Serotonin transporter gene polymorphism and susceptibility to a home-visiting maternal-infant attachment intervention delivered by community health workers in South Africa: Reanalysis of a randomized controlled trial
In light of genetic information Barak Morgan and colleagues re-analyze data from a randomized controlled trial to explore the genetic susceptibility to a psychosocial intervention.
Early childhood development is increasingly recognized as a public health priority that requires attention and investment, and specific targets and indicators addressing this area are included in the recent Sustainable Development Goal (SDG) framework and the United Nations Secretary General’s Global Strategy for Women’s, Children’s and Adolescents’ Health [1,2]. With the knowledge that more than 250 million children younger than 5 y globally will fail to reach their full developmental potential , clear recognition of the damaging effects of poverty on early childhood development has fueled an interest in psychosocial interventions aimed at mitigating these harmful consequences in order to promote lifelong health and prosperity.
Gains in child development have been generated by psychosocial interventions to improve child nutrition and development and to address the mental health of caregivers . Interventions in the early years are cost-effective , can reduce inequity , and have been shown to have an impact on adult health outcomes . Parenting programs focused on early child development are a good example of a delivery mechanism for the prevention and reduction of childhood disadvantage . However, the efficacy of early child development interventions has, for the most part, been quantified by averaging individual outcomes across an entire sample and then, to varying extents, controlling for other factors, mostly of an extrinsic nature (e.g., maternal, environmental, and demographic factors). However, if individual differences in susceptibility to intervention are not considered (e.g., temperament, biological stress reactivity, and genetic differences), averaging of outcomes could lead to a misleading assessment of efficacy, because weaker effects on less susceptible individuals would dilute the estimation of effects on those more liable to respond [9,10]. In order to optimize the benefit of interventions, accurate evaluation of efficacy—including for whom the intervention does and does not work—is essential.
In 2003, Caspi et al.  first suggested that there were genetic differences in susceptibility to environmental influence in relation to depression. They reported observational findings indicative of a gene × environment (G×E) interaction. When individuals had experienced childhood maltreatment or stressful life events in early adulthood, carriers of the short allele (short/short or short/long genotype) of a length polymorphism in the regulatory region of the serotonin transporter gene (termed the 5HTT-linked polymorphic region [5HTTLPR]) were more liable than carriers of two long alleles (long/long genotype) to suffer from depression . Subsequent studies have continued to yield evidence of G×E interaction at the 5HTTLPR locus .
While this evidence is important, only recently have investigators begun to make use of interventions where participants are randomly allocated to different environmental conditions to test the causal status of G×E interactions. Randomized controlled trials (RCTs) have been estimated to have 13 times the statistical power of G×E studies in which individuals (and therefore genotypes) are not randomly assigned to categorically different environments (e.g., intervention and no intervention) . RCTs have, inter alia, two other big advantages: precluding gene–environment correlations (where genes and environments “choose” one another) and overcoming many of the problems with accurately measuring the environment of interest (e.g., maltreatment of children) . A recent meta-analysis of 11 field RCTs investigated gene × intervention (G×I) interactions involving several genes thought to confer greater susceptibility to interventions, including 5HTTLPR . The meta-analysis showed that intervention benefits were significantly stronger in those with susceptible genotypes, with an overall odds ratio (OR) of 3.17 (p < 0.01) for individuals with susceptible genotypes, compared to only 1.16 (p = 0.6) among individuals with nonsusceptible genotypes. Although striking, this meta-analytic evidence of genetic differential susceptibility to intervention rests upon an empirical base beset by several limitations. These include reliance on small sample sizes and special populations (e.g., maltreated children [13–15] or stroke patients ), mixed results for some genes , lack of ethnic and socioeconomic diversity (mostly middle-class white individuals) , and inconsistencies across different ethnic groups . Crucially, given the interest in tackling poverty through early child development interventions in low- and middle-income countries (LMICs), it is striking that, to the best of our knowledge, no study has investigated differential susceptibility to intervention outside the US and Europe.
Security of attachment, which can be objectively and reliably assessed in infancy , is an important indicator of positive early socio-emotional development [19–21]. A range of studies have shown that, compared to insecure attachment, secure attachment is associated with better subsequent outcomes, including reduced externalizing behavior problems and better social competence [22,23]. There is also emerging evidence that secure infant attachment and the sensitive maternal care that promotes it are linked to better growth, physical health, and cognitive development [19–21,24]. Promoting secure mother–infant attachment is therefore an important focus for prevention studies, and, indeed, a wide range of interventions have been developed that appear to be successful in promoting secure attachment . The majority of these interventions target the responsiveness of the mother’s caregiving behavior in relation to the infant’s attachment cues and communications, and have been delivered as primary or secondary prevention in a wide range of contexts within high-income countries.
Here, we report the results of a reanalysis of data from a RCT in which we test for a G×I interaction in early child development. Our study focuses on 5HTTLPR as a potential genetic moderator of the efficacy of a home-visiting intervention that was designed to improve attachment in mother–infant dyads in an impoverished isiXhosa-speaking community in South Africa. The intervention, known as Thula Sana (“hush baby” in isiXhosa), was a manualized home-visiting parenting program that aimed to promote security of infant attachment (the primary outcome of the original trial) by enhancing maternal sensitivity to infant characteristics and communication and by supporting management of infant distress [26,27].
