Measurement and meaning of salivary cortisol: A focus on health and disease in children

Abstract

Measurement of salivary cortisol can provide important information about hypothalamic-pituitary-adrenal (HPA) axis activity under normal conditions and in response to stress. However, there are many variables relating to the measurement of cortisol in saliva which may introduce error and therefore may render difficult the comparison and interpretation of data between, and within, laboratories. This review addresses the effects of gender, age, time and location of sampling, units of measurement, assay conditions and compliance with the protocol, all of which have the potential to impact upon the precision, accuracy and reliability of salivary cortisol measurements in the literature. Some of these factors are applicable to both adults and children, but the measurement of salivary cortisol in children introduces aspects of unique variability which demand special attention. The specific focus of this review is upon the somewhat neglected area of methodological variability of salivary cortisol measurement in children. In addition to these methodological issues, the review highlights the use of salivary cortisol measurements to provide information about HPA axis dysfunction associated with psycho- and patho-physiological conditions in children. Novel applications for salivary cortisol measurements in future research into HPA axis activity in children are also discussed.

• Salivary cortisol
• children
• HPA axis
• awakening cortisol rise (ACR)

Introduction

The hypothalamic-pituitary-adrenal (HPA) axis is a major homeostatic system which under basal conditions maintains a circadian rhythm and is activated in response to stress. Stressors may include cognitive (e.g. excitement, fear, danger) or non-cognitive (e.g. infection) challenges. The endpoint of HPA axis activation in humans is the release of the glucocorticosteroid cortisol from the adrenal glands into the blood to exert multiple metabolic effects (Jacobson 2005). Integrity of the HPA axis is essential to life, and compromised performance through inappropriately low (e.g. Addison's disease) or high (e.g. Cushing's syndrome, chronic stress) production of cortisol can be life-threatening. A robust increase in cortisol secretion is necessary to determine appropriate behavioural reactions in response to the acute stress of psychological or physical threat, and resetting of the HPA axis with increased cortisol secretion is an important adaptive response influencing the onset, development and severity of inflammatory disease.

Measurement of salivary cortisol has provided a great deal of experimental information on activity of the HPA axis in health and disease and on the involvement of cortisol in the mechanisms determining our reactivity and adaptation to environmental changes. Salivary sampling is a well-established technique for cortisol measurement in adults (Al-Ansari et al. 1982) and children (Woolston et al. 1983). A strongly positive correlation between salivary and blood cortisol levels has been reported (Bober et al. 1988; Woodside et al. 1991; Gallagher et al. 2006) while measurement of salivary cortisol is more physiologically relevant since most of the analyte is in the free form and not bound to corticosteroid binding globulin (CBG) as it is in blood. A relatively non-invasive technique, salivary sampling has the dual advantage over blood sampling of (i) permitting cortisol to be measured without contamination of results by needle stress, and (ii) extending investigation beyond experimental laboratory studies through the ease of home sampling. These advantages may further serve to increase participation rates in studies from those who might be reluctant to provide blood samples, a particularly relevant issue in the study of children. Ethical issues are also substantially reduced by sampling saliva rather than blood.

However, anomalies, inconsistencies and contradictions are embedded throughout the literature of salivary cortisol measurements, many causing confusion, some leading to error. For example, wide variations in basal morning cortisol levels have been reported, many different units are employed to express cortisol measurement, and comparison of data between laboratories is often not possible for technical and methodological reasons. Not all of these issues are specific to juvenile studies, but there are sufficient problems in the literature relating to the measurement of salivary cortisol in children to justify this as the subject of a review in its own right. In addition to methodological issues, child studies have their own special set of independent variables which may influence cortisol measurements. Most studies are based on a pooled sample of children of mixed gender and age and few studies with sufficient statistical power have attempted a systematic analysis of the effects of these crucial variables on cortisol secretory activity. It is therefore valid to consider these problematic issues from a child-specific perspective because of the increasing number of studies on HPA axis activity in children, and the importance of these in providing insight into how children may anticipate and react to changes in their environment in ways which may shape their future health and behaviour as adults.

Measurement of salivary cortisol in children has been standard practice for more than 20 years, and major reviews have appeared on the subject (Gunnar et al. 2001a; Gunnar and Donzella 2002). We have referred to papers that have mostly been published since these reviews to highlight the benefits and problematic aspects of this methodology. We have not attempted to cite all papers published since these reviews, but instead we have focused on a selection of papers which best illustrates the points which we wish to emphasise. The purpose of this review is not to cover technical aspects of saliva collection methods, an area recently reviewed (Talge et al. 2005; Hanrahan et al. 2006). Rather it is our intention to focus on effects of gender, age, body weight, time and location of sampling, assay conditions, and compliance with the protocol which can compromise the precision and accuracy of salivary cortisol measurements in children and affect consequent interpretation of the data. Having discussed ways in which these variables may be more clearly defined, or in the case of artifacts, minimised or eliminated, we also highlight psycho- and patho-physiological conditions where reliable and accurate salivary cortisol measurements may provide important information about HPA axis dysfunction in children, and wherein may lie potential for therapeutic intervention. Finally, we provide some suggestions and applications for the future direction of research into HPA axis activity in children.

Factors introducing potential for variability in salivary cortisol measurement and analysis

It is apparent from the literature (Table I) that wide variations exist between laboratories in morning salivary cortisol values reported in healthy children, which in many studies are considerably greater than the normal range generally accepted for morning serum cortisol (300–600 nmol/l). This variability, which is greater than the usual physiological variation expected within and between individuals, has caused difficulty in establishing a normal range for basal salivary cortisol, and may be due to independent variables such as age, gender, weight or genotype, and also to confounding methodological factors such as sampling and assay conditions, units of cortisol measurement and compliance.

