Early Human Development 73 (2003) 39 52 www.elsevier.

com/locate/earlhumdev

Development of cortisol circadian rhythm in infancy

Carolina de Weerth a,*, Robbert H. Zijl a, Jan K. Buitelaar b
a

Child and Adolescent Psychiatry, University Medical Center of Utrecht, HP A01.468Postbox 85500, 3508 GA Utrecht, The Netherlands b Department of Psychiatry, University Medical Center Nijmegen, HP 333Postbox 9101, 6500 HB Nijmegen, The Netherlands Accepted 1 May 2003

Abstract Background and aims: Cortisol is the final product of the hypothalamus pituitary adrenal (HPA) axis. It is secreted in a pulsatile fashion that displays a circadian rhythm. Infants are born without a circadian rhythm in cortisol and they acquire it during their first year of life. Studies do not agree on the age of appearance of the circadian rhythm (varying between 2 weeks till the age of 9 months) nor on whether it is related to the appearance of the sleep wake circadian rhythm. The object of the present study was to find evidence of the age of appearance of the diurnal rhythm of cortisol and to compare the results obtained by several different analysis methods on a new data set. Design and measures: Cortisol was determined in salival samples of 14 normally developing infants who were followed monthly between the ages of 2 and 5 months. The data were analyzed with several previously published analysis methods as well as with Multilevel Analysis (Hierarchical Linear Modeling). Results: The previously published analysis methods each produced different results when applied to the current data set. Moreover, our results indicate striking differences between young infants in both age of appearance and stability of the diurnal cortisol rhythm. Also, a link was found between the appearance of the sleep wake circadian rhythm and the cortisol circadian rhythm. An important intraindividual variability in cortisol levels was found even after correcting for the different variables that affect cortisol (i.e. time of sampling, feeding, etc.). Conclusions: Although the choice of analysis method influences the age of appearance obtained, our use of HLM shows that the infants own

* Corresponding author. Tel.: +31-30-2509425; fax: +31-30-2505487. E-mail address: C.deWeerth@psych.azu.nl (C. de Weerth). 0378-3782/03/$ - see front matter D 2003 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/S0378-3782(03)00074-4

40

C. de Weerth et al. / Early Human Development 73 (2003) 3952

variability in onset and stability of the cortisol circadian rhythm greatly contributes to the different results. D 2003 Elsevier Ireland Ltd. All rights reserved.
Keywords: Cortisol circadian rhythm; Infancy; HPA

1. Introduction The hormone cortisol is the final product of the hypothalamus pituitary adrenal (HPA) axis in humans. It is a steroid hormone that is considered to be the major indicator of physiological alterations due to stressful stimuli [1,2]. It can be measured in saliva in a reliable, nonintrusive and stress-free way, as saliva flow rate does not affect cortisol levels. Therefore, the assessment of cortisol in saliva has become a popular tool for studying the adrenal cortical function. Cortisol is normally secreted in a pulsatile fashion that displays a circadian rhythm. Highest levels are encountered at early morning and lowest at around midnight. Although there is general agreement over the existence of a circadian rhythm of cortisol in children and adults [3], the age of appearance of this rhythm is still matter of discussion. Infants are born without a circadian rhythm in cortisol and they acquire it during their first year of life. Studies do not agree on the age of appearance of the circadian rhythm: by the age of 3 months [4,5], between 2 and 3 months [6], between 2 and 20 weeks of age [7,8], by 6 months [9,10] and not before 9 months [11]. Also, de Weerth and van Geert [12] found the diurnal circadian rhythm to gain in strength and become more adult-like between the ages of 5 and 8 months. Studies also differ in their conclusions on whether a link exists between the development of the cortisol circadian rhythm and the sleep wake circadian rhythm. While Price et al. [4], Antonini et al. [8] and Spangler [5] report finding a parallelism between the appearance of a sleep wake cycle characterized by uninterrupted night sleep and the appearance of the cortisol circadian rhythm, Santiago et al. [7] found no relation between the two. The lack of consensus from these studies is probably at least partly due to differences in (1) the working definition of circadian rhythm, (2) sampling times and frequencies and (3) methods of analysis of the data. In view of these differences in methodology, it would be helpful to have a study in which methodologies are compared by applying them to the same data set. In the present study, healthy, normally developing infants were followed between the ages of 2 and 5 months, with the object of finding evidence of the age of appearance of the cortisol diurnal circadian rhythm. In order to compare our results with those of other researchers, the data were analyzed with the previously published methods of Price et al. [4], Krieger et al. [13], Santiago et al. [7] and Antonini et al. [8]. We also analyzed the data with Multilevel Analysis or Hierarchical Linear Modeling [14,15]. This technique yields a more flexible way for analyzing repeated measures than the traditional repeated measure ANOVA or MANOVA. It also has the advantage of accepting missing data in the dependent variable without problem (i.e. the data set does not need to have the same