This intervention was evaluated in an individually randomized controlled trial over a 4-y period, beginning in 1999, with a sample of 449 pregnant women. At infant age 18 mo, infant attachment status was assessed using standardized laboratory procedures . Compared to controls who received no intervention, infants in the intervention group were significantly more likely to be securely attached to their primary caregiver (OR = 1.7, p = 0.029, 95% CI [1.06, 2.76]) . This result equates to an effect size of Cohen’s d = 0.29, consistent with previous reports of modest effect size estimates for such interventions . Further results on maternal sensitivity and maternal depression are reported in Cooper et al. . However, this first report of the trial did not take into account the issue of differential susceptibility, and it is therefore possible that differences in efficacy for susceptible and nonsusceptible individuals may have been overlooked.
A follow-up study of the original Thula Sana cohort at 13 y of age provided the opportunity to address the possibility of genetic differential susceptibility by collecting DNA from both children and mothers. We focused on 5HTTLPR, the polymorphism most frequently investigated with respect to attachment outcomes and related processes in G×E studies [13,28–31]. To date, all studies implicating 5HTTLPR in genetic differential susceptibility to environment for attachment outcomes have been observational and have yielded mixed results (for reviews, see [17,32]). The current study circumvented the inherent limitations of observational research designs [9–11] by testing for a G×I interaction between 5HTTLPR and a home-visiting intervention on attachment security. On the basis of previous studies implicating the short 5HTTLPR allele as the “susceptibility allele” [11,17], genetic differential susceptibility to the intervention was predicted to be high for children carrying at least one short allele and low for children carrying two long alleles.
The Health Research Ethics Committee of Stellenbosch University approved this study (Ethics Reference #S12/04/113). Adult caregivers provided written consent for their and their child’s participation, and adolescents signed assent forms prior to participating in the study.
This is a reanalysis of results from the original Thula Sana RCT. In the original trial, mothers were randomized during pregnancy to receive the Thula Sana intervention or usual care during pregnancy and the first 6 mo after birth. The primary outcomes of the original trial were maternal sensitivity and infant attachment security. The aim of this investigation was to test whether 5HTTLPR genotype moderated the intervention effect on infant attachment security measured at 18 mo. This report presents a reanalysis of the original trial’s primary outcome using genetic information gathered at 13 y.
Between April 1999 and February 2003, pregnant mothers from a racially and ethnically homogeneous black, isiXhosa-speaking population inhabiting two geographical areas within Khayelitsha were enrolled in the original Thula Sana study . We made efforts to identify and recruit women who were in their last trimester of their pregnancy (on the basis of the accounts of their gestation that women had received from antenatal clinics). Throughout the recruitment period, over 22 mo, a research assistant regularly visited all the homes door-to-door in both areas to inquire whether anyone had become pregnant or a pregnant woman had moved into the area, and to invite identified women to participate in the study. We identified a consecutive series of 452 women as pregnant within the study area and invited them to take part in the study. Of these, three refused to participate. We then assigned the remaining 449 women to the intervention or control group using minimization, balancing for antenatal depression, whether or not the pregnancy was planned, and which area within Khayelitsha they lived in. In the original trial, they were assessed antenatally and at 2 mo, 6 mo, 12 mo, and 18 mo after birth. The maternal socioeconomic profile for this sample at the time of antenatal interview was as follows: 85% lived in informal housing (shacks), 89% had no formal employment, 44% had no electricity, and 39% had no running water in their home . Later, from December 2012 to June 2014, we enrolled the sample for a long-term 13-y follow-up.
Lay community health workers, themselves all mothers, were selected from the local community, underwent an 8-wk training on delivering the intervention, and were given weekly support and supervision throughout the intervention period. The intervention began in the last trimester of pregnancy, and continued until 6 mo postpartum, during which a total of 16 visits of 1 h each were delivered . The intervention was designed to be suitable for routine delivery within low-resource settings. The content was based closely on The Social Baby , but it also incorporated the key principles of the World Health Organization’s report Improving the Psychosocial Development of Children  and the use of items from the Neonatal Behavioral Assessment Scale , to sensitize the mother to her infant’s individual capacities and needs. Women in the control group received standard services provided by the local infant clinic as well as fortnightly home visits by a community health worker who assessed the physical and medical progress of mothers and infants (women in the intervention group received these services as well).
Full data collection procedures for the early trial were reported in the first outcome paper . In the follow-up study, children and their caregiver were assessed at the Prevention Research for Community, Family and Child Health study center located in Khayelitsha for approximately 4 h. Only limited and out-of-date address information was available from the original study, and many of the names of areas and roads in the informal parts of Khayelitsha had changed in the period between the original study and the reenrollment period. In addition to going door-to-door to find participants at their old addresses, reenrollment strategies also included engaging local community structures. Most participants were still resident in the area, but one-quarter had migrated to other parts of the country since the infant age 18 mo assessment, with participants located in five different provinces of the country. Wherever possible, the team arranged for these child and mother participants to travel to Cape Town. However, there was a small subgroup of participants who were not able to travel to Cape Town. In these cases, a data collection team travelled to their homes for assessment purposes. At the time of assessment, saliva for DNA extraction was collected from children and whenever possible from their biological mothers as well.