Table I.  Selection of recent studies reporting morning salivary cortisol in children to show the wide variety of units of measurement and the wide range of values. One column lists levels of cortisol in the authors' choice of units; in a second column these have been converted into SI units of nmol/l. Conversion factors: μg/dl × 10 = ng/ml × 2.78 = nmol/l. Waking/30 min = single sample taken within 30 min of waking.

Mean age (or range) in years	Sample size	Sampling time	Cortisol in units provided (range, mean or median)	Cortisol (nmol/l)	Reference
(7–9)	68 boys	Waking	7–13 nmol/l	7–13
	72 girls		8–14 nmol/l	8–14	Jones et al. (2006)
11	109	0700–0800 h	0.48 μg/dl	13.3	Hessl et al. (2002)
11 (6–17)	90	Waking/30 min	0.14 μg/dl	3.9	Hessl et al. (2006)
(6–10)	217	0800–0900 h	0.5–0.9 μg/dl	13.9–25.0	Lupien et al. (2000)
11	178	0900 h	290 ng/dl	8.1	Moss et al. (1999)
(10–12)	262	0900 h	289 ng/dl	8.0	Hardie et al. (2002)
(10–12)	1768	0730 h	15.38 nmol/l	15.38	Rosmalen et al. (2005)
12.8 (8–16)	126	0800 h	3.30 ng/ml	9.17	Netherton et al. (2004)
(8–11)	340	Morning	7.5–8.1 nmol/l	7.5–8.1	Haines et al. (2001)
(8–10)	21	0900 h	291 nmol/l	291	Hershberger et al. (2004)
(7–15)	210 boys	0800–0900 h	8.8 nmol/l (highest value 53.9)	8.8 (53.9)
	176 girls		8.6 nmol/l (highest value 33.2)	8.6 (33.2)	Tornhage (2002)
4.5	56	Waking	2.88 ng/ml	8.0	Chryssanthopoulou et al. (2005)
(5–14)	1152	Waking/30 min	13.4 nmol/l for older children (highest value 231.2)	13.4 (231.2)
			14 nmol/l for younger children (highest value 246)	14 (246)	Koupil et al. (2005)
8.55 (Romanian)	18	Early morning	0.8 μg/dl	22.24
7.9 (Canadian)	27		0.68 μg/dl	18.9	Gunnar et al. (2001b)
9.4	12	Mid-morning	60–70 nmol/l	60–70	Jansen et al. (2003)
(11–17)	71	Early morning	0.4 mg/dl	11,120	Klimes-Dougan et al. (2001)
(6–7)	34	Waking/30 min	0.71 μg/dl	19.74	Bruce et al. (2002)

Gender

HPA axis activity in adults, basal and stress-stimulated, has pronounced and well-documented gender dimorphism, a phenomenon that has been linked to differences in circulating peptide and steroid hormones (Vamvakopoulos and Chrousos 1993; Young and Altemus 2004; Kudielka and Kirschbaum 2005; Kajantie and Phillips 2006). However, it is unclear whether HPA axis activity is gender-specific in children. One study in which test and control subjects were pooled demonstrated significantly increased morning salivary cortisol levels in girls compared to boys in the age range 10–12 years (Hardie et al. 2002), but there was a considerable disparity in group sizes (262 boys and 29 girls) which may have introduced bias into the statistical analysis. A similar finding was also reported in the large study of 1768 children by Sondeijker et al. (2007) in the same age group but only in morning, not evening, samples. This study also found a link between gender differences in cortisol and attention deficit/hyperactivity. Netherton et al. (2004) in a study of 8–16 year olds (60 boys, 69 girls) found morning (but not evening) salivary cortisol levels to be 20–30% higher in mid- to post-pubertal (MPP) girls compared to their male equivalents. There were no gender differences in cortisol in pre- to early pubertal groups. The authors speculate that “for at least a portion of each day the brains of MPP girls may be exposed to higher levels of cortisol than the brains of MPP boys”, an observation supported by the data from Tornhage (2002), who, in a study of 210 boys and 176 girls (age range 7–15 years) observed a maximum value of 53.9 nmol/l in morning salivary cortisol in girls compared to 33.2 nmol/l in boys.

Conversely, many studies have failed to show gender differences in basal salivary cortisol levels in children: in morning and evening samples over an age range of 1–15 years where all the girls were pre-pubertal (Groschl et al. 2003); in morning and evening samples (Chryssanthopoulou et al. 2005) and afternoon/evening samples (Essex et al. 2002) in 4.5 year old children; in a group of 12 year olds sampled from 0730 to 2030 h (Bartels et al. 2003); in 31 children approaching puberty, mean age 10.7 years, sampled from morning to evening (Carrion et al. 2002); and in a study of 1152 Swedish children aged 5–14 years sampled within 30 min of waking (Koupil et al. 2005). In the latter study, no significant influence of either gender or pubertal status on cortisol was observed although a small gender difference in cortisol was detected in older boys. No gender differences in total cortisol at any age (range 5–15 years) were observed in a study of 306 children (Tornhage and Alfven 2006), although in a previous study (Tornhage 2002) morning cortisol was higher in 7 year old boys compared to girls.

