Externalizing and Attentional Behaviors in Children of Depressed Mothers Treated With a Selective Serotonin Reuptake Inhibitor Antidepressant During Pregnancy

Objective  To evaluate attentional and activity behaviors in 4-year-olds following prenatal selective serotonin reuptake inhibitor (SSRI) exposure.

Design  Prospective cohort design.

Setting  Tertiary care center.

Participants  Twenty-two 4-year-olds with prolonged prenatal SSRI medication exposure and 14 children without prenatal exposure.

Main Exposure  Prenatal SSRI exposure.

Main Outcome Measures  Group differences in externalizing behaviors (according to the Child Behavior Checklist) and direct observations of child attention, activity, and impulsiveness in a laboratory setting using the procedure by Crowell and colleagues were compared, including measures of the duration of prenatal SSRI exposure, umbilical cord drug levels, a history of poor neonatal adaptation, and maternal mood.

Results  Externalizing behaviors did not differ between groups. Maternal depression and anxiety at the 4-year follow-up were associated with increased reports of externalizing behaviors. Increased externalizing behaviors were associated with increased umbilical cord drug levels (F_1,34 = 6.3; P = .02), but when controlling for maternal depressed mood at the 4-year follow-up, such levels only accounted for 11.2% of the behavioral outcomes (P>.05). On direct observation, the persistence score for child behavior was significantly lower in the exposed group. Increased aggressiveness scores were associated with a history of poor neonatal adaptation, even when parental report of stress was added to the model (F_1,34 = 4.0; P = .03); however, neither parental report of stress nor poor neonatal adaptation were significant (both P = .09), suggesting that both are important, if not unique, predictors of child behavior.

Conclusions  These findings suggest that the best predictors of externalizing behaviors at age 4 years are current maternal mood and parental stress, regardless of prenatal depressed mood and SSRI treatment during pregnancy. It remains uncertain whether poor neonatal adaptation can be excluded as a possible predictor of externalizing behaviors.

Selective serotonin reuptake inhibitor (SSRI) antidepressants are frequently used to manage maternal anxiety and depression during gestation,¹ and recent concerns about altered neonatal behaviors following prenatal exposure have raised questions about possible long-term neurodevelopmental effects.² Beyond altered neonatal behaviors,² little is known about the long-term neurobehavioral outcomes following in utero exposure to SSRIs.^3-6 Associations between prenatal SSRI exposure and cognitive and language development have not been demonstrated,^5,6 and a single study³ has shown subtly altered early-childhood fine motor development following SSRI exposure in the third trimester. In that study, children prenatally exposed to SSRIs had lower Apgar scores at birth and lower Bayley Scales of Infant Development psychomotor developmental scores at ages 6 to 40 months. Even with prenatal SSRI treatment, maternal mood at 4 years still appears to play a role in influencing internalizing behaviors⁵ during childhood.

Use of SSRIs during pregnancy also occurs in the context of maternal mental illness, which independently is a well-known risk to the fetus and developing child.⁷ Prenatal and postnatal exposure to maternal anxiety and depression have been widely associated with childhood attention, emotion, and arousal regulatory disorders.⁸ Higher levels of maternal anxiety in late pregnancy were related to an increased risk for behavioral-emotional problems in a longitudinal study⁹ of 4-year-olds. Similarly, postpartum depression has been linked to an increase in attentional and behavioral problems during childhood.⁸ However, whether prenatal exposure to SSRI medications and maternal depression independently or cumulatively contribute to the burden of adverse behavioral outcomes during childhood remains a central question.

Previously, we found that prenatal SSRI exposure was associated with an increased risk of poor neonatal adaptation (PNA) associated with increased umbilical cord drug levels, particularly of paroxetine hydrochloride,¹⁰ altered pain reactivity in exposed neonates and infants,^11,12 and increasingly difficult temperament at age 3 months among infants exposed to SSRIs and benzodiazepine.¹³ We expected that if these early behavioral outcomes, which reflected signs of altered serotonin-mediated regulatory processes, persisted beyond the first few months of life, they might be subsequently expressed as altered externalizing behaviors at age 4 years. Specifically, we expected parent reports of higher levels of externalizing behaviors and laboratory-based observations of increased levels of activity and aggressive behavior. To address this question, we studied children from our original cohort of infants to determine whether prenatal SSRI exposure was associated with altered arousal regulation (eg, externalizing behaviors, attention, and aggression) and activity levels during early childhood. For this study, we chose aspects of behaviors that reflect the serotonin-mediated process, namely, a spectrum of community-based and parent-reported attentional behaviors as well as levels of activity, impulsiveness, defiance, compliance, problem solving, task persistence, and control issues occurring in the context of the mother-child relationship in a laboratory setting.