At 18 mo of age, infant attachment status was assessed in 76% of the original 449 mother–infant pairs using the standardized strange situation procedure (SSP) developed by Ainsworth and colleagues [18,26], and used extensively in research in both high-income countries and LMICs [36,37]. To date, the attachment status of children in LMICs has received little research attention, but the construct of attachment and the SSP have been shown to be valid cross-culturally . To the best of our knowledge, there have been only two previous studies assessing attachment in Africa using the procedure [24,38].
At 18 mo, the primary caregiver who participated in the assessment was in all cases the biological mother. To conduct the SSP, the infant was filmed through a one-way mirror in an unfamiliar playroom over a 21-min period divided into seven 3-min episodes, including two episodes of separation and reunion with the mother. M. T. rated the video tapes for security of infant attachment, having been trained to criterion by an established US training program. He used the ABCD coding system to rate infants as secure (B) or as one of three categories of insecure (A, avoidant; C, anxious-resistant; or D, disorganized). These ratings were made blind to all other information about the infants and their mothers. Reliability was confirmed by assessing agreement between M. T. and a second trained UK rater on 16 tapes (four-way κ = 0.96). In the original trial, a total of 318 infants were successfully assessed in the SSP, 265 of whom were amongst the 334 children reassessed at 13 y of age. Of these 265, 40 had been classified as avoidant, 182 secure, 21 resistant, and 22 disorganized. In keeping with the literature and in order to maximize cell sizes in the analysis, we restricted our analyses to the binary distinction between secure and insecure classifications, where insecure included all three insecure categories (A/C/D) pooled together.
At the 13-y follow-up, for noninvasive collection of high-quality DNA, 2 ml of saliva was collected by trained and supervised data collectors using Oragene DNA OG-500 (DNA Genotek) saliva self-collection kits at the research center or at participants’ homes. Oragene kits were stored at room temperature and shipped to Germany for molecular genetic analysis. DNA was extracted from saliva samples and purified according to the kit protocol. All samples passed initial quality control, with OD260/OD280 ratios between 1.6 and 2.0. Participants were genotyped for the 43-bp insertion/deletion polymorphism in the regulatory promoter region of the serotonin transporter gene (5HTTLPR) with a standard PCR procedure, as previously described . There was no deviation from Hardy–Weinberg equilibrium (χ2 = 1.13, p = 0.29).
The current report tested a single a priori hypothesis that 5HTTLPR genotype, operationalized as the presence versus absence of the short form of 5HTTLPR, would moderate the intervention effect on the primary outcome (secure versus insecure attachment). Security of attachment cannot be measured in children under 11 mo, and therefore it was measured only at the post-intervention follow-up when the infants were 18 mo of age. The analysis was therefore a single-level (i.e., not repeated measure) logistic regression, with the hypothesized moderating effect specified as an intervention group × 5HTTLPR genotype interaction. The primary analysis was conducted without adjustment for covariates, but sensitivity analyses were also conducted, adjusting for covariates; in these analyses we also assessed the impact of missing data (for individuals without both attachment and genetic data, including all individuals lost to follow-up) using multiple imputation, as recommended by a reviewer. We used the fully conditional specification approach to multiple imputation, which is a highly flexible approach capable of accounting for nonlinearity in the relationship between covariates and outcome and which fits an imputation model that is consistent with the substantive model (i.e., explicitly includes the gene and intervention main effects and interaction). Multiple imputations included all model variables, maternal 5HTTLPR genotype, and the only baseline measures that were associated with missingness (time to entry into the trial from start of recruitment and whether the house had electricity and water). Imputation was conducted using the package smcfcs  and Stata’s MI procedure based on 100 imputed samples. Details of the imputation are provided in S1 Text and S2 Data.
From December 2012 to June 2014, we reenrolled 334 (74.1%) of the children (162 intervention, 172 control; 166 males, 168 females) from the original sample of 449 mother–child pairs. At 13 y of age, 115 of the original 449 children were lost to follow-up. Of these children, 24 had died since the original randomization process. The remaining 91 could not be contacted. Derivation of the sample used in this study is depicted in Fig 1.
Of the 334 adolescent participants at 13-y follow-up, 279 (131 males, 148 females; 134 intervention, 145 control) provided DNA samples, all of which yielded 5HTTLPR genotype results. There were 220 (104 males, 116 females; 110 intervention, 110 control) adolescents for whom there were both 5HTTLPR genotype and attachment security data. 5HTTLPR genotype results for the 220 adolescents with attachment security data are shown in Table 1. Of these 220 adolescents, 185 (89 males, 96 females; 97 intervention, 88 control) had mothers for whom 5HTTLPR genotype data were available. All analyses were performed on these 220 adolescents and 185 adolescent–mother pairs. No individuals changed from the intervention to the control arm or vice versa at any point in the trial, and no individuals with both genotype and attachment security data were excluded from analysis. The 220 participants with both genotype and attachment data were compared to the rest of the original sample of 449 on a range of demographic and socioeconomic variables. As shown in Table 2, with two exceptions—water and electricity in the home—there were no significant differences between the two groups (See S3 Table for actual values). There were no significant differences on any of the variables between the 110 adolescent participants in the intervention and control groups (S4 Table).