C. de Weerth et al. / Early Human Development 73 (2003) 3952

41

assessment moments for all subjects). The object of using this fourth more sophisticated analysis method is to give weight and conclusiveness to our findings on the age of appearance of the circadian rhythm and to investigate whether it is related to the appearance of the sleep wake circadian rhythm.

2. Method 2.1. Subjects Fourteen infants (six females and eight males) participated in the study. They were healthy full-term infants, from uneventful pregnancies, with normal Apgar scores and a mean birth weight of 3307 g (S.D. = 425 g). Two deliveries had been cesarean (one elective and one emergency) and 12 were vaginal, and the infants were developing well at the time of the study. Mean age of the mothers was 30.8 years at birth (S.D. = 2.37). Most of the infants were firstborns: 79%, while 21% were second children. All the infants belonged to two parent families. The mothers of the infants participated voluntarily in the study and were contacted through the University Hospital, well-baby clinics and information leaflets in shops, pharmacies, etc. The study was approved by the Medical Ethical Committee of our hospital and the mothers gave informed consent to the work. 2.2. Procedure The mothers were visited at their homes before the infant reached the age of 2 months. The objectives of the study and the sampling procedure were explained in detail. A folder about the study was given to the mother, and information about the pregnancy and delivery was obtained. The mothers were asked to fill in a short form of the babys activities on the assessment days and to register the age at which their infants slept through the night (defined as a sleeping period of at least 6 h after the last feeding session of the day). The saliva sampling procedure was also explained and demonstrated. The mothers were instructed to sample their babys saliva five times during 1 day (at 0800, 1100, 1400, 1700 and 2000 h) at the ages of 2, 3, 4 and 5 months. Because the mothers did not always collect saliva at the exact required time, the following time intervals were used for analysis: 0700 1000, 1000 1300, 1300 1600, 1600 1900 and 1900 0015 h. The saliva was collected by means of special cotton swabs (Salivettes, Sarstedt, Germany), if necessary after cleaning the buccal cavity to avoid food or milk contamination of the sample [16]. The mother was given citric acid crystals to put on the infants tongue if she felt the mouth to be too dry. Citric acid does not interfere with the cortisol determination method used in this study. The salivettes were stored frozen until the researcher collected them at the end of the period. At the Endocrinology Laboratorium, the salivettes were centrifuged and cortisol was determined by radiommunoassay (RIA) [17] in samples containing at least 100 Al of saliva. All samples from the same individual were measured in the same assay. The