With approval from the University of British Columbia Research Ethics Board, the Children's and Women's Health Centre of British Columbia Research Review Committee, and informed parental consent, a consecutive cohort of mothers and their infants were recruited during pregnancy as part of a prospective longitudinal study of prenatal psychotropic medication use. Mothers and their infants in the control group were recruited during the newborn period and were eligible if no psychotropic, illicit drug use or antidepressant medication use occurred during the pregnancy, the pregnancy was full term (37-42 weeks), and there was no history of maternal mental illness, as previously described.^10,11

Mothers were originally recruited between January 1, 1997, and April 1, 1999, while undergoing treatment for mood and/or anxiety disorders through the Reproductive Mental Health Program at British Columbia Women's Hospital, Vancouver. As previously described,¹¹ 46 mothers in the SSRI group participated in the first phase of this study. All of the mothers were contacted for participation as their children approached their third and fourth birthdays and were studied between July 9, 2002, and July 16, 2003. Twenty-four mothers and children were lost to follow-up, leaving 22 children prenatally exposed to SSRIs for the follow-up study. In the initial cohort, 23 mothers and their infants formed the control group; of these, 9 declined to participate in the follow-up assessment, leaving 14 control children.

The widely used and well-validated Child Behavior Checklist¹⁴ (CBCL) and Child-Teacher Report Form¹⁴ are self-administered and can be completed by the child's mother (CBCL) and teacher (Child-Teacher Report Form) in less than 15 minutes. They were used to provide standardized ratings and descriptive details of the behavior of children aged 1.5 to 5 years in the prior 2 months. The complete CBCL (at ages 1.5-5 years) and Child-Teacher Report Form questionnaires were administered in the published format; however, for the current study, we focused on attention and aggressive behaviors, examining the subscales specific to externalizing problems comprising the attention subscale (including the child's ability to sit still and concentrate) and the aggressive behaviors subscale (including the child's defiance, demandingness, temper, and acting-out behaviors). In addition, we used the clinical scales of attention deficit/hyperactivity problems and oppositional defiance problems. Scoring was carried out using standard computer scoring software that generates t scores.

An assessment of each child's activity was obtained from observations of caregiver-child interactions using a procedure developed by Crowell and colleagues.^15-17 This involved a series of 8 activities designed to elicit behaviors that allow the clinician to focus on the relationship between a child and his or her caregiver in a setting that is unstructured enough to allow for “real-life” or spontaneous interactions. The 8 episodes included free play, clean up, a bubble-blowing episode, 4 increasingly difficult problem-solving tasks, and a separation and reunion episode, and they were completed in 45 to 60 minutes. Child behavior was tabulated to assess the degree to which the child and caregiver were comfortable with each other, the ways they negotiated transitions, their ability to problem solve together, their use of shared affect (positive and negative) to communicate, and attachment behaviors. Behaviors¹⁸ were scored to yield measures of 7 child behaviors: positive affect, withdrawn or depressed, irritability or anger, noncompliance, aggression, persistence, and enthusiasm. Five caregiver scales were also derived: behavioral responsiveness, emotional responsiveness, positive affect, withdrawn or depressed, and irritability or anger. The scales are scored on a 7-point anchored system (a score of 0 being the lowest and 7 the highest).¹⁵ Child affect–irritability, child affect–withdrawn, and child affect–positive scales are scored from 0 to 4, and the remaining scales were scored from 1 to 7. Fear of mother was scored as yes or no. This procedure has been previously used to demonstrate different caregivers' representations and specificity of the interaction between mothers and their children as well as to distinguish between clinic-referred and comparison toddlers¹⁵ and between delayed and nondelayed toddlers.^15,19 All of the sessions were videotaped and later coded by a trained infant psychiatrist (P.R.) blind to the assignment of groups.