Gene × intervention interaction
Because the presence of at least one short 5HTTLPR allele frequently confers susceptibility to environmental influence [11,17], individuals with short/long and short/short genotypes were treated as one genotype category. Individuals carrying at least one short allele comprised 40% of the 220 participants (Table 1). Logistic regression revealed a significant G×I interaction: for infant security of attachment, the efficacy of the intervention varied as a function of serotonin transporter genotype (OR = 4.07, p = 0.028, 95% CI [1.16, 14.20]). As shown in Table 3 and in Fig 2, for those with the susceptible genotype (short/long and short/short), the intervention increased the odds of secure infant attachment nearly 4-fold relative to controls (OR = 3.86, p = 0.008, 95% CI [1.42, 10.51], d = 0.75). By contrast, for those with the nonsusceptible genotype (long/long), the intervention had no impact on the odds of secure attachment relative to controls (OR = 0.95, p = 0.89, 95% CI [0.45, 2.01], d = 0.03). Expressed in terms of absolute risk, for those with the short allele, the probability of secure attachment being observed in the intervention group was 84% (95% CI [73%, 95%]), compared to 58% (95% CI [43%, 72%]) in the control group. For those with two copies of the long allele, the probability of being secure was 70% (95% CI [59%, 81%]) in the intervention group, compared to 71% (95% CI [60%, 82%]) in the control group (Table 4; Fig 3). The results show that, on average, individuals carrying at least one short allele were susceptible to the intervention and those carrying two long alleles were nonsusceptible.
The efficacy of the home-visiting intervention on the attachment outcome in terms of percentage secure and insecure individuals according to group and genotype is shown in Fig 3. From left to right in Fig 3, for individuals carrying at least one short allele, the percentage showing secure attachment was 84% and 58% in the intervention and control groups, respectively. For individuals carrying two long alleles, the percentage showing secure attachment was 71% and 70% in the intervention and control groups, respectively. In the absence of genetic information, when results are averaged over all individuals and all genotypes (“unknown”), the apparent percentage of individuals showing secure attachment was 75% and 65% in the intervention and control groups, respectively. The numbers of individuals in each group and in each genotype category are given in Table 1.
The above logistic regression analysis was rerun controlling for covariates observed to be different between the two groups. The result was not affected by sex or any other covariates (Table 2).
Further, to address the possibility that the observed interaction effect was attributable to maternal rather than child genotype, we reran the logistic regression including terms for maternal 5HTTLPR genotype and the interaction between maternal 5HTTLPR and group (intervention versus control) in the model. The child 5HTTLPR × group interaction remained significant (OR = 4.8, p = 0.041), while neither the main effect of maternal 5HTTLPR (OR = 0.96, p = 0.93) nor its interaction with group (OR = 1.97, p = 0.37) was significant.
Finally, multiple imputation analyses based on 100 imputed samples of n = 499 (the total number of mother–infant dyads originally randomized) were also run to check the robustness of the result. These analyses confirmed the 5HTTLPR × group interaction in the logistic regression analysis, with the interaction OR equal to −1.41 (standard error = 0.64, p = 0.029, 95% CI [−2.68, −0.015]).
The current study aimed to test the hypothesis that 5HTTLPR genotype would moderate the impact of an early child development intervention aimed at promoting the security of mother–infant attachment in a middle-income country. Our reanalysis of the original trial in light of recently acquired genetic information provided support for this hypothesis. Specifically, for children with one or two copies of the short allele of 5HTTLPR, the intervention appeared to be highly effective in improving the rate of attachment security (from 58% in the control group to 84% in the intervention group), but for those with only the long allele, the intervention led to no measurable benefits (secure attachment rate 70% in the control group versus 71% in the intervention group).
There are few studies, and none outside the US and Europe, that have used the framework of experimental trials to test for G×E interaction in early child development . In the specific area of attachment, which is a key domain of psychosocial functioning among young children, we are aware of only two other studies of G×I interaction. In a study of maltreated children who were randomized either to a parenting intervention or to a control condition, Cicchetti and colleagues  found no G×I interaction for 5HTTLPR in relation to attachment. There are, however, important differences between this study and our own that may account for the difference in findings. First, the Cicchetti et al. study involved a smaller sample size (in total, ignoring genotype: 49 intervention, 47 control) than our study (110 intervention, 110 control). Second, their sample was racially and ethnically diverse, with the frequency of short allele carriers differing markedly between the black (45.9%), white (78.6%), and other/multiracial (67.5%) categories. By contrast, our study sample was drawn from an ethnically homogeneous population. Third, maltreated children represent a special population that is not comparable to the community sample included in the Thula Sana study. In that regard, it is notable that the disorganized class of insecure attachments, which carries the highest clinical risk, was present in 88% of the maltreated sample and in only 8% of the Thula Sana sample, a figure typical of community samples. Given these important sample differences, little can be concluded from the difference in G×I findings between the two studies. The only other study in this area also relied on a special population sample, in this case, children raised in Romanian institutions. The Bucharest Early Intervention Project randomly allocated institutionalized children to either high-quality foster care or continuing institutional care before 30 mo of age. At 54 mo of age, among children carrying the short/short 5HTTLPR genotype, relative to the outcome of those in the continuing institutional care group, those provided with high-quality foster care had lower symptom levels of attachment disorder (specifically “indiscriminate social behavior”). For the children with at least one long allele (short/long and long/long), there was no difference in terms of attachment disorder symptoms between foster and institutional care conditions .
