Work package framework

FAMILY puts in place six empirical work packages (WPs3-8) to reach its objectives (Fig. 2). This WP framework is characterised by strong interconnections and synergistic methodological approaches between the proposed research in objectives 1–4 (WPs3-7). WPs3-5 work with human data and each focus on their own biological level, i.e., the genome (e.g., polygenic risk; WP3), epigenome (e.g., DNAm, microRNA; WP4), and the brain (e.g., magnetic resonance imaging (MRI); WP5). Information from these data domains may act as predictors, mediators, or moderators of intergenerational transmission of risk. Furthermore, to increase mechanistic understanding of the role of intergenerational transmission, WP6 will leverage two unique animal models: one where maternal disorder-like behaviour originates from postnatal severe stress, and one where parental disorder-like behaviour results from purely genetic risk. As in the human studies (i.e., WPs3-5), effects on behaviour and biology in the offspring will be investigated. WPs4-5 will share biological readouts with WP6, i.e., epigenetic and neuroimaging markers to allow translation between animal and human findings. Overall, results from WPs3-5 and mechanistic information from WP6 will be integrated and validated in humans in WP7. WP7’s multilevel and multimodal integration is not limited to the biological domains but also includes environmental and behavioural factors. Ultimately, WP7 will develop an individualised risk or resilience prediction model of mental health problems using multimodal normative prediction approaches informed by causal pathways, allowing for prediction systems, which also provide etiological insights. Lastly, to address objectives 5 and 6, WP8 will utilise methods from the social sciences to map social and ethical consequences of risk prediction models as a first step to prepare clinical practice for its future implementation.

Fig. 2

FAMILY’s work package framework

Population and familial high-risk cohorts

FAMILY leverages existing longitudinal general population cohorts and familial high-risk offspring cohorts (with at least one parent with a confirmed diagnosis in the mood-psychosis spectrum) spanning childhood to adulthood with available data from children as well as their mother and/or father (i.e., triads and dyads). Both these general population and familial high-risk offspring cohorts include longitudinal assessments of clinical, behavioural, environmental, genetic, and neuroimaging information (Table 2). These already rich data resources will be expanded by (i) generating circulating microRNA (miRNA) profiles at birth from cord blood plasma in 1600 children from Generation R as an additional type of epigenetic process on top of more commonly investigated DNAm; and (ii) collecting parental brain imaging and blood or saliva samples (for those where samples are missing or of poor quality) in familial high-risk cohorts to obtain triad or dyad MRI and genetic datasets of child, mother and/or father (Table 2).

Table 2 Overview of cohorts available to FAMILYWork packagesWP3: Genetically informed designs to disentangle routes of transmission

In WP3, using DNA-variants from triads (father, mother, child) we aim to provide insight into genetic and environmental routes of risk transmission from parents to offspring. In an intergenerational context (Fig. 3), a risk factor, e.g., parental psychotic symptoms, can be associated with offspring psychotic symptoms (i) via the transmission of parental genes (entirely genetic transmission; bottom arrow); and (ii) via an environmentally transmitted effect (top arrow), where parental psychotic symptoms play a role in the emergence of offspring's symptoms. This relates to the concept of ‘genetic nurture’ [30], i.e., the environmentally mediated effect of parental genetics, reflecting the fact that the offspring's nurturing environment is partially shaped by their parents' genetic risk. For example, a risk variant increasing symptoms of psychosis in the parent can impact offspring symptoms via disrupting parenting abilities even when the child does not inherit this variant.

Fig. 3

The virtual parent design: genetic and environmental transmission in an intergenerational context. PS, psychotic symptoms; U, unobserved confounders

Genetic variants measured throughout the genome can be summarised in a polygenic score, i.e., an individual-level score that captures genetic contribution for a given trait. The virtual parent design [30,31,32,33,34] splits polygenic scores corresponding to a particular disorder into transmitted and non-transmitted polygenic scores (Fig. 3). The transmitted scores from both parents form the child’s polygenic score. The path from the parental polygenic score to offspring psychotic symptoms via the offspring polygenic scores corresponds to genetic transmission. Conversely, the path from the non-transmitted polygenic score to the offspring outcome via parental behaviour corresponds to genetic nurture. This genetic nurture path must be environmentally mediated, as genetic transmission is already accounted for.