42

C. de Weerth et al. / Early Human Development 73 (2003) 3952

inter-assay coefficient of variation was 10.0%, 6.4% and 6.0% for, respectively, 5.1, 11.9 and 20.6 nmol/l. The intra-assay coefficient of variation was 4% for all three levels. 2.3. Sleep 2.3.1. Naps Larson et al. [18] found morning naps to be associated with decreases in cortisol in 9month-old infants with returns to pre-nap levels 45 min after waking up. de Weerth and van Geert [12] also found that in 5- to 8-month-old cortisol levels were lower when the infant had just slept than when he/she had been awake for longer than half an hour. Therefore, the mothers in this study were asked to register whether the infant had awoken from a nap during the half-hour preceding the saliva assessment. This was the case in 73 out of 221 saliva assessments, which amounts to 33% of the samples. A dummy variable was created, with 0 meaning that the infant had not just awoken from a nap and 1 meaning that the infant had just awoken from a nap. 2.3.2. Sleeping through the night The mothers were asked to register from which month on their infant slept through the night. This was defined as regularly sleeping at least 6 h after the last feeding of the day, without waking up for a feeding during this interval of time. Defined in this way, six infants slept through the night from the age of 2 months on, two infants from 3 months on, three infants from 5 months on and one infant did not yet sleep through the night at the age of 5 months. A dummy variable was created with 0s for all the cortisol assessments taken in months in which the infant did not sleep through the night and 1s for all the assessments of months in which the infant did sleep through the night. 2.4. Feeding Reports of how feeding affects cortisol levels are conflicting. While some authors report no effects of feeding on cortisol [19], others report different effects according to the type of feeding, namely liquid or solid [20]. Solid feedings appear to decrease cortisol levels immediately after a feed and increase them a short time afterwards [5,21]. In order to control for possible effects of feeding on cortisol, the mothers in the present project were asked to register whether the infant had eaten in the hour previous to the saliva sampling. Out of the 221 cortisol assessments used in the analyses, in 32 cases (14.5%), the infant had had a liquid feeding in the half-hour preceding the saliva assessment, and in 7 cases (3.2%), the infant had had a solid feeding in the hour preceding the assessment. A 0 1 variable feeding liquid was created in which a 1 was given to assessments linked to liquid feedings and a 0 to assessments not linked to liquid feedings in the manner described above. A 0 1 variable feeding solid was also created, in which a 1 was given to assessments linked to solid feedings and a 0 to assessments not linked to solid feedings in the manner described above.

C. de Weerth et al. / Early Human Development 73 (2003) 3952

43

2.5. Gender Gender was scored in a 1 2 variable in which 1 was given to male infants and 2 to female infants. 2.6. Missing data The mother of one infant did not take samples at 2 months, that of another did not take samples at 3 and 5 months and the mother of a third collected unusable data on months 4 and 5 due to corticosteroid medication. Of the 255 samples left, 34 (13.3%) were either forgotten by the mothers or contained too small amounts of saliva for analysis. Thus, 221 samples could be included in the analyses. 2.7. Statistical analysis Multilevel Analysis calculations (Hierarchical Linear Modeling) were carried out with the aid of MLWiN [22]. Models were compared by the likelihood ratio v2 test, calculated by subtracting deviances (defined as minus twice the log-likelihood). A three-level model was used, in which the subjects constituted the level 3 units, the months the level 2 units and the assessment times the level 1 units. Because cortisol distributions are often skewed, the data were normalized by calculating the logarithm.

3. Results Table 1 shows the mean cortisol values, the standard deviations and the number of infants sampled at each age for the time intervals around the sampling times. As can be seen from the table, a tendency to display a circadian rhythm in cortisol is apparent for the group as a whole from the age of 2 months on. However, as the analyses presented below reveal, closer inspection of the data discovered important inter- and intraindividual differences. 3.1. Analyses using previously published methods As an example of the data, Fig. 1a d shows the cortisol assessments for four different infants. The figure illustrates the important intra- and interindividual variability that exists
Table 1 Means of diurnal salivary cortisol levels of healthy infants sampled five times per day Time period 2 months Mean 0700 1000 1000 1300 1300 1600 1600 1900 1900 0015 h h h h h 9.5 9.0 7.1 4.1 4.6 S.D. 6.8 5.3 3.3 1.3 3.9 n 12 10 10 11 10 3 months Mean 10.1 8.5 8.4 6.0 3.6 S.D. 7.4 3.8 4.8 2.9 3.2 n 11 12 11 9 14 4 months Mean 13.3 9.9 5.8 5.5 3.1 S.D. 6.3 7.3 3.9 4.5 3.0 n 11 13 10 9 13 5 months Mean 11.0 7.1 7.3 5.1 2.8 S.D. 7.8 4.1 2.9 4.1 1.9 n 11 10 11 12 11

44 C. de Weerth et al. / Early Human Development 73 (2003) 3952

Fig. 1. (a d) Examples of cortisol levels of four different infants.