This lack of consensus in the literature does not necessarily mean that there are no gender differences in HPA activity in normal healthy children, only that this issue has not been systematically addressed. Gender and age in children are variables that are inextricably linked through hormone production and behaviour and any study designed to investigate gender differences in cortisol should take samples from boys and girls within well-defined age groups with sufficient power to provide meaningful data. The literature suggests that gender differences in salivary cortisol may exist but it is possible that they are too subtle to be revealed within current experimental protocols. In most child studies, saliva samples are collected outside of the laboratory, which necessitates a more pragmatic approach to sampling. Some studies utilise samples collected on school days while in others samples are collected over school days and weekends. One study collected daytime samples at school and evening samples at home (Tornhage and Alfven 2006) while another collected only on “quiet days at home” in an attempt to reduce the “noise” of background stress (Groschl et al. 2003). Since variability in school or domestic activity can be a stress factor affecting child cortisol, perhaps a controlled laboratory environment may be more suitable to elucidating pre-pubertal gender, and other, differences in cortisol. It is also possible that gender differences in cortisol may not be manifest in early morning samples since this is when the largest cortisol variations occur as a consequence of pulsatile release. Afternoon sampling, when the HPA axis is relatively quiescent and the diurnal slope flattened, may be a preferable time to determine any influence of gender on basal cortisol.

Few studies have investigated gender differences in the HPA axis response to experimental challenges which, unlike most basal sampling studies, are performed in a laboratory setting. In a meta-analysis of five independent studies, no gender difference was observed in the cortisol responses of children (mean age 12.1 years) to the Trier Social Stress Test (TSST) administered in the afternoon/evening (Kudielka et al. 2004). No gender effects were reported in a group of 36 control subjects, mean age 13.8 years (18 boys, 18 girls), either prior to or following a 5% CO2 stress or a tilt test (Coplan et al. 2002). Increased anticipatory and decreased reactive cortisol responses were observed in boys compared to girls, age range 7–9 years (68 boys, 72 girls), in response to a modified TSST performed in the early afternoon (Jones et al. 2006). Although this was not the primary question asked of the study design, which was to investigate an association between HPA axis activity and birth weight, it is certainly worth further investigation as possible evidence for early gender-determined coping behaviour in stressful situations.

Age

The HPA axis is believed to reach adult-equivalent maturity at an early age. The circadian pattern of cortisol secretion develops early in breastfed infants, generally not being present before 4 weeks (Groschl et al. 2003) but becoming apparent between 2 and 20 weeks post-partum, with no apparent gender dimorphism in early or late development during this period (Santiago et al. 1996). However, studies in this area have employed small group size, there are inherent difficulties such as the effects of stress and birthweight (Ong 2005) in establishing a normal range, and there are some discrepancies in the data depending on whether blood or salivary sampling was conducted. In a careful and comprehensive meta-analysis, Gunnar and Donzella (2002) found that, while there were no statistically significant age-related differences in morning (0900–1100 h) salivary cortisol values between 2 and 36 months, there was considerable variability between groups, and the slope of mid-morning to mid-afternoon cortisol appeared flatter compared to 4 year olds. In a similar study, cortisol values tended to be higher throughout the day with greater variability in younger children compared to those at 30 and 36 months (Watamura et al. 2004), suggesting that the HPA axis circadian rhythm continues to develop throughout the first 3 years of life. This is broadly consistent with the report by Groschl et al. (2003) of wide variability in cortisol within and between groups ranging from one month to 15 years. Although morning cortisol levels in this study were not correlated with age, midday and evening values were generally higher in the younger children. This suggests that the circadian rhythm of basal HPA axis activity exhibits some hypervariability over the first 4 years of life before settling into an adult-equivalent rhythm. It is as yet unclear whether this represents early developmental processes in neuroendocrine systems reflecting ongoing programming of stress responsiveness during mother-child bonding, or whether it is a consequence of daytime sleeping behaviour in young children.

Age-dependent changes in HPA axis activity during the periods of adrenarche and puberty in humans have not been adequately studied. Development of the adrenal cortex and consequent increased sex and glucocorticoid hormone secretion during the adrenarche should be taken into account for cortisol measurements in children between ages 6 and 10 years, while the highly complex repertoire of steroids, peptides and other compounds secreted during puberty will exert significant effects on HPA axis activity in studies on older children. In response to maturation of the hypothalamic gonadotrophin releasing hormone pulse generator during puberty, compounds such as oestrogen, progesterone, testosterone and leptin are secreted (Grumbach 2002), all of which have been shown to profoundly influence cortisol secretion in humans (Young and Altemus 2004), although the literature on stimulatory or inhibitory effects of individual steroids lacks consensus.

The interactions between hormones secreted during puberty and the HPA axis are highly complex and not well understood. What is certain is that this time of sexual development is a difficult period in which to obtain consistent and reproducible experimental cortisol data. Age of onset of puberty occurs earlier in girls than in boys and can differ significantly between individuals and ethnic groups. There are also diurnal changes in sex steroid secretion which may affect HPA axis activity. Pulsatile secretion of luteinising hormone and follicle stimulating hormone, and consequent release of testosterone, oestradiol and progesterone, is largely a nocturnal phenomenon prior to puberty while during puberty daytime pulsatile patterns begin to emerge. To minimise variability in data, strict attention should be paid to the subjects' pubertal stage when designing and recruiting for experiments.