In WP3, we will develop and apply cutting-edge approaches to estimate genetic nurture effects and identify risk and resilience factors underlying intergenerational transmission of risk of mental illness. That is, genetic nurture and genetic transmission effects will be estimated from polygenic scores from theoretically-based candidate parental risk and resilience factors for offspring psychotic and mood symptoms, including polygenic scores for depression, bipolar disorder and schizophrenia, other psychiatric disorders, and neurodevelopmental conditions, as well as parental education, lifestyle, and physical health. Polygenic scores will also be constructed based on the main independent components from thousands of genome-wide association studies (GWAS) of brain-related traits from WP7. Methodological extensions, such as Mendelian Randomisation (MR), can test whether such factors have a causal effect on child outcomes. MR exploits non-transmitted parental variants related to a risk factor as instruments to assess the causal effect of this risk factor on the outcome. In intergenerational MR, the non-transmitted variants associated to parental risk can be used to estimate the effect of parental risk on offspring outcome (path βxy in Fig. 3) independently of unobserved confounders of this association (U in Fig. 3) [35]. Risk or resilience factors for which a genetic nurture effect is detected will be followed up by MR analyses to test causal relationships directly. Findings in large population cohorts will be tested in familial high-risk offspring cohorts.

WP4: Epigenetic routes of transmission

In WP4, we aim to characterise the relevance of epigenetic processes for the transmission of psychiatric risk. Epigenetic processes have been hypothesised to underlie transmission of risk of mental illness based on observations that (i) epigenetic patterns are influenced by both genetic and environmental factors, starting in utero [36]; (ii) they play an essential role in normative development, including brain maturation and function [37]; and (iii) they are associated with numerous psychiatric disorders, including psychosis and mood disorders [38, 39]. DNAm has by far been the most widely examined epigenetic process, showing huge promise as a biological marker for disorder prediction, early detection, and risk stratification [40,41,42,43]. However, evidence for a role of DNAm in intergenerational transmission of mental illness remains scarce. In contrast to DNAm, circulating miRNAs have received less attention in humans but are increasingly implicated in intergenerational transmission based on animal studies. For example, rodent and C. elegans models have identified miRNAs mediating the effect of parental preconceptional environmental exposures (e.g., starvation, chemical, and stress-related exposures) on offspring health outcomes [44, 45]. Currently, no large-scale data of circulating miRNAs exists in early life, which can be linked to parental exposures and offspring outcomes [46]. Importantly, we are unaware of any largescale study featuring paired samples of both DNAm and miRNA profiles, limiting insights into which processes contribute most to intergenerational transmission.

In WP4, we will apply innovative approaches, including the genetic triad design (from WP3) and machine learning methods. Analyses will be based on data from Generation R [47], ALSPAC [48], and MoBa [49] (Table 2) and will primarily focus on epigenetic patterns at birth, linking parental mental illness to offspring mental health outcomes, pre-symptom manifestation. Relevant markers will then be tested for stability vs change across development in studies with long follow-up using the available repeated epigenetic assessments. The generated results will be replicated using data from the Pregnancy and Childhood Epigenetics (PACE) consortium [50] and tested in the RegionH-VIA familial high-risk offspring cohort (Table 2).