As Belsky recently observed, attachment research, like much research on early child development, has proceeded for the most part with the assumption that all children are equally susceptible to the effects of sensitive and insensitive care . The current findings suggest otherwise and highlight the significance of genetic differential susceptibility in shaping developmental trajectories during early infancy.
An important limitation of this study is that we were not able to follow up all of the individuals from the original trial, and there were missing data for attachment and genotype. In total, our primary analysis included 49% (220/449) of the original sample of children whose mothers were randomized to intervention and control conditions. Although the intervention and control groups were highly similar in our follow-up sample, and the follow-up sample was generally very similar to the original sample, there was some evidence of selective loss to follow-up on two variables (Table 2). This means that randomization within our follow-up sample may have been imperfect. Attribution of the primary outcome to causal effects of the intervention in the present sample should therefore be treated with caution.
Another limitation of this study is its focus on only one gene. Despite extensive evidence from research using both observational and experimental approaches showing that the 5HTT promoter polymorphism influences organisms’ sensitivity to environmental influences , and despite evidence that 5HTTLPR influences the development of functional and structural brain networks involved in emotion regulation, stress processing, and threat sensitivity , it is unlikely that one single gene will explain all individual differences in intervention efficacy. Rather, it can be assumed that differential susceptibility to environmental influences is a complex, polygenic trait influenced by the combination of hundreds of common genetic variants of small effect. The first studies in the field of therapygenetics have started using genome-wide approaches  and polygenic scoring , which allow researchers to aggregate the effects of multiple variants. The use of these approaches in sufficiently large samples, much larger than the present study, could open new avenues in G×I interaction research.
A final limitation is that our attachment finding could be culturally-specific and therefore not generalizable, and certainly the study needs replicating in other cultural contexts. For example, in contexts such as the one described here, where there are multiple caregivers, infants and children are able to develop attachments to more than one person, and when the attachment status is discordant between different caregivers, it remains unclear what the longer-term outcomes are . Nevertheless, we believe it is unlikely that our results would be confined to this cultural context. First, both this sample and one we previously studied in Khayelitsha  showed a distribution of attachment categories similar to that in high-income countries. Second, our previous research showed the same association between attachment security and the main parenting antecedents that we would expect (i.e., sensitivity and lack of intrusiveness) from a substantive body of research globally .
Beyond illuminating the role of genetic differential susceptibility in early childhood development, the current finding also speaks to a fundamental issue in the quest to understand and mitigate the developmental effects of poverty through psychosocial intervention. The near-large effect size reported here for the intervention in children with susceptible genotypes (d = 0.75) is at variance with the general conclusion that psychosocial interventions in the context of poverty produce only small to medium effect sizes . Without taking account of genetic susceptibility, it is possible that other intervention studies have, at least in some subpopulations, underestimated the impact of their interventions, as we originally did. By the same token, as was originally reported for Thula Sana , other studies might also have underestimated the negative impact on susceptible subpopulations of not receiving an intervention (Figs 2 and 3). In short, averaging outcomes across all participants may well lead to an invalid conclusion about the efficacy of an intervention .
The launch of the SDGs and the Global Strategy for Women’s, Children’s and Adolescents’ Health in late 2015 has focused attention on a life-course perspective towards the understanding of child and adolescent development—the “thrive agenda” [48,49]. To stand a chance of meeting the ambitious SDGs and Global Strategy targets by 2030, an enhanced understanding will be required of the biological and psychological mechanisms underlying interventions aimed at improving the lives of young children. In the context of the resource constraints that characterize LMICs, ensuring that psychosocial interventions are implemented in the most efficacious manner will take on an added urgency. In this regard, it is instructive to note a parallel between genetic differential susceptibility to psychosocial interventions and genetic differential susceptibility in the emergence of personalized medicine, specifically pharmacogenomics. Just as genetic information is being used to guide the choice of medication for different individuals diagnosed with the same condition (e.g., ), it has been suggested that in a world of limited resources, psychosocial interventions could, once more is known, also be selectively targeted at genetically susceptible individuals [41,51]. This possibility would precipitate the daunting moral challenge of balancing equity (equal treatment for all) and efficacy (treating only those likely to benefit) [41,47].
However, while such targeting in LMICs is technically feasible, provision of intervention services on the basis of genotyping is not currently a realistic prospect. First, as already noted above, genetic prediction of intervention efficacy based on variation at one gene locus is far from sufficiently sensitive or specific to provide a reliable basis for intervention recommendations. Second, quite apart from the science, the prospect of discriminating individuals on the basis of their genetic makeup is controversial and likely to encounter strenuous social resistance.
Nevertheless, other avenues of investigation do suggest themselves. A promising approach might be to incorporate intermediate phenotypes, such as easily accessible physiological or temperamental characteristics, with genetic and epigenetic markers  to improve prediction by use of multiple data types. Physiological measures might include hormonal and/or sympathetic and parasympathetic nervous system stress markers ; relevant temperamental characteristics could include emotion regulation abilities  or approach avoidance tendencies . Indeed, the short/long 5HTTLPR polymorphism has been associated with individual differences in epigenetic methylation , stress physiology [57–59], and temperament [60,61]. A combination of biological and behavioral markers could be used to identify meaningful subgroups and thus target interventions to those likely to respond. Moreover, such measures could be used not only to better target interventions to those likely to respond, but also to clarify where new or additional interventions are required.