C. de Weerth et al. / Early Human Development 73 (2003) 3952

45

in the development of the cortisol circadian rhythm. In the following paragraphs, the results of the analyses of the data using three previously published method will underline this first visual impression. For all three analyses, morning values were defined as those samples taken between 0700 and 0915 h. Price et al. [4] describe the appearance of the circadian rhythm by the age of 3 months. They define the circadian rhythm as a higher value in the morning (0600 0800 h) than in the evening (2200 2400 h), with a steady decline throughout the day. When this pattern remains, the infant is considered to have developed a circadian rhythm. We applied Price et al.s [4] visual method to our data by means of the following rules: An infants month data was scored if the morning and late evening assessments were present together with at least one assessment in between. The circadian rhythm was considered to be present if the evening value was at least 2 nmol/l smaller than the morning value (implementation of Price et al.s . . .higher value in the morning than in the evening. . ., p. 456) and all values after the morning value were not more than 2 nmol/l larger than their immediate preceding value (implementation of Price et al.s . . .steady decline throughout the day. . ., p. 456). The results of this analysis are presented in Table 2a, where YES denotes the presence of the cortisol circadian rhythm and NO the absence of this rhythm. Santiago et al. [7] and Antonini et al. [8] defined a normal circadian pattern as one in which both afternoon and night values were less than 83.5% (100% 3 mean intraassay CV) of the morning value for each subject. Thus, they determined that the circadian rhythm emerged and persisted between the ages of 2 and 20 weeks for the individual infants (for the groups as a whole at 8 weeks). In this study, the intra-assay CV was 4%, which meant that the afternoon and evening values should be less than 88% of the morning value. An infants month data were scored if the morning and both 1700- and 2000-h assessments were present. A YES for the presence of the circadian rhythm was scored when both evening values were less than 88% of the morning value (see Table 2b) and a NO was scored when this was not the case. Finally, Krieger et al. [13] defined a normal circadian pattern of cortisol as one in which all values after 0800 h were less than 75% of the 0800-h level. In the present study, an infants month data was scored if cortisol had been assessed at least twice after the morning assessment. A YES was scored when all samples after the morning sample were lower than 75% of the morning sample, and a NO was scored when this was not the case (see Table 2c). As can be seen from Table 2, the application of the different methods of analysis produced results that differed importantly from one another. Also, infants showed great interindividual variation in patterns of development of the circadian rhythm. The subjects were also variable intraindividually; in other words, the circadian pattern of cortisol did not always persist after emerging. 3.2. Analyses using multilevel approach The infant cortisol data were modeled in search of the multilevel model that best fitted the data. The empty model for the infant cortisol had a deviance of 197.089. The empty model is the simplest case of the hierarchical linear model, in which the explanatory variables do not figure. It only contains random groups and random variation within

46

C. de Weerth et al. / Early Human Development 73 (2003) 3952

Table 2 (a c) Results of analyses with the methods of Price et al. [4] (a), Santiago et al. [7] and Antonini et al. [8] (b) and Krieger et al. [13] (c) Infant (a) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 (b) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 (c) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 2 months NO NO NO NO YES NO NO NO YES NO 3 months YES NO NO YES NO YES NO YES YES NO 4 months YES YES NO YES NO YES NO NO YES 5 months YES YES NO YES YES YES NO NO YES NO

YES NO NO YES YES NO NO YES YES

YES YES NO YES YES YES NO YES YES NO

YES YES YES YES YES NO YES YES

YES YES NO YES YES YES NO YES YES YES

NO NO NO YES NO YES NO NO NO YES NO

NO NO NO YES NO NO NO NO NO NO

NO NO YES NO NO NO NO YES NO YES YES

YES YES NO YES YES YES NO NO NO NO

YES means that according to the method, the circadian rhythm of cortisol is present, NO means that it is not present. Hyphens represent cases that due to missing data could not be scored using a particular method.