Videotaped child activity behaviors were used to derive measures of activity using a technique that is similar to those used by ethnologists. Hutt et al²⁰ used this method of observation of children in experimental settings. We adapted this method to observe children through a 1-way mirror, noting the child's explorations away from the anchor point when the child was left alone at the separation sequence. As all of the children had an opportunity to explore the room in a similar fashion, this was an important opportunity to see how the child exercised the freedom of activity without parental or adult supervision.

These ecobehavioral measures of child activity were tabulated from videotaped behaviors observed during 5 structured mother-child interactions to yield measures of exploratory activity of the child. Observations of a secure base exploration made while the child was playing with his or her mother were contrasted with behavior when the mother was asked to leave the room and during a reunion phase to yield measures of exploration, passivity, and fidgetiness. A free-play period in which mothers played with their child (10 minutes) was followed by these periods: (1) clean up toys; (2) transition 2, bubble time (5 minutes); (3) transition 3, 4 tasks for the mother-child session (total time, 12 minutes) (age-dependent tasks were the following: for 2- to 3-year-olds, stacking activity and simple puzzle; for 2.5- to 3.5-year-olds, a more-complicated puzzle and age-specific Developmental Learning Materials cards used for duplication of shapes and colors; and for 3.5- to 4.5-year-olds, sorting task, matching cards tasks, and playing with a toy house with complex shapes that fit into different slots); (4) transition 4, separation of mother and child (maximum time, 3 minutes); and (5) transition 5, mother-child reunion in the observation room (3 minutes). Videotaped behaviors were also coded to yield measures of 3 aspects of the children's affect, ie, positivity, withdrawal, and irritability.

To ensure that differences in attention and behavior were not a function of another developmental disability, cognitive screening was carried out using the information subtest from the verbal scale and the block design subtest from the performance scale of the Wechsler Preschool and Primary Scale of Intelligence–Revised.²¹

Five-milliliter infant umbilical cord blood samples were collected during labor and delivery. All of the blood samples were collected, stored, and analyzed as previously described.²² Gas chromatography–mass spectrometry electron impact ionization²² was used to measure fluoxetine hydrochloride, its metabolite, and paroxetine.²² Sertraline hydrochloride and its metabolite desmethylsertraline (norsertraline) were measured using liquid chromatography–mass spectrometry.

The Hamilton Rating Scale for Depression²³ is a 21-item clinician-administered scale that measures the severity of depression in adults. Scores range from 0 to 63, with higher scores being associated with higher levels of depression in the patient. Scores ranging from 0 to 7 suggest no or minimal levels of depression; 8 to 17, mild depression; 18 to 25, moderate depression; and 26 and higher, severe depression. The Hamilton Rating Scale for Anxiety²⁴ is a 14-item clinician-administered scale that measures the severity of anxiety. Total scores on this scale have a possible range of 0 to 56, with higher scores on this measure being associated with higher levels of anxiety in the patient. Scores ranging from 0 to 7 suggest no or minimal levels of anxiety; 8 to 17, mild anxiety; 18 to 25, moderate anxiety; and 26 and higher, severe anxiety. The Symptom Questionnaire Score is a self-administered measure used to assess symptoms and well-being in depression, anxiety, anger-hostility, and somatic conditions.²⁵ The Parenting Stress Index (PSI) short version was used in this study to identify parent-child systems that were under stress and at risk for the development of dysfunctional parenting behaviors or behavioral problems in the child. The PSI has been used as a screening, diagnostic, and research instrument for parents of children younger than 10 years.²⁶

A composite measure of activity was derived from the CBCL (parent's CBCL externalizing score + teachers' CBCL externalizing scores) from the ecobehavioral laboratory-based observations, based on the work of Crowell and Feldman,^15,16 summed from the following empiric measures: (1) movement (numbers of moves + numbers of quadrants entered); (2) aggressiveness (negativity with mother + interaction with mother + roughness with toys); (3) attention (compliance + persistence + avoidance); and (4) emotion (negativity + anxiety). The primary analyses compared child behavioral outcomes between prenatally exposed and control groups using t tests. Along with SSRI exposure, the factors of maternal mood, prenatal clonazepam exposure, a history of PNA, and umbilical cord drug levels were examined in regression models as predictors of child behaviors. Reliability was determined using 2 scorers (P.R. and Ursula Brain, BA) reviewing 7 cases (18%), yielding Cohen κ coefficients. Using the methods of Crowell and Feldman,¹⁵ scores within 1 scale point of each other were considered to be in agreement. The κ coefficients for the child behaviors ranged from 1.0 (enthusiasm) to 0.5 (positive affect) for child scores.