Salivary cortisol levels may be correlated with age in the period before and during sexual development but the literature is conflicting. This is largely because most studies have used a heterogeneous sample of age ranges rather than focusing on defined age groups. In a study over the age range of 7–15 years, median cortisol values in 7–9 year olds between 0800 and 0900 h were lower than 10–12 year olds independent of gender (Tornhage 2002) but in a more recent study (Tornhage and Alfven 2006) there was a wide variation in mean cortisol levels in samples taken at 0800–0900 h over the age range 6–15 years in both boys and girls. The highest mean cortisol in boys was observed in the group with a mean age of 6.4 years and the lowest at a mean of 9.4 years while the groups with the highest and lowest mean values of cortisol for girls were 12–14 and 11.5 years. When cortisol was measured in pooled gender groups using area under the curve values for three sampling times at 0800–0900 h, 1300 and 2000 h, higher cortisol was observed in 6–7 year olds compared to 9 year olds (Tornhage and Alfven 2006). The range of individual cortisol values in all groups in these studies was large and conclusions are difficult to draw, but this example serves to illustrate some of the more inconsistent aspects of the literature. In a large study of children ranging from 5 to 14 years (n = 1152), there was a tendency towards lower cortisol with increasing age (Koupil et al. 2005). Increased morning (0800–0900 h) basal cortisol has been reported following onset of puberty in both boys and girls (Tornhage 2002) and in premature adrenarche in 7-year-old girls (Cizza et al. 2001), suggesting a correlation of salivary cortisol with puberty but this was not observed in 8–16 year olds at either 0800 or 2000 h (Netherton et al. 2004).

Body mass index

Although a negative correlation between body mass index (BMI) and basal morning and total salivary cortisol has been observed in children (n = 31, age range 6–18 years) with psychosomatic abdominal pain (Tornhage and Alfven 2006), no association between basal morning cortisol and BMI has so far been reported in healthy children (Netherton et al. 2004; Koupil et al. 2005). However, these were general studies without defined obese groups and were not specifically designed to study the association between cortisol and BMI as the primary outcome. In a small laboratory-based study of children aged 8–10 years, fasting morning cortisol was not different in obese subjects with BMI in the 95th percentile or greater for their specific age and gender (n = 10) compared to lean controls (n = 11) but there was a significant increase in cortisol in response to a standard breakfast in the obese group (Hershberger et al. 2004). Physical exercise in the early afternoon resulted in a slight increase in cortisol in the control group and a 32% decrease in cortisol in the obese group. These data suggest that, while basal HPA axis activity may not be correlated to BMI in children, variables such as meal patterns and composition and physical activity, which are relatively uncontrolled in the domestic environment where most studies on children are performed, may impact significantly on cortisol secretion in children with a BMI outside the normal range.

Given the compelling evidence for HPA axis dysfunction in obese adults, it would be prudent to avoid recruiting obese children as controls in studies measuring salivary cortisol until such an association can be confidently ruled out. The extremely wide range of 9.4–75 nmol/l given for cortisol at 0900 h, and < 1.7–9.5 nmol/l at 2300 h, in 21 obese children aged 1–16 years (Martinelli et al. 1999) suggests that some obese children have abnormal cortisol secretion. BMI is notoriously difficult to calculate accurately in children since it depends upon rate of growth (Wells and Fewtrell 2006). In the light of current moves to replace the BMI as an indicator of underlying pathology with a more appropriate measure such as percentage of visceral fat, a study into the association of the latter with salivary cortisol in children would be appropriate.

Sampling conditions

Location and circumstances of salivary sampling are also important variables which can affect cortisol levels in children. Protocols generally require either home sampling with the aid of the parents or sampling within a laboratory setting. The latter is more convenient for testing HPA axis responses to stress tests, and also is more likely to ensure accuracy of sample timing and compliance with the study protocol, but it may be perceived by the child as a stressful environment in itself which may elevate basal cortisol levels, particularly if the tests are performed in a laboratory within a hospital. In a population of 7–9 year olds, it has been shown that basal cortisol increased in a laboratory setting immediately prior to a TSST compared to samples taken at home at the equivalent time of the afternoon (Jones et al. 2006). This has also been observed in a study of 9-year-old girls (Gunnar 2003). However, several studies conducted in the morning in younger children of pre-school age have observed decreased basal cortisol in a laboratory setting compared to home sampling (Gunnar and Donzella 2002; Gunnar 2003). It is not easy to explain these paradoxical observations other than through a possible association with age and gender (Gunnar and Donzella 2002; Jones et al. 2006). While an increase in cortisol secretion may be a response to the novel stressor of a laboratory environment, it is more difficult to explain the opposite effect of decreased cortisol. It is possible that this phenomenon represents age-related anticipatory excitement and consequent HPA axis hyperactivity overnight prior to the visit, resulting in reduced secretion of cortisol the following morning.

In spite of these paradoxical reports, however, the literature is consistent in that home and laboratory cortisol baseline values do differ, so it may be prudent to establish a home salivary cortisol baseline in the experimental group prior to any laboratory visit. However, even home baseline measurements of cortisol must be established under well-defined conditions since they are subject to alteration by factors which may be difficult to control. For example, participation in physical activities in the evening can elevate bedtime levels of cortisol in boys but not girls in the age range 7–10 years (Kertes and Gunnar 2004). It is of interest that a laboratory study in which anxious children mounted an enhanced cortisol response to a 5% CO₂ inhalation test (Coplan et al. 2002) was reproducible in the home environment (Terleph et al. 2006). Therefore HPA axis responses to an experimental stressor may be less subject to influence by the sampling environment than basal cortisol secretion.

