Psychological Resilience in Adolescence as a function of Genetic Risk for Major Depressive Disorder and Alzheimer’s Disease

  1. Raluca Petrican
  2. Alex Fornito

  • Curated by eLife

    eLife logo

    eLife assessment

    This work presents a multimodal approach to ascertain links between risk and resilience to depression and Alzheimer's disease in a large pediatric sample. The authors find two latent imaging variables that may be associated with resilience to adverse life events and disease risk, which show some spatial overlap with disease relevant gene-expression patterns and neurotransmitter expression. Such findings could be important for understanding mechanisms underlying resilience in neurological disorders, however, the analyses are inadequate for fully supporting the interpretation of the variables involved in these models, or for supporting some of the overall conclusions of the work.

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Major Depressive Disorder (MDD) and Alzheimer’s Disease (AD) are two pathologies linked to prior stress exposure and altered neurodevelopmental trajectories. As a putative antecedent to AD, MDD could be key to understanding the neurobiological changes that precede the clinical onset of AD by decades. To test this hypothesis, we used longitudinal data from the Adolescent Brain and Cognitive Development study (N total = 980, 470 females) and investigated overlapping connectomic, transcriptomic, and chemoarchitectural correlates of adjustment to stressors (i.e., resilience) among adolescents at genetic risk for AD and MDD, respectively. The potential for perinatal adversity to directly and/or indirectly, via accelerated biological ageing, foster resilience (i.e., “inoculation” effects) was also probed. We identified two distinguishable neurodevelopmental profiles predictive of resilience among MDD-vulnerable adolescents. One profile, expressed among the fastest developing youth, overlapped with areas of greater dopamine receptor density and reflected the maturational refinement of the inhibitory control architecture. The second profile distinguished resilient MDD-prone youth from psychologically vulnerable adolescents genetically predisposed towards AD. This profile, associated with elevated GABA, relative to glutamate, receptor density, captured the longitudinal refinement and increasing context specificity of incentive-related brain activations. Its transcriptomic signature implied that poorer resilience among AD-prone youth may be associated with greater expression of MDD-relevant genes. Our findings are compatible with the proposed role of MDD as a precursor to AD and underscore the pivotal contribution of incentive processing to this relationship. They further speak to the key neuromodulatory role of DA-gonadal hormone interactions in fostering resilience in adolescence.

Significance Statement

Environmental stressors can substantially alter brain maturation and incur lifelong costs. Using longitudinal data, we characterise two developmental profiles correlated with positive adjustment to environmental challenges (i.e., resilience) among adolescents at genetic risk for two stress-related conditions, Alzheimer’s Disease (AD) and Major Depressive Disorder (MDD), respectively. One dopamine-related profile typified the fastest developing MDD-prone adolescents and reflected the neural maturation of the inhibitory control architecture. The second profile, neurochemically linked to excitation/inhibition balance, indicated the developmental refinement of motivational pathways, distinguishing resilient MDD-prone from psychologically vulnerable AD-prone teens. Its transcriptomic signature supported the posited role of MDD as an antecedent to AD. Our results unveil candidate neurobiological mechanisms supporting lifespan resilience against both psychiatric and neurological conditions linked to stress exposure.

Article activity feed

  1. eLife

    eLife assessment

    This work presents a multimodal approach to ascertain links between risk and resilience to depression and Alzheimer's disease in a large pediatric sample. The authors find two latent imaging variables that may be associated with resilience to adverse life events and disease risk, which show some spatial overlap with disease relevant gene-expression patterns and neurotransmitter expression. Such findings could be important for understanding mechanisms underlying resilience in neurological disorders, however, the analyses are inadequate for fully supporting the interpretation of the variables involved in these models, or for supporting some of the overall conclusions of the work.

  2. eLife

    Reviewer #1 (Public Review):

    Motivated by the premise that Alzheimer's disease (ADD) and major depressive disorder (MDD) have shared underlying environmental and genetic risk factors, Petrican and Fornito combine non-imaging risk factors and executive task-based functional network change indices into latent variables of resilience to AD and MDD. The authors find two latent variables (LVs): LV1 represents change in network membership over time of distributed nodes during task, which is associated with greater genetic MDD risk, less psychopathology, and more advanced puberty, all while adjusting for age and indices of environmental stressors. LV2 represents occipital lobe nodal flexibility across task and time, decreased AD genetic risk, increased MDD genetic risk and less psychopathology, again adjusted for age and environmental stressors. The authors validate the latent network variables by assessing their overlap with genes for which SNPs have been associated with both depression risk and change in gene expression. Finally, the authors create simple path models in order to break down the relationships between genetic risk, latent variables, and what the authors term "resilience", finding distinct path for MDD and (non-APOE) AD genetic risk. All of these analyses are then re-run using a different brain parcellation. LV2 replicates, while a new LV1 emerges with similar non-imaging variables now being correlated with a different set of distributed network nodes.