WP5: Neuroimaging routes of transmission

Within WP5, we aim to investigate the role of brain structure and function in the transmission of risk of mental illness from parents to offspring. Brain structural and functional metrics and their developmental trajectories have been hypothesised to underlie the development of mental illness, based on observations that (i) brain metrics are influenced by genetic and familial environmental factors [51]; (ii) they play an essential role in typical and atypical development during adolescence [52]; (iii) brain alterations are associated with numerous psychiatric disorders and may or may not overlap between disorders [53]; and (iv) family members of patients with severe mental illness show similar brain deviations, albeit with smaller effect sizes [54]. Together with WP7, a multimodal dimension reduction approach to the study of psychiatric disorders will be applied to thousands of variables across different neuroimaging modalities, i.e., T1-weighted imaging, diffusion tensor imaging, and resting-state functional MRI. The predictive power of integrated multimodal components, in addition to ‘classical’ unimodal metrics (such as brain volume or functional connectivity strength), can be tested by predicting offspring's mental health problems later in life, taking measures of parental mental health into account, both in the general population and in families at high risk for mental illness. Moreover, multimodal components also aid biological interpretation across the various modalities. This will facilitate mechanistic understanding about which brain regions or structures are more likely to ‘work together’. In addition, we aim to, for the first time, examine the degree to which parental mental health problems during the offspring’s prenatal phase or during early childhood are associated with overlapping or distinct features of brain structure or function between ill parent, partner, and child. Recent research suggests that parent–child relationships influence children's brain development [55], while parental brain networks that are associated with bonding-related behaviour adapt when becoming a parent [56]. While suggestive, previous neuroimaging studies did not jointly examine the neural pathways of intergenerational transmission of mental health problems in parents and children. Strength of this trio modelling approach to the intergenerational transmission of brain developmental variations is that familial and genetic confounding can be controlled for.

Analyses will focus on brain metrics from childhood into young adulthood, linking parental mental illness to offspring mental health outcomes in the age range where symptoms may occur. Relevant predictors will then be tested for stability vs change across development using the available repeated MRI assessments in population cohorts and in the familial high-risk offspring cohorts (Table 2).

WP6: Animal modelling to increase mechanistic understanding of risk transmission

In WP6, we aim to establish causal mechanisms of intergenerational transmission of risk by exploiting two different mouse models, thereby going beyond what is possible in humans. First, environmentally induced molecular and epigenetic changes in the germline will be investigated as possible route of transmission of disorder-like traits from mother to offspring. Inbred mouse models of risk for maladaptive behavioural responses and cognitive functions can determine to what extent molecular/epigenetic factors, maternal behaviours, and nurturing contribute to transmission of disorder traits to the offspring independently or synergistically. We will use a model of early postnatal life adversity known to cause severe behavioural and cognitive impairment in adulthood, and strategies to distinguish the effects of maternal behaviour from molecular/epigenetic factors in reproductive cells [57]. Quantitative and qualitative measures of maternal care during offspring postnatal development will be obtained from ‘mentally-ill’ females and molecular/epigenetic analyses will be conducted in oocytes and female reproductive tissue. To distinguish between prenatal and maternal factors after birth and assess the causal relationship between symptoms in mothers and symptoms in the offspring, embryo transfer will be conducted in a way to grow offspring from ‘mentally-ill’ females in a normal intrauterine environment and vice versa. Cross-fostering will also be utilised to assess the contribution of maternal behaviours. For translational perspectives, the effects of intervention by environmental enrichment will be examined to determine if disorder-like traits can be attenuated or corrected in mothers and if their transmission to the offspring can be prevented, as previously demonstrated for paternal trauma [58].

