogenous among women, then the likeli-hood for N such observations of inactivephases is

L = )i=1

N

rn exp@−r~ai + tine−rai!#

and the values of parameters r and n thatmaximize the likelihood are the estimatesfor the rate of atresia and the initial pool offollicles, respectively.


We can also estimate the independent ef-fects of covariates (e.g., parity, cigarettesmoking, age at menarche, breastfeeding,particular hormonal environments, geneticmakeup) on the rate of atresia. Covariatesthat are fixed across the lifespan as well asthose that change over time can be exam-ined, although here we only illustrate howto do this for fixed covariates with thesimple model. Suppose covariates act by di-rectly increasing or decreasing the rate ofatresia. Call xi the value of a measured co-variate for the ith individual. The effect ofthe covariate will be modeled by parameterb. For the ith individual the rate of atresiais thus ri 4 r exp(xib), and we replace r withri in the likelihood above. In addition tofinding estimates for r and n, we can alsoestimate a value for the b parameter, whichgives us the magnitude of the covariate ef-fect on the rate of atresia. Additional covari-ates can be included as an additive array ofcovariates and parameters so that ri 4 xi1b1+ xi2b2 + . . . . Other extensions control forthe non-independence of repeated observa-tions within women and provide for covari-ates that vary with time.

CONCLUSIONS

We have proposed that follicular deple-tion and the inactive phase are the mecha-nisms underlying the increasing variabilityin menstrual cycle length and hormone pat-terns of the perimenopause. The cases of in-active phases that we have found in ourlaboratory studies and the literature arebroadly consistent with our theoreticalmodel. We are currently collecting the datanecessary to critically test this model: pro-spective daily hormone and bleeding datafrom approximately 120 women aged 30–58years, followed for 6 months of each year for5 years.

If the model is correct, then the process offollicular depletion can, by itself, explainmost of the age-related menstrual cycle fea-tures of the perimenopause and menopause.In particular, the process of follicular deple-tion can account for the age-related increasein the frequency of long menstrual cycles,the increasing variability in menstrual cyclelength with age, the increase in follicularphase length with age, and the higher LHand especially FSH levels observed in peri-menopausal women (Metcalf et al., 1982;MacNaughton et al., 1992; Klein et al.,1996, Santoro et al., 1996). The follicular

depletion model is attractive because it isparsimonious: one biological process can ex-plain many of the salient menstrual cyclefeatures of the perimenopause and meno-pause. In contrast, hypothalamic agingmodels cannot account for all of these vari-ous characteristics of reproductive aging,nor do they specify the mechanism(s) pro-ducing, for example, elevated gonadotropinlevels. Moreover, if an aging hypothalamuswere the primary biological mechanismdriving the breakdown in communicationbetween the ovary and pituitary, we mightexpect a more random pattern of gonadotro-pin secretion throughout the ovarian cycle.

The current formulation of the model doesnot offer an explanation for one other sa-lient feature of the perimenopausal period:the slight increase in shorter than averagemenstrual cycles observed in the years be-fore menopause (e.g., Treloar et al., 1967,see Fig. 1). These short cycles may becaused by an altogether different mecha-nism, related to the higher frequency of an-ovulation seen during the perimenopausalyears (Metcalf, 1983). Further work will benecessary to examine the relationship ofshort cycles with anovulation.

ACKNOWLEDGMENTS

We thank Eleanor Brindle, SusannahBarsom, Kenneth Campbell, Fortune Ko-hen, Bill Lasley, John O’Connor, PhyllisMansfield, and Cheryl Stroud for their as-sistance with the laboratory component ofthis work. Anti-human LHb1 and FSHb an-tisera (AFP #1) and LH (AFP4261A) andFSH (LER907) standards were provided bythe National Hormone and Pituitary Pro-gram through NIDDK, NICHD, and USDA.

LITERATURE CITED

Austad SN. 1997. Postreproductive survival. In:Wachter KW, Finch CE, editors. Between Zeus andthe salmon: the biodemography of longevity. Wash-ington, DC: National Academy Press. p 161–174.

Austin CR, Short RV. 1982. Germ cells and fertilization.Cambridge: Cambridge University Press.

Block E. 1952. Quantitative morphological investiga-tions of the follicular system in women: variations atdifferent ages. Acta Anat 14(Suppl 16):108–123.

Block E. 1953. A quantitative morphological investiga-tion of the follicular system in newborn female in-fants. Acta Anat 17:201–206.

Burger HG. 1996. The endocrinology of the menopause.Maturitas 23:129–136.

Faddy MJ, Gosden RG. 1995. A mathematical model offollicle dynamics in the human ovary. Hum Reprod10:770–775.

Flaws JA, Abbud R, Mann RJ, Nilson JH, Hirshfield


AN. 1997. Chronically elevated luteinizing hormonedepletes primordial follicles in the mouse ovary. BiolReprod 57:1233–1237.