Units of measurement

An unwieldy and confusing range of units of cortisol measurement exists in the literature, probably more than for any other analyte. This problem pertains to general human studies on cortisol, but units as diverse as μg/dl, μg/l, ng/ml, ng/dl, mg/dl and nmol/l (SI, standard international, units) have been utilised in studies on cortisol concentrations in children. Some groups publish log-transformed data without giving the raw data in the text, making it difficult to directly compare values with other laboratories. We have also found papers that include no units, and we have observed more than one unit in use in the same paper, e.g. mmol/l and nmol/l. It would be very helpful to the harmonisation of data and results in the field if a consensus of agreement could be reached on a common unit of measurement. Furthermore, it is not uncommon to find quite inappropriate units for salivary cortisol such as nmol/ml and also data expressed in μg/dl or nmol/l but at concentrations equivalent to serum. This type of error multiplies the cortisol values by a factor of 1000. It is quite possible that some laboratories have the wrong conversion factor programmed into their data analysis software, a matter which is simple to check.

Assay conditions

Many types of assays for salivary cortisol measurement are in common use, such as radioimmunoassays, immunofluorescence assays and ELISAs, and the output of these may not be easily comparable. Differences in assay conditions such as buffer composition, antiserum specificity and free fraction separation may be critical. Antisera may cross-react with other steroids present in saliva, in particular cortisone, a metabolite of cortisol produced by the enzyme 11beta-hydroxysteroid dehydrogenase 2 (11beta-HSD2) which is widely distributed within tissues including the parotid gland. Cortisone can represent up to 30% of the total glucocorticosteroid content in saliva and therefore an assay utilising an antiserum with significant cross-reactivity with cortisone will give a markedly different result compared to an assay specific for cortisol alone. 11beta-HSD2 activity can be influenced by a range of factors including diet, and therefore the ratio of cortisol to cortisone in control or experimental groups may vary between individuals. In addition, most antisera utilised in cortisol assays exhibit significant cross-reactivity with the synthetic steroid prednisolone which may be present in samples taken from children on glucocorticoid therapy for respiratory disorders.

There is no imperative to use an antiserum which is specific for cortisol only; such antisera are in fact rare, but what is necessary to the operator, and helpful to the reader of the literature, is to know exactly the degree of cross-reactivity reported in an assay. Without this information, which few papers give, it is impossible to determine how much immunoreactivity reported in salivary samples is cortisol and how much is due to cross-reactivity with other steroids.

Secondly, it is widely assumed that cortisol exists in saliva solely in the free, bioactive form unbound to CBG. In fact CBG is present in saliva (Hammond and Langley 1986; Chu and Ekins 1988), possibly secreted by the parotid gland as well as the liver (Hammond et al. 1987). Although concentrations of salivary CBG represent only 0.1–0.2% of plasma CBG, this has been calculated as having the effect of binding 10–15% of total salivary cortisol in normal subjects. In other words, although there is a strong positive correlation between salivary and blood cortisol concentrations (Kirschbaum and Hellhammer 1994), the ratio may be considerably less than unity.

There is no evidence that salivary CBG levels are different in children older than 12 months compared to adults, and therefore CBG may be present in saliva from subjects of all ages, although researchers planning studies in babies should be aware that synthesis of CBG increases throughout the first year of infancy with a wide variation between individuals (Hadjian et al. 1975). Little is known about CBG levels in children and the conditions under which its production may be influenced. Concentrations of CBG may vary in individual samples in response to medication, or under psycho- or patho-physiological states, introducing a further degree of uncertainty into measurements of salivary cortisol. However, this issue may be readily addressed. The extent to which salivary CBG will bind cortisol in saliva samples, and therefore the accuracy of measurement of free cortisol, can depend on assay conditions. Many assays currently utilise neutral buffer which will not dissociate cortisol from CBG in saliva. It is possible to use a low pH (3–4) buffer to dilute saliva samples, denaturing CBG and resulting in accurate measurements of unbound cortisol.

Compliance

Compliance with protocol is a major issue, particularly when subjects collect their saliva samples outside of a laboratory or clinical environment, and there are special factors to consider when children are involved in the investigation. To date, there are two published studies that address the specific issue of adult compliance with saliva sampling protocols, one in a sample of participants drawn from the community using a one day protocol (Kudielka et al. 2003) and the other in women with fibromyalgia compared to healthy controls employing a seven day protocol (Broderick et al. 2004). In both of these studies, an electronic sampling device enabled accurate recording of collection time. Results consistently revealed a lower rate of objective compliance for participants who were unaware of the sampling device, compared to subjective self report and in comparison with participants who were aware of the device. Furthermore, a flatter awakening cortisol rise, or response (ACR), and diurnal decline were observed in the non-compliance groups.

Equivalent compliance studies have not so far been reported for children. One factor which would improve compliance in home studies is saliva sampling under parental supervision but the effect of this on the ACR and diurnal rhythm of cortisol in children has not been studied directly. In addition, there are issues of saliva sampling relating to different stages of childhood and adolescence, e.g. differences in school routines at different ages, and age-related behavioural variables, e.g. diet and sleep. A whole range of variables can impact on compliance in salivary sampling across the childhood/adolescent age spectrum. Compliance may be relatively reliable in the parent-controlled environment of early childhood (pre-school) but less so during the onset of puberty and the early teenage years. Compliance with timekeeping is of special importance for measurement of the ACR (see Section 3.3) but close attention should be paid to the overall issue of protocol compliance for salivary sampling in children. To this end, supplying the parents with colour-coded sample collection tubes, along with clear instructions emphasising the importance of timekeeping to the study, is a useful strategy. Novel incentives and reminder techniques such as text messaging may prove effective, particularly with adolescent populations.