C. de Weerth et al. / Early Human Development 73 (2003) 3952

47

groups. The best model that could be found for the infant cortisol had a deviance of 67.632 and is shown in Table 3. The reduction in deviance was obtained by subsequently adding the different variables that are now part of the model. Their significance and the building of the model will be explained in the following paragraphs. The variable time of sampling was centered by subtracting the mean from each value. The introduction of this variable into the fixed part of the model resulted in a significant reduction of the deviance (deviance = 107.670, v2 = 89.42, df = 1, p < 0.001). The negative effect of the time of sampling means that cortisol samples taken earlier in the day tended to have higher values. When the squared variable time of sampling was introduced into the fixed part of the model, the deviance was further reduced (deviance = 98.460, v2 = 9.2, df = 1, p < 0.01). This means that the reduction in cortisol levels over the day followed a quadratic trajectory. These results confirm the effects of the circadian rhythm on basal cortisol for the group as a whole and without taking age (i.e. month) into account. The variable month was also centered by subtracting the mean from each value. The next step in the analysis was to introduce the centered month into the fixed part of the model. This resulted in a nonsignificant reduction of the deviance (deviance = 97.73, v2 = 0.73, df = 1, NS) showing that the infants did not show general positive or negative trends in their cortisol values between 2 and 5 months. Next, a variable representing the interaction between month and time of sampling was obtained by multiplying both centered variables. The new variable was introduced into the model, first only into the fixed part and then into both the fixed part and the random part of the model at the individual level (level 3). In both models, month was kept in the fixed part of the model. The deviance decreased nonsignificantly in both cases (respectively,
Table 3 Estimates for the best multilevel model of the infant cortisol data Fixed effects Intercept Time of sampling Squared time of sampling Month Sleeping through night time of sampling Feeding solid Random effects Level 3 (i.e. individual) variance Intercept variance Month variance Intercept month covariance Level 2 (i.e. month) variance Intercept variance Time of sampling variance Intercept time of sampling covariance Level 1 (i.e. occasion) variance Residual variance Deviance Coefficient 0.744 0.027 0.002 0.045 0.025 0.342 Parameter 0.014 0.013 0.013 0.026 0.0004 0.0004 0.050 67.632 S.E. 0.046 0.009 0.001 0.040 0.010 0.107 S.E. 0.010 0.009 0.007 0.011 0.0002 0.001 0.006

48

C. de Weerth et al. / Early Human Development 73 (2003) 3952

deviance = 94.322, v2 = 4.14, df = 2, p = 0.13 and deviance = 92.541, v2 = 5.92, df = 4, p = 0.20). Although a significant negative interaction effect between the time of sampling and the month (t = 2.33, df>200, one-sided p = 0.01) could be found, because the deviance does not decrease significantly ( p = 0.13), it is at this point not possible to state that the effect of the time of sampling on the cortisol level depended on the age of the infant. In other words, it is not possible to say that the effects of the circadian rhythm increased with age between the ages of 2 and 5 months. Because the second model is nonsignificant ( p = 0.20), the results also mean that it is not possible to state that the different subjects showed effects of the circadian rhythm of cortisol at different ages. The following step in the modeling was to introduce a random effect of month at the individual level (level 3) and a random effect of time of sampling at the month level (level 2). In order to be able to do this, month was also introduced into the fixed part of the model. The deviance decreased significantly in this new model (deviance = 82.432, v2 = 16.03, df = 5, p = 0.01). Next, feeding solid was introduced into the model. The deviance again decreased significantly (deviance = 73.043, v2 = 9.39, df = 1, p < 0.01). The positive effect of feeding can be interpreted straightforwardly; infants that had had a solid meal within 1 h of the saliva sampling had higher cortisol levels than those that had not. A final reduction in deviance was obtained by introducing the interaction variable sleeping through night time of sampling into the fixed part of the model. This variable was obtained by multiplying sleeping through night and time of sampling. The decrease in deviance was significant (deviance = 67.632, v2 = 5.41, df = 1, p = 0.02). The interaction variable had a significant negative effect on the cortisol levels, which means that infants that slept through the night showed greater circadian rhythm effects on their cortisol levels. Since the predictor variables month and time of sampling were centered, the intercept variances of levels 3 and 2 have a clear meaning: they represent the amount of variation in cortisol levels across individuals and across months, respectively, for the average occasion while correcting for circadian rhythm effects. As can be seen in Table 3, although these effects account for some variation at the individual and month levels, the greatest random variation can be found at the occasion level (variation coefficient = 0.050). This means that even after a fair amount of variance in cortisol levels has been accounted for by different variables in the fixed and random parts of the model, an important variability in cortisol levels can still be found between assessment occasions. In other words, the average occasion displays important random variability in the levels of the hormone cortisol. Finally, the introduction of the following variables into the fixed part of the model did not further reduce the variance: feeding liquid (deviance = 66.091, v2 = 1.54, df = 1, p = 0.21), gender (deviance = 67.551, NS), nap (deviance = 67.555, NS) and sleeping through the night (deviance = 67.477, NS). Thus, these variables had no general effect on the cortisol levels.