Second, we will dissect the behavioural and neuroanatomical signatures of genetic and maternal environment risk underlying the intergenerational transmission of psychosis-related behaviour using the 22q11 microdeletion (22q11.2DS) mouse model [59, 60]. Human adults with 22q11.2DS exhibit a range of behavioural, cognitive, and neuroanatomical alterations that put them at increased risk for psychosis. This mouse model will enable the distinction between genetic transmission and genetic nurture (see also Fig. 3). A 2 × 2 breeding design will be implemented with adult female and male mice carrying 22q11.2DS. The four resulting experimental groups are (i) wild type (WT) offspring raised by WT mothers (healthy control); (ii) WT offspring raised by 22q11.2DS mothers (maternal environmental risk, i.e., genetic nurture); (iii) 22q11.2DS offspring raised by WT mothers (genetic transmission); and (iv) 22q11.2DS offspring raised by 22q11.2DS mothers (genetic transmission and genetic nurture). This model allows investigation of quantitative and qualitative measures of genetic transmission and genetically driven maternal environmental risk by evaluating psychosis-related behaviour and cognition during offspring postnatal development. Again, cross-fostering will be utilised to assess causality.

Further integration of findings between the two mouse models will be implemented in parallel experiments to assess maternal behaviour during offspring postnatal development using maternal care measures.

WP7: Multilevel, multimodal integration, causal pathway models, and normative risk prediction

WP7 wraps around the biological domain-specific WPs to integrate the different types of data and information to achieve two fundamental goals: (i) to facilitate interpretation in terms of potential mechanistic insights into intergenerational transmission of risk; and (ii) to improve prediction of mental health outcomes using all the information and individual variation, in ways that are transferable across cohorts and individuals. Recent advances in independent component analysis (ICA) have been specifically addressing the integration of high dimensional data: genomic ICA [61, 62] and linked ICA [63, 64]. Genomic ICA transforms genome-wide associations of thousands of (brain) traits into a smaller set of genomic components that essentially 'group' SNPs according to the similarity of their effects on different brain traits. Genomic ICA achieves efficient data reduction in a way that aids interpretation in terms of biological mechanisms that drive gene-brain associations [61]. The resultant components increase reproducibility of the GWAS signal [62], and can be applied to any new genotyped cohort or individual to calculate polygenic scores per component, creating a new set of variables of individual's loadings on each multivariate component (used in WP3). Linked ICA is applied for the meaningful data reduction and integration of multimodal voxel-wise neuroimaging data (used in WP5), but will in WP7 be extended to include a range of other biological, environmental, and cognitive-behavioural variables (i.e., ‘SuperBigFLICA’) for supervised data reduction [64]. Integrating association patterns across biological levels and modalities from human cohort data through genomic and linked ICA can reveal putative mechanisms: metrics and voxels loading highly on the same component are statistically dependent, and reflect joint biological processes [65].

In stage 1, WP7 will use a fully data-driven way to generate different types of multimodal components with putative predictive value from UK Biobank and ABCD data using both genetic and imaging data. These will be used as new, more interpretable, and more efficient machine-learning features in subsequent mechanistic as well as predictive analyses. To this end, these new feature sets will be passed on to WP3 (genetics) and WP5 (neuroimaging), to test their clinical relevance and potential predictive values in independent population and familial high-risk cohorts. In stage 2, clinically relevant markers from the specialised data domains (WPs3-5) and from the experimental testing of the hypothesised mechanisms in the animal (WP6) will be used to create, test, and generalise predictive models in and between the general population and offspring at high familial risk (Fig. 4); and to test putative mechanisms in dedicated causal pathway models. For example, the temporal ordering in the data, i.e., repeated assessments within individuals, allows for testing cross-lagged models that postulate pathways linking early exposures to later life health outcomes or examining how traits develop with increasing age within-person and assess time-dependent interrelationships. These analyses will integrate several biological levels (e.g., genetics, epigenetics, imaging, psychosocial) to build extended chains of risk transmission (mediation). Factors earlier identified as putative resilience factors will be entered as moderators, to test their risk-dampening effects.