Salivary cortisol measurements—application and utility

Pathophysiology

A dysfunctional HPA axis has been reported in many diseases in adults including cancer (Reiche et al. 2004), fibromyalgia (McBeth et al. 2005) and autoimmune diseases such as multiple sclerosis (Huitinga et al. 2003), rheumatoid arthritis (RA) (Jessop and Harbuz 2005) and asthma (Rook 1999). However, rarely have investigations extended into this association in childhood diseases. For example, it is not known whether the defect in HPA axis activity associated with RA in adulthood is also present in juvenile RA. One study suggests that it may not be, since a normal cortisol response to the stress of venepuncture was reported in mixed-sex groups (age range 7–17 years, n = 16) with RA or fibromyalgia compared to controls (Conte et al. 2003). This and other diseases are an untapped area for important research into the development of abnormalities in HPA axis function in children and in particular whether HPA axis dysfunction precedes, or is consequent to, the onset of inflammation.

Despite the paucity of research in this area, a dysfunctional HPA axis has been identified in some diseases prevalent in childhood. Autism, an early onset neurodevelopmental condition that is characterised by marked inability to interact with a novel environment, has been associated with HPA axis dysfunction but the historical literature is conflicting, possibly reflecting problems in diagnosis and also the use of venepuncture to obtain samples for cortisol analysis in this group of subjects who are highly sensitive to physical contact. Recent studies in autistic children using better defined methods of diagnosis and the use of salivary sampling have demonstrated a subtly dysfunctional HPA axis. In a laboratory-based study of 8–10-year-old children between 1000 and 1600 h, the autistic group (n = 10) had similar baseline salivary cortisol to the normal control subjects (n = 12) but showed an increased cortisol response to a psychosocial, but not a physical exercise stressor compared to controls and a group with multiple complex developmental disorder (n = 10), a condition which shares many but not all symptoms with autism (Jansen et al. 2003). Similar findings have been reported in a study of 6–11-year-old children who provided salivary samples in the home environment (Corbett et al. 2006). Diurnal variation in cortisol secretion in autistic children (n = 10) was not different to normal controls (n = 12), although there was a wider range of individual cortisol values observed in the former group. In response to the novel environmental stressor of a simulated MRI scan in the laboratory, autistic children mounted a cortisol response while no response was observed in the control subjects. Although these studies are important in highlighting possible HPA axis hyperactivity in autism, interpretation is hampered by the low subject numbers, and larger studies with more frequent sampling times are necessary to further examine this phenomenon.

Children with fragile X syndrome, a genetic disease associated with behavioural, physical and psychiatric abnormalities, demonstrated elevated diurnal levels of salivary cortisol compared to normal siblings, which was particularly evident in males (Hessl et al. 2002), with a significant correlation of cortisol with abnormal behaviour (Hessl et al. 2002, 2006). Major depression, which in adults is characterised by a hyperactive HPA axis, has also been investigated in children using salivary cortisol measurements. In a longitudinal study of chronically depressed 8–16 year olds sampled at presentation and at 36 and 72 weeks, basal cortisol secretion was persistently elevated in the evening but not in the morning compared to a control group with non-chronic depression (Goodyer et al. 2001). This increase in total cortisol secretion and flattened diurnal curve is characteristic of HPA axis hyperactivity in depressed adults. Cortisol levels in depressed pre-schoolers (range 3–5.6 years) subjected to a regimen of psychosocial tasks in a laboratory setting were significantly increased compared to controls (Luby et al. 2003). Clinical depression is becoming increasingly more prevalent in children and adolescents and more information is required about HPA axis dysfunction, its relationship to depressive symptoms, and the influence of antidepressants on cortisol secretion in this population.

In contrast to the elevated cortisol reported in depression in adults and children, attenuated cortisol secretion has been observed in adults with post traumatic stress disorder (PTSD). However the opposite pattern emerged in a study of children (mean age 10.7 years, n = 51) with a history of exposure to trauma and PTSD who had similar basal morning cortisol levels compared to the control group but exhibited an elevated and flattened diurnal cortisol pattern consistent with cortisol hypersecretion (Carrion et al. 2002). Morning and afternoon basal cortisol levels were elevated in children suffering sudden unexpected parent death compared to normal controls but the bereaved group suffering from PTSD showed decreased afternoon cortisol secretion and a hyper-sensitivity to the dexamethasone suppression test (Pfeffer et al. 2006). PTSD in children may represent an example of the phenomenon of allostatic load (McEwen 2003), with the potential for impaired immune responses and adverse mental and physical health in adulthood. Detailed longitudinal studies are required to investigate HPA axis dysfunction in children with PTSD or severe stress, and any links with long-term ill-health.

HPA axis dysfunction is well-established in a number of chronic diseases in adults (Harbuz 2002) but information is scarce in children. Basal cortisol levels in a laboratory-based afternoon study were similar in asthmatic children (n = 17) compared to controls (n = 18) but the asthmatic children showed a blunted cortisol response to the TSST (Buske-Kirschbaum et al. 2003). Awakening cortisol levels in salivary samples collected at home did not differ between the groups. A similar blunted response to the TSST was observed in children with atopic dermatitis (Buske-Kirschbaum et al. 1997), suggesting that salivary cortisol may provide a marker of stress sensitivity in this and possibly other chronic inflammatory diseases. It is well-established that stress can precipitate the onset and exacerbate the severity of inflammatory disease (Jessop et al. 2004) but the underlying mechanisms are poorly understood. Further information about HPA axis activity and cortisol secretion in childhood, both basal and in response to stress, may provide insight into the early onset of inflammatory illnesses such as juvenile arthritis, serious allergies and asthma. It is also possible to speculate that detection of abnormal HPA axis circadian patterns and responses to psychological stress in children may be causally related to later predisposition to adult onset of chronic pathologies such as autoimmune diseases, metabolic syndrome and cardiovascular disease, permitting early identification and therapeutic intervention.