Maternal pregnancy and neonatal or infant characteristics are tabulated in Table 1. In the 22 mothers in the group exposed to SSRIs, symptoms of depressed mood persisted and all except 1 mother returned on receiving an SSRI medication at follow-up (the mother had stopped receiving paroxetine and began receiving doxepin hydrochloride). Comparing those studied at the 4-year follow-up with those not followed up, there were no significant demographic differences (Table 1). All of the mothers in the SSRI group had a diagnosis of depression during their pregnancies. Two mothers who had been in the no-exposure control group had since begun receiving an SSRI medication and returned at the 4-year follow-up with a diagnosis of depression (Table 2). Previously, maternal mood at the 2 months' postpartum assessment was not substantially different between groups; by 4 to 5 years later, mothers in the SSRI treatment group reported significantly more depressed and anxiety-related symptoms (Table 3). Child demographic characteristics are tabulated in Table 4.

Child externalizing behaviors (attention, aggression, attention deficit/hyperactivity, and oppositional or defiant behaviors) rated by parent and teacher report did not differ between the exposed and nonexposed groups (Table 5). However, a greater proportion of children with prenatal SSRI exposure had externalizing scores greater than 60 (the clinical cutoff) compared with those in the nonexposed group, although this difference was not statistically significant (7 children [31.8%] vs 2 children [14.3%], respectively; χ² = 1.4, P>.10). Externalizing behaviors reported by parents but not by teachers were significantly associated with measures of maternal mood at the 4-year follow-up (Table 6).

Using a regression model, parental reports of CBCL-measured externalizing behaviors were significantly associated with depressed parental mood even when a history of prenatal SSRI exposure, a history of prenatal clonazepam exposure, the length of prenatal drug exposure, and a history of PNA (PNA, respiratory distress, irritability, jitteriness, and an increased rate of admission to a special-care nursery¹⁰) were added to these models.

Parental reports of increasing externalizing behaviors were associated with increased umbilical cord drug levels at birth (Table 2) (F_1,17 = 6.33, P = .02); however, when maternal depressive symptom measures at the time of the 4-year follow-up were added to the regression model, umbilical cord drug levels accounted for only 11.2% of the variance in behavioral outcomes and were no longer significant (P>.10) (although maternal mood was also not significant [P = .07]).

Mean clinician-coded measures of avoidance, compliance, control, and enthusiasm were derived from a structured laboratory study of interactions between mothers and their children. Average interrater reliability on this scale has been shown to be 0.89.¹⁵ With the exception of persistence, which was significantly lower in the exposed group (F_1,34 = 5.35, P = .03), individual measures and composite behavioral scores of movement (activity), aggressiveness, attention, and emotion were not significantly different between the 2 study groups (Table 7). Mean aggressiveness (negativity with mother + interaction with mother + roughness with toys) composite scores were significantly higher in children with a history of PNA compared with children without a history of these symptoms (10.1 children vs 7.3 children, respectively; F_1,34 = 4.71, P = .04). These neonatal symptoms continued to account for 13.3% of the variance in aggression observed during the mother-child interaction, even when controlling for maternal depressed mood at the time of this follow-up study (P = .03).

Using a regression model, increased composite laboratory-observed aggressive behavioral scores were associated with a history of PNA. However, when the PSI score was added to the model, the overall model remained significant (F = 4.02, P = .03) but neither PSI score nor history of PNA were significant (both P = .09), suggesting that both are important, if not unique, predictors of child behavior.

Maternal mood at the 2 months' postpartum assessment or Bayley Scales of Infant Development developmental scores at age 8 months were not predictive of subsequent child behavior or activity.