In summary, despite a considerable body of evidence on how cumulative risk is implicated in poor child development, our understanding of pathways and mechanisms, and how dose, timing, and adversity impact on outcome, is to date quite limited . Measuring genetic susceptibility together with epigenetic, physiological, temperamental, and behavioral markers in RCTs will allow better examination and greater insight into these mechanisms and pathways in LMICs. This could enhance our understanding of why certain individuals do not respond to a particular treatment and facilitate the development of new interventions for them.
1. Chan M. Linking child survival and child development for health, equity, and sustainable development. Lancet. 2013;381(9877):1514–5. doi: 10.1016/S0140-6736(13)60944-7 23642687
2. Lake A, Chan M. Putting science into practice for early child development. Lancet. 2014;385(9980):1816–7. doi: 10.1016/S0140-6736(14)61680-9 25245180
3. Black MM, Walker SP, Fernald LC, Andersen CT, DiGirolamo AM, Lu C, et al. Early childhood development coming of age: science through the life course. Lancet.2017;389(10064):77–90. doi: 10.1016/S0140-6736(16)31389-7 27717614
4. Shonkoff JP. Building a new biodevelopmental framework to guide the future of early childhood policy. Child Dev. 2010;81(1):357–67. doi: 10.1111/j.1467-8624.2009.01399.x 20331672
5. Heckman JJ. Skill formation and the economics of investing in disadvantaged children. Science. 2006;312(5782):1900–2. doi: 10.1126/science.1128898 16809525
6. Marmot M, Friel S, Bell R, Houweling TA, Taylor S, Commission on Social Determinants of Health. Closing the gap in a generation: health equity through action on the social determinants of health. Lancet. 2008;372(9650):1661–9. doi: 10.1016/S0140-6736(08)61690-6 18994664
7. Campbell F, Conti G, Heckman JJ, Moon SH, Pinto R, Pungello E, et al. Early childhood investments substantially boost adult health. Science. 2014;343(6178):1478–85. doi: 10.1126/science.1248429 24675955
8. Knerr W, Gardner F, Cluver L. Improving positive parenting skills and reducing harsh and abusive parenting in low-and middle-income countries: a systematic review. Prev Sci. 2013;14(4):352–63. doi: 10.1007/s11121-012-0314-1 23315023
9. van Ijzendoorn MH, Bakermans-Kranenburg MJ. Genetic differential susceptibility on trial: meta-analytic support from randomized controlled experiments. Dev Psychopathol. 2015;27(01):151–62.
10. Bakermans-Kranenburg MJ, Van IJzendoorn MH. The hidden efficacy of interventions: gene×environment experiments from a differential susceptibility perspective. Annu Rev Psychol. 2015;66:381–409. doi: 10.1146/annurev-psych-010814-015407 25148854
11. Caspi A, Sugden K, Moffitt TE, Taylor A, Craig IW, Harrington H, et al. Influence of life stress on depression: moderation by a polymorphism in the 5-HTT gene. Science. 2003;301(5631):386–9. doi: 10.1126/science.1083968 12869766
12. Karg K, Burmeister M, Shedden K, Sen S. The serotonin transporter promoter variant (5-HTTLPR), stress, and depression meta-analysis revisited: evidence of genetic moderation. Arch Gen Psychiatry. 2011;68(5):444–54. doi: 10.1001/archgenpsychiatry.2010.189 21199959
13. Cicchetti D, Rogosch FA, Toth SL. The effects of child maltreatment and polymorphisms of the serotonin transporter and dopamine D4 receptor genes on infant attachment and intervention efficacy. Dev Psychopathol. 2011;23(02):357–72.
14. Drury SS, Gleason MM, Theall KP, Smyke AT, Nelson CA, Fox NA, et al. Genetic sensitivity to the caregiving context: the influence of 5httlpr and BDNF val66met on indiscriminate social behavior. Physiol Behav. 2012;106(5):728–35. doi: 10.1016/j.physbeh.2011.11.014 22133521
15. Brett ZH, Humphreys KL, Smyke AT, Gleason MM, Nelson CA, Zeanah CH, et al. 5HTTLPR genotype moderates the longitudinal impact of early caregiving on externalizing behavior. Dev Psychopathol. 2015;27(1):7–18. 25640827
16. Kohen R, Cain KC, Buzaitis A, Johnson V, Becker KJ, Teri L, et al. Response to psychosocial treatment in poststroke depression is associated with serotonin transporter polymorphisms. Stroke. 2011;42(7):2068–70. 21847802
17. Van IJzendoorn M, Belsky J, Bakermans-Kranenburg M. Serotonin transporter genotype 5HTTLPR as a marker of differential susceptibility? a meta-analysis of child and adolescent gene-by-environment studies. Transl Psychiatry. 2012;2(8):e147.
18. Ainsworth MDS, Blehar MC, Waters E, Wall SN. Patterns of attachment: a psychological study of the strange situation. Hove (United Kingdom): Psychology Press; 2015.