Fig. 4

The work package 7 framework

Further, the results from WPs3-5 will be used by WP7 to develop normative prediction models where symptoms in offspring will be modelled as a function of predictive multimodal risk features along developmental trajectories and individual deviations from predicted risk. This allows for predicting the risk of symptoms in new (independent) individuals [66, 67]. The method builds age- and sex-specific ‘normative’ (i.e., average range reference) trajectories from relevant predictors, notably FAMILY's multilevel, multimodal, and genomic components from the ICAs. Given normative models, new out-of-sample individuals can then be characterised in terms of their deviation from those reference trajectories, while controlling for specific demographics. This may constitute either risk or resilience if symptoms are respectively higher or lower than predicted symptoms. Key advantages of FAMILY's novel normative modelling approach are that (i) the reference models can be flexibly adapted to the characteristics of different target populations by including different additional reference samples [67]; and (ii) it is sensitive to heterogeneity in different clinical populations because the deviation scores do not assume that all patients (or target group members) differ from the reference in a consistent way [68]. The deviation scores will be entered as predictors for prospective, future risk- and resilience estimation, in addition to psychological, clinical, and environmental factors. FAMILY will further develop the normative modelling approach towards more individualised realistic applications by implementing the family context.

WP8: Ethical and social consequence of individualised risk prediction

Risk prediction can give rise to ethical and social issues, e.g. stigma, risk communication, and risk perception [69, 70]. Within a family context, additional issues arise, such as reproductive choices or the right of offspring not to know. Moreover, questions arise with respect to the ethical and professional duty of mental healthcare professionals to recognise and address risk and/or resilience indicators in families of their patients [71]. Although legal frameworks in different countries vary, ethical frameworks are likely to be consistent in requiring special safeguards for both offspring exposed to parental mental illness and parents suffering from mental illness. In WP8, we aim to systematically investigate how ethical and legal normative frameworks should be amended in the context of research and clinical settings. Specifically, cultural, age, and sex/gender aspects in relation to ethical and social consequences of risk prediction (e.g., stigma, shame, guilt) from neurobiological, psychological, environmental, and clinical information will be considered.

We will analyse ethical and normative issues in the context of intergenerational transmission of risk of mental illness and the use of prediction models. That is, we will reflect on how ‘structural discrimination’ exacerbates individual patients’ health problems [72]. Furthermore, empirical data will be gathered through qualitative (semi-structured interviews) and quantitative (survey) research methods, according to standards of practice for empirical bioethics research [73] across several European member states. These empirical data gathered will be integrated with ethical arguments to develop ethical guidelines for professionals to assist in the clinical use of prediction models, as well as empowerment of people with mental disorders and their families. Knowledge gained in WPs3-7 will be shared with WP8 and serve as input for the interviews and surveys.

WP2: Data management and infrastructure for data sharing

Given the large number of data, WP2 is dedicated to guarantee data management in line with the Findable, Accessible, Interoperable and Reusable (FAIR)-principles and General Data Protection Regulation (GDPR)-compliant storage of and access to data sets. Standard Operating Procedures (SOPs) for data harmonisation (following the FAIR-principles), data-merging across, and data access among partner sites for all data types will be centrally developed within FAMILY. Harmonisation of datasets will leverage on existing efforts and plans already in place with large, EU-funded consortia using similar data types and structures (e.g. LifeCycles [74], Early Cause [75]) which jumpstarts the harmonisation process. Requests for data access will be supervised by a Data Access Committee based on the data access policy, which will be developed. The Data Access Committee will conduct scientific evaluation of proposals requesting data.

In FAMILY, all processing of data (when local regulations allow) will be implemented on a dedicated research infrastructure, the Digital Research Environment (DRE: https://www.andrea-cloud.eu/) following GDPR-standards. The DRE provides a secure, flexible, scalable cloud-based platform where researchers have access to and can work (together) with the data, methods, and models available in FAMILY. The DRE operates on the Microsoft Azure platform (which respects intellectual property rights), and the hardware is located within the EU. The architecture of the DRE allows researchers to use a solution within the boundaries of data management rules and regulations as will be put in place by WP2. To ensure the longevity of the FAMILY infrastructure, FAMILY will develop a long-term data (re)use model. Importantly, this strategy also allows for new groups to incorporate their data into the DRE, expanding the potential of this resource.