Psychophysiology

Salivary cortisol measurements have played an important role in allowing evaluation of how children respond physiologically to naturalistic stressors such as entering kindergarten or other pre-school programmes (Quas et al. 2002; Smider et al. 2002; Doussard-Roosevelt et al. 2003; Gunnar et al. 2003; Blair et al. 2004) and the first week of school (Bruce et al. 2002; Turner-Cobb et al. 2006) at an age when they may find it difficult to verbally express their feelings. Low basal salivary cortisol has been recorded in children who experienced problems integrating into the school environment (McBurnett et al. 2000; Oosterlaan et al. 2005). Low basal morning cortisol in a longitudinal study of boys (n = 313, age range 10–12 years) has been correlated with increased likelihood of aggressive acts 5 years later (Shoal et al. 2003). While the use of hormonal markers to identify those at risk of developing anti-social behaviour is controversial and research in this area is at an early stage, there is no doubt of the potential of its long term utility in identifying at-risk individuals and providing necessary social support. For this reason, properly designed longitudinal studies of this nature are to be encouraged.

One important area of research is that of child-parent interactions and in particular the ways in which parental psychological wellbeing and socio-economic status can impact on the developing child. In a classroom study, basal salivary cortisol was elevated in young children (ages 6–10 years) from families with low compared to high socio-economic status (Lupien et al. 2001), a difference which disappeared after the age of 10 years. Morning basal cortisol levels were also elevated in children aged 6–10 years whose mothers were depressed (Lupien et al. 2000; Ashman et al. 2002); morning, evening and total cortisol levels were higher in pre-school children (n = 56, age 3.5–4.5 years) whose mothers were suffering from job or domestic dissatisfaction (Chryssanthopoulou et al. 2005); and afternoon cortisol levels were higher in children exposed to marital violence (Saltzman et al. 2005). Maternal levels of stress or depression may also predict later development of HPA axis hyperactivity in children (Essex et al. 2002; Gutteling et al. 2004, 2005; Halligan et al. 2004). The degree of bonding between mothers and infants has also been assessed and salivary cortisol has been used to identify correlations between maternal behaviour, mother-child separations and child insecurity (Ahnert et al. 2004; Schieche and Spangler 2005). Cortisol levels increased in infants aged 15 months during the maternal separation phase following the start of childcare and had returned to baseline 6 months later (Ahnert et al. 2004). Cortisol tended to increase throughout the day in children in childcare compared to relatively constant levels in children sampled at home; this increase was more pronounced in toddlers (mean age 29.7 months) than in infants (mean age 10.8 months) (Watamura et al. 2003). In a study of 3–5 year olds, a flatter diurnal cortisol slope was associated with low quality childcare programmes and a steeper diurnal slope with high quality programmes (Sims et al. 2006). Children aged 20–60 months undergoing fostering exhibited a flatter diurnal cortisol slope from morning to evening, with lower morning levels (Dozier et al. 2006).

It is very important that a distinction is made between a normal healthy increase in cortisol in response to a novel environment such as starting school, with consequent normalisation of cortisol as the children adapt (Turner-Cobb et al. 2006), compared to a situation where children exhibit chronically elevated cortisol over a period of time in their new environment. This may represent evidence of failure to adapt to new surroundings which may be associated with chronic stress and unhappiness. These studies have the potential to shed important light on HPA axis development during childhood and reveal differential mechanisms of adaptation or mal-adaptation to adverse and unsatisfactory domestic or institutional situations. Identification of individuals or groups of children with elevated cortisol secretion, and consequent risk of developing stress-related disorders, allows the possibility of therapeutic intervention at an early stage. Incidentally, almost all of these reported studies measured cortisol only in children. There is the potential to learn more about family interactions through studies designed to simultaneously measure cortisol secretion in parent-child dyads, studies of father-child dyads being notably rare.

Awakening cortisol rise—a special case for investigation

In the great majority of the studies cited above, morning cortisol levels have been measured in a single sample, usually taken on or soon after awakening. An alternative to single sampling is measurement of the early period of increased cortisol secretory activity known as the ACR. This requires an initial sample immediately upon waking, followed by a second at 30 min (Clow et al. 2004), and often with interval measurements between these points and for approximately 15 min afterwards. The ACR is most frequently measured in saliva since blood sampling over the awakening period would require an indwelling cannula and a comparatively stressful intervention in sample collection unless careful sampling conditions are observed (Born et al. 1999). A robust ACR is believed to be associated with good health, although there is a paucity of convincing evidence for ACR dysfunction in any pathological condition (Clow et al. 2004). The ACR is a consistently reported physiological phenomenon in normal healthy adults but there have been few reports on the ACR in children. Whilst there is a small but growing body of evidence that has included children when looking at the ACR (Pruessner et al. 1997; Wust et al. 2000; Bartels et al. 2003; Buske-Kirschbaum et al. 2003; Kudielka and Kirschbaum 2003; Feder et al. 2004; Rosmalen et al. 2005), these studies have used heterogeneous samples with respect to age, many of them combining adults and children within the same cohort. Some evidence from mixed age group studies does suggest that the ACR may be present in children but is less pronounced than in adults (Pruessner et al. 1997; Rosmalen et al. 2005).