4. Discussion The different data analyses presented in this paper suggest that while at group level there is evidence for the presence of a circadian rhythm of cortisol from the early age of 2

C. de Weerth et al. / Early Human Development 73 (2003) 3952

49

months, individuals can vary greatly in the age at which they acquire the rhythm. There are also important differences between individuals in how persistent or stable the rhythm is once it appears. Moreover, as has been reported before for slightly older infants (12), infant cortisol levels can also be subject to important day-to-day variability that would make it difficult to assume that the variable curves found in the present population are stable on a weekly or even daily basis. As shown by the analyses, the different working definitions of circadian rhythm and the differences in methods of analysis of the data contribute greatly to obtaining varying conclusions with respect to the age of appearance of the circadian rhythm. The methods used in previously published articles produced different results when applied to the current data set. More importantly, however, is to stress the striking individual differences that were found in the development of the diurnal circadian rhythm of cortisol. The differences between individuals could be due to underlying physiological and/or behavioral differences. White et al. [23] for example found that infants with colic had a less clearly defined cortisol circadian rhythm than infants without colic. The authors suggest that infants with colic might have a delay in the establishment of the cortisol circadian rhythm. Also, in studies in the rat [24], the age of appearance of the adrenocortical rhythm has been found to be influenced by early (stressful) experiences in the infant rat. It is therefore possible that age of appearance (i.e. early versus late), stability, curve form, etc., are characteristic of developmentally distinct groups of infants. However, as in other developmental features, a certain degree of variability in the appearance, age and stability of the circadian rhythm could also be a characteristic of normal development in healthy infants. Furthermore, de Weerth and van Geert [12] have discussed the possibility that the important intraindividual variability that healthy 5- to 8month-old display in cortisol levels obscures the effects of the circadian rhythm between morning and afternoon hours. In the present study, an important intraindividual variability in cortisol levels was found even after correcting for the different variables that affect cortisol (i.e. time of sampling, feeding, etc.), making it plausible for this variability to be characteristic of normal development also in infants between 2 and 5 months. A large intraindividual variability in normally developing infants has also been found in past studies with respect to emotion-related behaviors such as crying and fussing [25 27]. In sum, the individual differences in age of appearance together with the intraindividual variability in cortisol levels would account for the variation in results found in the present and earlier studies. Both would be features of healthy development. Infants that had had a solid meal in the hour preceding the saliva sample tended to display higher cortisol levels than those that had not, probably due to the postprandial surge in cortisol [21]. On the other hand, liquid feedings were found to have no effect on the cortisol levels, confirming earlier findings [20]. Although a negative effect of sleep on cortisol levels was expected [12,18], in the present study naps did not have any effect on cortisol. Perhaps this is due to age differences between the studies (5 9 months versus 2 5 months). Finally, the data analysis showed that infants that slept through the night showed greater circadian rhythm effects on their cortisol levels. This suggests a possible synchronicity between the establishment of the sleep wake circadian rhythm and the cortisol circadian rhythm, confirming earlier reports by other authors [4,5,8].