Using a variety of parent and teacher reports and direct laboratory-based observations, behavioral differences were observed in the follow-up of 4- to 5-year-old children of depressed mothers treated with an SSRI antidepressant during pregnancy. Although the overall levels of parent- and teacher-reported externalizing behaviors (on the CBCL) were not significantly different between groups, mothers reporting higher levels of stress, anxiety, and depressed mood at the 4-year follow-up were more likely to report increased levels of externalizing behaviors. Within the prenatal exposure group, increased umbilical cord SSRI drug levels at birth were associated with an increased incidence of externalizing behaviors. However, when maternal mood at the time of the study was controlled for, umbilical cord drug levels no longer accounted for a significant percentage of the variance in these behaviors. Using a laboratory-based observational measure, persistence was found to be significantly lower in the exposed group, even when controlling for maternal mood at the time of the study. In addition, increased levels of laboratory-based aggressive behaviors were observed in children with a history of PNA, even when accounting for parental stress. Importantly, however, the history of PNA and PSI scores overlap to some extent, so they both contributed to the prediction of aggression but not uniquely; both seem to be important, but their unique contribution to child behavior in this study could not be teased apart. Such findings suggest that even with appropriate prenatal treatment of depressed maternal mood during and following pregnancy, it appears that concurrent maternal mood still influences child behavior regardless of clonazepam exposure or the length of prenatal SSRI treatment.

To date, there have been limited studies^3-6 of the long-term neurobehavioral outcomes following in utero exposure to SSRIs. To our knowledge, this is the first study to examine activity and attentional behaviors 4 years after prenatal SSRI exposure. Previous work⁶ reported that measures of intelligence and language development in preschool-aged children were not influenced by fluoxetine exposure in the first trimester, although the study did not specify the maternal daily dose of the SSRI, the measure of maternal depression, or the measure of maternal intelligence. In this prospective study of children aged 15 to 71 months with prenatal fluoxetine exposure, cognitive, language, and behavioral development did not differ with prenatal exposure. Using population-based data linking hospital and community data with pharmacy records of SSRI antidepressants, Simon et al²⁷ found no differences in motor or speech development during the first 2 years of life. Mattson et al²⁸ also reported no differences in neuropsychological evaluation results of preschool-aged and school-aged children (aged 48-72 months) between children with and without prenatal fluoxetine exposure. In contrast, Casper et al³ examined 31 children aged 6 to 40 months and found that prenatal SSRI exposure had subtle effects on both motor development and motor control (tremulousness and fine motor movements) as well as lower psychomotor and developmental scores on the Bayley Scales of Infant Development. To date, none of these studies have examined the dual impact of prenatal SSRI exposure and concurrent exposure to maternal mental illness.

The association between maternal mood and child behavior is consistent with previous findings.^7,8,29 Mounting evidence has shown links between adverse maternal mood during pregnancy and preterm delivery as well as lower birth weight, and it suggests that maternal distress may have lasting effects on postnatal development.^30,31 Such effects may include altered motor development,³² impaired regulatory control of behavior in childhood, and psychiatric disorders in adulthood.^33-36 Exposure to postpartum maternal depression alone has also been widely recognized as a key determinant of child development and behavior. Altered sleep and feeding as well as irritability have been observed in infants of depressed mothers, and a negative impact on cognitive, social, and behavioral development during childhood has been observed.³² Beyond prenatal SSRI exposure, prenatal exposure to depressed maternal mood itself was associated with an increased risk for PNA symptoms³⁷ and maternal anxiety was associated with increased internalizing behaviors in 4- to 5-year-olds.⁵ However, higher levels of externalizing behaviors have been reported by depressed mothers but not by teachers.³⁸ Moreover, when prenatal maternal anxiety is accounted for, prenatal depression does not independently affect development.^9,39

Our findings have potential implications for pediatricians, family physicians, and psychiatrists alike. Child development following prenatal psychotropic medication exposure occurs in the context of maternal mental illness, a substantial risk factor for problems in child development. Four years after treatment for perinatal mental illness, maternal symptoms still persisted; even concurrent pharmacotherapy did not appear to confer protection against the ongoing influence of perinatal mental illness on early child behavior. While there are many reasons for increased externalizing behaviors at age 4 years, the association with maternal mood in this study remains an important reminder that close observation and mental health follow-up of both mother and child are essential to a healthy outcome. Perinatal mental health does not appear to be a time- or impact-limited condition.