19. Zaslow MJ, Weinfield NS, Gallagher M, Hair EC, Ogawa JR, Egeland B, et al. Longitudinal prediction of child outcomes from differing measures of parenting in a low-income sample. Dev Psychol. 2006;42(1):27–37. doi: 10.1037/0012-16126.96.36.199 16420116
20. Pettit GS, Bates JE, Dodge KA. Supportive parenting, ecological context, and children’s adjustment: a seven-year longitudinal study. Child Dev. 1997;68(5):908–23.
21. World Health Organization. The importance of caregiver-child interactions for the survival and healthy development of young children: A review. Geneva: World Health Organization; 2004.
22. Fearon R, Bakermans-Kranenburg MJ, Van IJzendoorn MH, Lapsley AM, Roisman GI. The significance of insecure attachment and disorganization in the development of children’s externalizing behavior: a meta-analytic study. Child Dev. 2010;81(2):435–56. doi: 10.1111/j.1467-8624.2009.01405.x 20438450
23. Groh AM, Fearon RP, Bakermans-Kranenburg MJ, Van IJzendoorn MH, Steele RD, Roisman GI. The significance of attachment security for children’s social competence with peers: a meta-analytic study. Attach Hum Dev. 2014;16(2):103–36. doi: 10.1080/14616734.2014.883636 24547936
24. Tomlinson M, Cooper P, Murray L. The mother–infant relationship and infant attachment in a South African peri-urban settlement. Child Dev. 2005;76(5):1044–54. doi: 10.1111/j.1467-8624.2005.00896.x 16150001
25. Bakermans-Kranenburg MJ, Van Ijzendoorn MH, Juffer F. Less is more: meta-analyses of sensitivity and attachment interventions in early childhood. Psychol Bull. 2003;129(2):195–215. 12696839
26. Cooper PJ, Tomlinson M, Swartz L, Landman M, Molteno C, Stein A, et al. Improving quality of mother–infant relationship and infant attachment in socioeconomically deprived community in South Africa: randomised controlled trial. BMJ. 2009;338:b974. doi: 10.1136/bmj.b974 19366752
27. Mother-Infant Intervention Programme for the Khayelitsha Treatment Trial. http://www.reading.ac.uk/web/files/cls/Khayelitsha_manual.pdf.
28. Pauli-Pott U, Friedl S, Hinney A, Hebebrand J. Serotonin transporter gene polymorphism (5-HTTLPR), environmental conditions, and developing negative emotionality and fear in early childhood. J Neural Transm (Vienna). 2009;116(4):503–12.
29. Barry RA, Kochanska G, Philibert RA. G× E interaction in the organization of attachment: mothers’ responsiveness as a moderator of children’s genotypes. Journal of Child Psychology and Psychiatry. 2008;49(12):1313–20. doi: 10.1111/j.1469-7610.2008.01935.x 19120710
30. Spangler G, Johann M, Ronai Z, Zimmermann P. Genetic and environmental influence on attachment disorganization. J Child Psychol Psychiatry. 2009;50(8):952–61. 19673052
31. Luijk MP, Roisman GI, Haltigan JD, Tiemeier H, Booth-LaForce C, Van Ijzendoorn MH, et al. Dopaminergic, serotonergic, and oxytonergic candidate genes associated with infant attachment security and disorganization? In search of main and interaction effects. J Child Psychol Psychiatry. 2011;52(12):1295–307. doi: 10.1111/j.1469-7610.2011.02440.x 21749372
36. Fearon RMP, Belsky J. Precursors of attachment security. In: Cassidy J, Shaver PR, editors. Handbook of attachment: theory, research and clinical applications. 3rd edition. New York: Guilford Press; 2016. pp. 291–313.
37. Mesman J, van IJzendoorn MH, Sagi-Schwartz A. Cross-cultural patterns of attachment: universal and contextual dimensions. In: Cassidy J, Shaver PR, editors. Handbook of attachment: theory, research and clinical application. 3rd edition. New York: Guilford Press; 2016. pp. 852–77.
38. True McMahan M, Pisani L, Oumar F. Infant–mother attachment among the Dogon of Mali. Child Dev. 2001;72(5):1451–66. 11699681
39. Wendland J.R., Martin B.J., Kruse M.R., Lesch K.-P., Murphy D.L. Simultaneous genotyping of four functional loci of human SLC6A4, with a reappraisal of 5-HTTLPR and rs25531. Molecular Psychiatry; 2006:11, 224–226
40. Carpenter JR, Goldstein H, Kenward MG. REALCOM-IMPUTE software for multilevel multiple imputation with mixed response types. J Stat Softw. 2011;45(5):1–14.
41. Belsky J. Beyond vulnerability: attachment, adversity, gene–environment interaction, and implications for intervention. J Dev Behav Pediatr. 2015;36(6):464–6. doi: 10.1097/DBP.0000000000000184 26154715
42. Caspi A, Hariri AR, Holmes A, Uher R, Moffitt TE. Genetic sensitivity to the environment: the case of the serotonin transporter gene and its implications for studying complex diseases and traits. Focus. 2010;8(3):398–416.