50

C. de Weerth et al. / Early Human Development 73 (2003) 3952

Although our study design was in line with earlier studies on infant cortisol circadian rhythmicity [4 8,11,23] and with those of many leading researchers on cortisol [28 31], the methodology used does have important shortcomings. More sophisticated studies on hormonal circadian rhythmicities and sleep, such as those by van Cauter et al. [32,34] and Linkowski et al. [33], use higher (serum) sampling frequencies (i.e. at 15-min intervals) that go on through the night and eventually during several days, combined with continuous EEG monitoring of sleep. This type of data gives more precise insight into circadian cycles. Despite the obvious hurdles of applying similar methodologies in infancy studies, given our results, we feel that future research would greatly benefit by, for example, increasing the saliva-sampling frequency, by sampling through the night and by sampling on consecutive days at different ages. Our findings imply that clinicians and researchers should not take the diurnal circadian rhythm of cortisol for granted at certain ages during the first half-year of life. Individual infants vary greatly in the age of appearance and stability of the cortisol circadian pattern. This makes generalizations unnecessary and even unreliable. During the second half-year of life, the circadian rhythm probably consolidates, although even then infants display great intraindividual variability in cortisol levels that defy the expected decline during the day [12]. It would be very interesting to investigate when this variability decreases and children acquire a more adult-like stability in cortisol levels and circadian patterns, whether certain infants do not follow this apparently normal developmental path and what the consequences are of such abnormal developmental paths. Acknowledgements The authors wish to thank the families that participated in the study, and also Tom Snijders and Paul Westers for their advice on the statistical analysis of the data. This research was supported by the Netherlands Organization for Scientific Research (NWO), grant number 575-25-009. References
[1] Kirschbaum C, Hellhammer DH. Salivary cortisol in psychobiological research: an overview. Neuropsychobiology 1989;22:150 69. [2] Kirschbaum C, Hellhammer DH. Salivary cortisol in psychoneuroendocrine research: recent developments and applications. Psychoneuroendocrinology 1994;19:313 33. [3] Van Cauter E, Turek FW. Endocrine and other biological rhythms. In: De Groot LJ, editor. Endocrinology. Philadelphia: Saunders; 1995. p. 2487 548. [4] Price DA, Close GC, Fielding BA. Age of appearance of circadian rhythm in salivary cortisol values in infancy. Arch Dis Child 1983;58:454 6. [5] Spangler G. The emergence of adrenocortical circadian function in newborns and infants and its relationship to sleep, feeding and maternal adrenocortical activity. Early Hum Dev 1991;25:197 208. [6] Mantagos S, Moustogiannis A, Vagenakis AG. Diurnal variation of plasma cortisol levels in infancy. J Pediatr Endocrinol Metab 1998;11:549 53. [7] Santiago LB, Jorge SM, Moreira AC. Longitudinal evaluation of the development of salivary cortisol circadian rhythm in infancy. Clin Endocrinol 1996;44:157 61.