There is, to date, no published work that specifically examines the ACR with respect to age and gender. In a study of 68 boys and 72 girls aged 7–9 years, a significant ACR was present in boys but not in girls when expressed in terms of the means of the groups (Jones et al. 2006). The broad overlap of data points between the groups almost certainly indicates presence of the ACR in some girls and absence in some boys but nevertheless these data are supported by another study in which a significant ACR was observed in a small population (mean age around 9 years) of boys, but not girls (Yeap et al. 2006). These data suggest that the ACR may mature earlier in boys than in girls. In contrast, a large study of 1768 healthy children aged 10–12 years found that the ACR was significantly greater in girls (Rosmalen et al. 2005), while in a Dutch study of 180 twin pairs of 12 year olds a similar ACR was observed in both boys and girls (Bartels et al. 2003). Therefore, it is possible that the ACR develops earlier in pre-pubertal boys while becoming manifest in girls quite rapidly around the onset of puberty. A comprehensive study with well-defined gender and age groups is required to fully characterize the ACR in children prior to and during the onset of puberty. Development of a robust ACR may be instrumental in determining the way a child reacts to waking up, a situation which may be perceived as a novel environmental stress. It is possible that the ACR may be dysfunctional in syndromes in which children find it difficult to relate to their environment and which are more prevalent in boys, such as attention deficit and hyperactivity disorder, autism and related conditions.

There is little information about the ACR in sick children. The ACR is lacking in the majority of children who have recovered from leukemia following chemotherapy or bone marrow transplant (Yeap et al. 2006), which may suggest that the ACR is compromised by serious disruption of immune system development and function. The consequences of an impaired ACR in childhood on future health are unknown. Design of future studies to investigate any disruption of the ACR in pathological conditions should take into account the reported effects of awakening time on the ACR (Federenko et al. 2004).

Conclusions and future directions

We have identified a number of factors that have the potential to influence accurate measurement and evaluation of salivary cortisol in children. There is a body of evidence which suggests that salivary cortisol may be age and gender-dependent, but these variables have not been systematically investigated. There are physiological variables which may influence salivary cortisol such as body weight, issues of protocol design such as sampling environment and compliance, and technical aspects relating to assay conditions. Differences in genotype may also influence HPA axis activity (Wust et al. 2000) and may underpin evidence now emerging for race differences in cortisol secretion (Gozansky and Kohrt 2006). Several polymorphisms have been identified in the glucocorticoid and mineralocorticoid receptors (van Rossum et al. 2002; Wust et al. 2004) which may affect cortisol feedback. These genetic influences on cortisol production in children have not been systematically studied.

Given these caveats, where does this leave the measurement of salivary cortisol as a useful research tool in children? There is no doubt that its utility cannot be over-emphasised as a non-invasive marker of HPA axis activity in children where blood sampling is not usually a viable option. When careful experimental and methodological conditions are observed, the measurement of free salivary cortisol may better reflect normal adrenal function than total serum cortisol (Gozansky et al. 2005). However, research into the HPA axis in children is now at a crossroads where the accumulated weight of publications may become counterproductive and even misleading unless there is an attempt to harmonise the methods and procedures underlying data generation within and between laboratories in a more rigorous manner.

Although this review has focused on salivary cortisol as a marker of HPA axis activity in children, it is worthwhile considering the utility of measuring other analytes. Assays are now commercially available for the adrenal steroid dehydroepiandrosterone (DHEA) and its sulphated form DHEAS in saliva, and also for the sex steroids oestradiol, oestriol and testosterone, all of which can influence HPA axis activity. Correlation ratios between these steroids and cortisol can amplify the power of data over and above that which can be gained from measurement of individual analytes, and can provide different kinds of useful information about HPA axis integrity under basal and stressful conditions (see review by Granger and Kivlighan 2003). This strategy has proved invaluable in revealing subtle but significant changes in the ratios between serum DHEA and cortisol in RA compared to healthy control subjects, creating a milieu in the blood which predisposes to inflammation, where measurement of these analytes alone has been less useful (Straub et al. 2002). This approach may also be applied to correlate adrenal steroid output with psycho- or patho-physiological profiles in children. It is interesting in this context to note that individual variability in cortisol secretion was not correlated with pubertal status in a group of 8–16 year old children but there was a significant correlation between puberty and the ratio of salivary cortisol/DHEA (Netherton et al. 2004). In another study, an elevated morning salivary cortisol/DHEA ratio was a predictor for onset of major depression in adolescents (Goodyer et al. 2003).

Finally, an important question to ask is whether a “gold standard” for cortisol measurement in children is really attainable. This is a difficult question to pose for any hormone in humans, and it is particularly pertinent in childhood when so many neuroendocrine and behavioural changes are occurring which can profoundly affect HPA axis activity. It is probably unrealistic to aim for a standardised “normal range” for basal cortisol across all laboratories. Rather we propose the need for major studies specifically designed to highlight the effects of potentially confounding variables such as age, gender or BMI on cortisol in children. The most pressing requirement is for a large and systematic study to determine whether salivary cortisol is age and/or gender-dependent in healthy children, prior to and during the onset of puberty. This would inform decisions for protocol design and study recruitment which would permit better defined control groups in future studies. Once we have a clearer understanding of how the variables of age and gender can impact on the circadian pattern of cortisol secretion in normal children, priority research areas to focus on include (i) dysfunctional HPA axis activity as a marker for behavioural problems in children unable to fully adapt to and engage with their environment; (ii) the consequences in later life of anomalous cortisol responses to stressors, environmental or domestic, experienced in childhood; (iii) determination of dysfunctional HPA axis activity in diseases prevalent in childhood such as juvenile arthritis, leukaemia, asthma and diabetes mellitus.