A number of limitations to this study need to be mentioned. Our study cohort was small, and a larger sample will be needed to demonstrate group differences in behavior. Although we had a follow-up rate of approximately 50%, differences between those studied did not differ significantly from those not studied at the 4-year follow-up. In this study, we were unable to recruit and study children of depressed but not pharmacologically treated mothers; therefore, we could not directly compare the impact of maternal mood on child development with the effects of prenatal SSRI exposure. For logistical reasons, we also could not examine the impact of SSRI exposure in the absence of exposure to maternal depression. Maternal measures and a history of prenatal SSRI exposure were highly correlated; therefore, it remains difficult to draw definitive conclusions from individual measures treated as independent measures. Further, we were only able to study this cohort during the perinatal period and again at ages 4 to 5 years and could not assess the contribution of other early life exposures (such as social experience at home, preschool, etc) that may have also contributed to developmental outcomes. We only found associations between maternal observations and child behaviors. Because we could not confirm this with teacher reports, we could not independently confirm that this was not just an observatory bias. However, it could have been that our preschool observations may have been incomplete, as preschools run twice or thrice a week for 2.5 hours and some teachers may not have observed the dysfunctional behavior. Given that only 1 drug level was studied, it remains difficult to know how this relates to other levels during pregnancy and beyond or whether these levels were associated with symptomatic levels of depressed mood. The clinical use of SSRI drug levels remains one of the key questions facing clinicians and researchers alike. Finally, given that maternal depression is a common correlate of preschool behavioral problems,⁴⁰ there is a close relationship between maternal depression and maternal report of child behavior. One needs to ask whether we truly observed an increased level of aggression or whether we were just observing lower maternal thresholds for tolerating preschool behavior.

In summary, our findings suggest that SSRI exposure in the second and third trimesters was not independently associated with increased parental reports of externalizing behaviors and laboratory-based observed behaviors, even when accounting for maternal mood. Importantly, depressed maternal mood at the 4 years' postpartum assessment, even following long-term perinatal SSRI therapy, was associated with increased reports of externalizing behaviors at age 4 years. Prenatal SSRI dose and length of prenatal drug exposure were not independently associated with behavioral outcomes whereas elevated umbilical cord drug levels were associated with increased parental reports of externalizing behaviors. This was also reflected in our findings showing that increased levels of laboratory-based observed measures of aggression were associated with both a history of PNA and increased maternal reports of parental stress at the 4-year follow-up. These findings highlight the importance of recognizing and treating maternal depression associated with pregnancy. Although SSRIs or other antidepressants may reduce symptoms, these findings point to the need to apply other strategies to reduce or eliminate the impact of maternal mental illness on child development and behavior.

Correspondence: Tim F. Oberlander, MD, FRCPC, Early Human Experience Unit, Centre for Community Child Health Research, Room L408, 4480 Oak St, Vancouver, British Columbia, Canada V6 3V4 (toberlander@cw.bc.ca).

Accepted for Publication: August 4, 2006.

Author Contributions:Study concept and design: Oberlander, Reeybe, Misri, and Grunau. Acquisition of data: Oberlander, Reeybe, Misri, Kim, and Grunau. Analysis and interpretation of data: Oberlander, Reeybe, Papsdorf, Kim, and Grunau. Drafting of the manuscript: Oberlander, Reeybe, Misri, Papsdorf, and Grunau. Critical revision of the manuscript for important intellectual content: Oberlander, Reeybe, Misri, Papsdorf, Kim, and Grunau. Statistical analysis: Oberlander, Papsdorf, and Grunau. Obtained funding: Misri. Administrative, technical, and material support: Reeybe and Kim. Study supervision: Oberlander, Misri, and Reeybe.

Financial Disclosure: None reported.

Funding/Support: This work was supported by research grant BCM96-152 from the British Columbia Medical Services Foundation and by unrestricted funding (Dr Misri) for this study (protocol number 29060/830) from GlaxoSmithKline, Inc, Canada. Dr Oberlander is supported by a Human Early Learning Partnership (Graduate Studies, University of British Columbia) Senior Career Award and has the R. Howard Webster Professorship in Child Development (Faculty of Graduate Studies, University of British Columbia).

Acknowledgment: We are grateful to the mothers and their children for the time and effort in participating in our study, Dan Rurak, PhD, and Wayne Riggs, PhD, for the pharmacological data analysis, Ursula Brain, BA, and Colleen Fitzgerald for their thoughtful and dedicated work in making this study possible, and Kristin Kendrick, BA, for data processing. We are also grateful to the anonymous reviewers of the ARCHIVES, who provided thoughtful and important comments that improved the manuscript.