Gilardi KVK, Shideler SE, Valverde CR, Roberts JA,Lasley BL. 1997. Characterization of the onset ofmenopause in the rhesus macaque. Biol Reprod 57:335–340.

Gougeon A. 1996. Regulation of ovarian follicular devel-opment in primates: facts and hypotheses. EndocrineRev 17:121–155.

Gougeon A, Ecochard R, Thalabard JC. 1994. Age-related changes of the population of human ovarianfollicles: increase in the disappearance rate of non-growing and early-growing follicles in aging women.Biol Reprod 50:653–663.

Hee J, MacNaughton J, Bangah M, Burger HG. 1993.Perimenopausal patterns of gonadotropins, immuno-reactive inhibin, oestradiol and progesterone. Matu-ritas 18:9–20.

Hsueh AJW, Billig H, Tsafriri A. 1994. Ovarian follicleatresia: a hormonally controlled apoptotic process.Endocrine Rev 15:707–724.

Kaipia A, Hsueh AJW. 1997. Regulation of follicle atre-sia. Ann Rev Physiol 59:349–363.

Klein NA, Battaglia DE, Clifton DK, Bremner WJ,Soules MR. 1996. The gonadotropin secretion patternin normal women of advanced reproductive age in re-lation to the monotropic FSH rise. J Soc Gynecol In-vest 3:27–32.

Leidy LE, Godfrey LR, Sutherland MR. 1998. Is follic-ular atresia biphasic? Fertil Steril 70:851–859.

MacNaughton J, Banah M, McCloud P, Hee J, BurgerH. 1992. Age related changes in follicle stimulatinghormone, luteinizing hormone, estradiol and immu-noreactive inhibin in women of reproductive age. ClinEndocrinol 36:339–345.

Metcalf MG. 1983. Incidence of ovulation from the men-arche to the menopause: observations of 622 NewZealand women. NZ Med J 96(738):645–648.

Metcalf MG, Donald RQ, Livesey JH. 1982. Pituitary–ovarian function before, during and after the meno-pausal transition: a longitudinal study. Clin Endocri-nol 17:484–489.

McGee EA, Hsueh AJW. 2000. Initial and cyclic recruit-ment of ovarian follicles. Endocrine Rev 21:200–214.

Miller PB, Charleston JS, Battaglia DE, Klein NA,Soules MR. 1999. Morphometric analysis of primor-dial follicle number in pigtailed monkey ovaries: sym-metry and relationship with age. Biol Reprod 61:553–556.

Munro CJ, Stabenfeldt GH, Cragun JR, Addiego LA,Overstreet JW, Lasley BL. 1991. Relationship of se-rum estradiol and progesterone concentrations to theexcretion profiles of their major urinary metabolitesas measured by enzyme immunoassay and radioim-munoassay. Clin Chem 37:838–844.

Norris DO. 1997. Vertebrate endocrinology. San Diego:Academic Press.

O’Connor KA, Holman DJ, Wood JW. 1998. Decliningfecundity and ovarian aging in natural fertility popu-lations. Maturitas 30:127–136.

Richardson SJ, Senikas V, Nelson JF. 1987. Folliculardepletion during the menopausal transition: evidenceof accelerated loss and ultimate exhaustion. J ClinEndocrinol Metab 65:1231–1237.

Santoro N, Brown JR, Adel T, Skurnick JH. 1996. Char-acterization of reproductive hormonal dynamics inthe perimenopause. J Clin Endocrinol Metab 81:1495–1501.

Sherman BM, West JH, Korenman SG. 1976. Themenopausal transition: analysis of LH, FSH, estradioland progesterone concentrations during menstrualcycles of older women. J Clin Endocrinol Metab 42:629–636.

Shideler SE, DeVane GW, Kalra PS, Benirschke K,Lasley BL. 1989. Ovarian–pituitary hormone interac-tions during the perimenopause. Maturitas 11:331–339.

Treloar AE, Boynton RE, Behn BG, Brown BW. 1967.Variation of the human menstrual cycle through re-productive life. Int J Fertil 12:77–126.

van Look PFA, Lothian H, Hunter WM, Michie EA,Baird DT. 1977. Hypothalamic–pituitary–ovarianfunction in perimenopausal women. Clin Endocrinol7:13–31.

vom Saal FS, Finch CE, Nelson JF. 1994. Natural his-tory and mechanisms of reproductive aging in hu-mans, laboratory rodents, and other selected verte-brates. In: Knobil E, Neill JD, Greenwald GS,Markert CL, Pfaff DW, editors. The physiology of re-production, Vol. 2, 2nd edition. New York: RavenPress. p 1213–1314.

Wise PM, Krajnak KM, Kashon ML. 1996. Menopause:the aging of multiple pacemakers. Science 273:67–70.

Wise PM. 1999. Neuroendocrine modulation of the“menopause”: insights into the aging brain. Am JPhysiol 277:E965–E970.


Documents

Menstrual cycle variability and the perimenopause