43. Pezawas L, Meyer-Lindenberg A, Drabant EM, Verchinski BA, Munoz KE, Kolachana BS, et al. 5-HTTLPR polymorphism impacts human cingulate-amygdala interactions: a genetic susceptibility mechanism for depression. Nat Neurosci. 2005;8(6):828–34. doi: 10.1038/nn1463 15880108
44. Keers R, Coleman JR, Lester KJ, Roberts S, Breen G, Thastum M, et al. A genome-wide test of the differential susceptibility hypothesis reveals a genetic predictor of differential response to psychological treatments for child anxiety disorders. Psychother Psychosom. 2016;85(3):146–58. doi: 10.1159/000444023 27043157
45. Coleman JR, Lester KJ, Keers R, Roberts S, Curtis C, Arendt K, et al. Genome-wide association study of response to cognitive–behavioural therapy in children with anxiety disorders. Br J Psychiatry. 2016;209(3):236–43. doi: 10.1192/bjp.bp.115.168229 26989097
46. van IJzendoorn MH, Sagi A, Lambermon MW. The multiple caretaker paradox: data from Holland and Israel. New Dir Child Adolesc Dev. 1992;1992(57):5–24.
47. Ellis BJ, Boyce WT, Belsky J, Bakermans-Kranenburg MJ, Van IJzendoorn MH. Differential susceptibility to the environment: an evolutionary–neurodevelopmental theory. Dev Psychopathol. 2011;23(01):7–28.
48. Every Woman Every Child. The global strategy for women’s children’s and adolescent’s health (2016–2030). New York: United Nations; 2015 [cited 2017 Jan 20]. http://globalstrategy.everywomaneverychild.org/.
49. United Nations. Transforming our world: the 2030 agenda for sustainable development. New York: United Nations; 2015 [cited 2017 Jan 20]. https://sustainabledevelopment.un.org/post2015/transformingourworld.
50. Tse SM, Tantisira K, Weiss ST. The pharmacogenetics and pharmacogenomics of asthma therapy. Pharmacogenomics J. 2011;11(6):383–92. doi: 10.1038/tpj.2011.46 21987090
51. Belsky J. The differential susceptibility hypothesis: sensitivity to the environment for better and for worse. JAMA Pediatr. 2016;170(4):321–2. doi: 10.1001/jamapediatrics.2015.4263 26831915
52. Feder A, Nestler EJ, Charney DS. Psychobiology and molecular genetics of resilience. Nat Rev Neurosci. 2009;10(6):446–57. doi: 10.1038/nrn2649 19455174
53. Ellis BJ, Del Giudice M. Beyond allostatic load: rethinking the role of stress in regulating human development. Dev Psychopathol. 2014;26(01):1–20.
54. McRae K, Jacobs SE, Ray RD, John OP, Gross JJ. Individual differences in reappraisal ability: links to reappraisal frequency, well-being, and cognitive control. J Res in Pers. 2012;46(1):2–7.
55. Heuer K, Rinck M, Becker ES. Avoidance of emotional facial expressions in social anxiety: the approach–avoidance task. Behav Res Ther. 2007;45(12):2990–3001. doi: 10.1016/j.brat.2007.08.010 17889827
56. Palma-Gudiel H, Córdova-Palomera A, Leza JC, Fañanás L. Glucocorticoid receptor gene (NR3C1) methylation processes as mediators of early adversity in stress-related disorders causality: a critical review. Neurosci Biobehav Rev. 2015;55:520–35. doi: 10.1016/j.neubiorev.2015.05.016 26073068
57. Klucken T, Alexander N, Schweckendiek J, Merz CJ, Kagerer S, Osinsky R, et al. Individual differences in neural correlates of fear conditioning as a function of 5-HTTLPR and stressful life events. Soc Cogn Affect Neurosci. 2013;8(3):318–25. doi: 10.1093/scan/nss005 22258800
58. Gilissen R, Bakermans-Kranenburg MJ, van Ijzendoorn MH, Linting M. Electrodermal reactivity during the Trier Social Stress Test for children: interaction between the serotonin transporter polymorphism and children’s attachment representation. Dev Psychobiol. 2008;50(6):615–25. doi: 10.1002/dev.20314 18683185
59. Taylor SE, Way BM, Welch WT, Hilmert CJ, Lehman BJ, Eisenberger NI. Early family environment, current adversity, the serotonin transporter promoter polymorphism, and depressive symptomatology. Biol Psychiatry. 2006;60(7):671–6. doi: 10.1016/j.biopsych.2006.04.019 16934775
60. Weeland J, Slagt M, Brummelman E, Matthys W, de Castro BO, Overbeek G. 5-HTTLPR expression outside the skin: an experimental test of the emotional reactivity hypothesis in children. PLoS ONE. 2015;10(11):e0141474. doi: 10.1371/journal.pone.0141474 26560754
61. Gressier F, Calati R, Serretti A. 5-HTTLPR and gender differences in affective disorders: a systematic review. J Affect Disord. 2016;190:193–207. doi: 10.1016/j.jad.2015.09.027 26519640
62. Zeanah CH, Sonuga-Barke EJ. Editorial: the effects of early trauma and deprivation on human development—from measuring cumulative risk to characterizing specific mechanisms. J Child Psychol Psychiatry. 2016;57(10):1099–102. doi: 10.1111/jcpp.12642 27647049