C. de Weerth et al. / Early Human Development 73 (2003) 3952

51

[8] Antonini SRR, Jorge SM, Moreira AC. The emergence of salivary cortisol circadian rhythm and its relationship to sleep activity in preterm infants. Clin Endocrinol 2000;52:423 6. [9] Onishi S, Miyazawa G, Nishimura Y, Sugiyama S, Yamakawa T, Inagaki H, et al. Postnatal development of circadian rhythm in serum cortisol levels in children. Pediatrics 1983;72:399 404. [10] Lewis M, Ramsay DS. Developmental change in infants responses to stress. Child Dev 1995;66:657 70. [11] Kiess W, Meidert A, Dressendo rfer RA, Schriever K, Kessler U, Ko nig A, et al. Salivary cortisol levels throughout childhood and adolescence: relation with age, pubertal stage, and weight. Pediatr Res 1995;37:502 6. [12] de Weerth C, van Geert P. A longitudinal study of basal cortisol in infants: intra-individual variability, circadian rhythm and developmental trends. Infant Behav Dev 2002;25:375 98. [13] Krieger DT, Allen W, Rizzo F, Krieger HP. Characterization of the normal temporal pattern of plasma corticosteroid levels. J Clin Endocrinol Metab 1971;32:266 84. [14] Bryk AS, Raudenbush SW. Hierarchical linear models, applications and data analysis methods. Newbury Park, CA: Sage Publications; 1992. [15] Snijders T, Bosker R. Multilevel Analysis. An introduction to basic and advanced multilevel modeling. London: Sage Publications; 1999. [16] Magnano C, Diamond E, Gardner J. Use of salivary cortisol measurements in young infants: a note of caution. Child Dev 1989;60:1099 101. [17] Thijssen JHH, Van den Berg JHM, Adlercreutz H, Gijzen AHJ, De Jong FH, Meijer JC, et al. The determination of cortisol in human plasma: evaluation and comparison of seven assays. Clin Chim Acta 1980;100:39 46. [18] Larson MC, Gunnar MR, Hertsgaard L. The effects of morning naps, car trips, and maternal separation on adrenocortical activity in human infants. Child Dev 1991;62:362 72. [19] Lobo ML. Stress in infancy. In: Arnold LE, editor. Childhood stress. New York: Wiley; 1990. p. 173 92. [20] Hertsgaard L, Gunnar M, Larson M, Brodersen L, Lehman H. First time experiences in infancy: when they appear to be pleasant, do they activate the adrenocortical stress response? Dev Psychobiol 1992;25:319 33. [21] Riad-Fahmy D, Read GF, Hughes LA. Corticosteroids. In: Gray CH, James VHT, editors. Hormones in blood, vol. 4. New York: Academic; 1983. p. 285 315. [22] Goldstein H, Rasbash J, Plewis I, Draper D, Browne W, Yang M, et al. A users guide to MLWiN. London: Multilevel Models Project, Institute of Education, University of London; 1998. [23] White BP, Gunnar MR, Larson MC, Donzella B, Barr RG. Behavioral and physiological responsivity, sleep, and patterns of daily cortisol production in infants with and without colic. Child Dev 2000;71:862 77. [24] Ader R. Neonatal stimulation and maturation of the 24-hour adrenocortical rhythm. In: Gispen WH, van Wimersma-Greidanus TB, de Wied D, editors. Hormones, homeostasis and the brain. Progress in brain research, vol. 42. Amsterdam: Elsevier; 1975. p. 333 41. [25] de Weerth C. Emotion-related behaviors in infants. A longitudinal study of patterns and variability. PhD Thesis. Groningen: The Netherlands; 1998. [26] de Weerth C, van Geert P, Hoijtink H. Intra-individual variability in infant behavior. Dev Psychol 1999;35:1102 12. [27] de Weerth C, van Geert P. The dynamics of emotion-related behaviors in infancy. In: Lewis MD, Granic I, editors. Emotion, development, and self-organization. Dynamic systems approaches to emotional development. New York: Cambridge Univ. Press; 2000. p. 324 48. [28] Adam EK, Gunnar MR. Relationship functioning and home and work demands predict individual differences in diurnal cortisol patterns in women. Psychoneuroendocrinology 2001;26:189 208. [29] Wust S, Federenko I, Hellhammer DH, Kirschbaum C. Genetic factors, perceived chronic stress, and the free cortisol response to awakening. Psychoneuroendocrinology 2000;25:707 20. [30] Stone AA, Schwartz JE, Smyth J, Kirschbaum C, Cohen S, Hellhammer D, et al. Individual differences in the diurnal cycle of salivary free cortisol: a replication of flattened cycles for some individuals. Psychoneuroendocrinology 2001;26:295 306. [31] Kudielka BM, Broderick JE, Kirschbaum C. Compliance with saliva sampling protocols: electronic monitoring reveals invalid cortisol daytime profiles in noncompliant subjects. Psychosom Med 2003;65:313 9. [32] Van Cauter E, Leproult R, Kupfer DJ. Effects of gender and age on the levels and circadian rhythmicity of plasma cortisol. J Clin Endocrinol Metab 1996;81:2468 73.

52

C. de Weerth et al. / Early Human Development 73 (2003) 3952

[33] Linkowski P, Van Onderbergen A, Kerkhofs M, Bosson D, Mendlewicz J, Van Cauter E. Twin study of the 24-h cortisol profile: evidence for genetic control of the human circadian clock. Am J Physiol 1993;264: E173 81. [34] Van Cauter E. Diurnal and ultradian rhythms in human endocrine function: a minireview. Horm Res 1990;34:45 53.