    The authors conclude from this work that they have identified imaging indices of resilience manifest during adolescent brain development, and that they have found further evidence linking MDD to AD. However, the analyses do not fully support the conclusions. The premise of this work - to examine links between MDD and AD and to try to define indices of resilience during development - is fascinating and will hopefully motivate future work in this direction. However, the impact of this work as currently presented may be limited.

    *STUDY STRENGTHS*

    There are two premises motivating this study that deserve praise for their innovation and creativity. First, in the introduction the authors present several fairly new papers showing shared environmental and risk factors between AD and MDD. This is a very interesting line of study that certainly deserves more attention. Second, the authors are interested in finding aspects of adolescent brain development that may be helpful to understanding resilience to genetic or environmental risk later in life. The AD resilience community is very interested in contributions of early life experiences and development, but there is still very little research in this domain. I hope the authors continue to conduct research in the direction of these pursuits.

    The authors demonstrate great methodological and statistical rigor in some aspects of data preprocessing and analysis. This is particularly salient in null modeling and permutation, graph-based analysis, treatment of motion for functional imaging, using eQTLs to inform disease-relevant genes, statistical considerations in PLS and path modeling, processing of Allen Brain Atlas gene expression data, and validating certain study variables. The methodology of these steps displays great attention to detail and a mastery of certain data types.

    The authors reproduce all analyses using a second parcellation and carefully report the results. This type of painstaking analysis is nonetheless important in the context of network-based graph analysis that is reliant on nodal information.

    *STUDY LIMITATIONS*

    1. The overarching limitation of this study is that the study variables, both independent and dependent, are abstracted to the point where interpretations are challenging. The authors' own interpretations are not sufficiently justified and are often taken at face value rather than supported by analysis. These are further combined into latent variables with weak conceptual foundation, which are then abstracted even further to other analyses with cortical molecular data maps. It is not clear that the conclusions drawn are convincingly supported by this highly abstracted analysis.

    2. The other major limitation of this study is that several PLS models are run but, while appropriate null modeling is used to identify "significant" LVs, none of the LVs are cross-validated. Null modeling can help to protect against overfitting to noise in data, but it does not necessarily provide a good index of generalizability nor reliability. Without cross-validation, I question the reliability of the LVs irrespective of how they are interpreted. This is once again partially driven by the fact that changing the atlas resulted in a different imaging LV.

    3. The study notes that participants were selected based on "having contributed high-quality data on all measures of interest". This is of course meritorious from a methodological perspective, but the authors should be aware that this may create an important selection bias (10.1007/s11682-022-00665-2, 10.1016/j.ynirp.2022.100085, 10.1016/j.neuroimage.2022.119296)

    4. The premise of this paper was interesting, as described in the Strengths section above. However, what was missing was a clear theory or hypothesis as to how resilience to AD and MDD are related, and how the analyses in this study were conducted in order to support that hypothesis. The relevance of the results to AD was not clear; a clear biological model would help put the pieces together.

    5. The selection of relevant features involved in LVs was inconsistent. At several points, the authors use an arbitrary threshold of bootstrap ratio (BSR) > 4, which they equated to a p-value. A p-value doesn't make sense in this context, since bootstrap samples are not independent samples. Instead, features should be selected based on 95% CIs that don't cross 0, which the authors do in some places but not in others.

  3. eLife

    Reviewer #2 (Public Review):

    The authors' manuscript has several strengths. First, the authors consider multiple relevant levels of biology including genomics, transcriptomics, structural and functional neuroimaging, cognitive neuroscience, and psychological/environmental factors. Such an approach is often necessary to deconvolute the complexities of psychiatric phenotypes. The authors have taken careful steps to think about potential confounds (e.g., ancestry for PRS) and to try to define their phenotypes (e.g., psychological resilience and biological aging) as best as they can, given the data they have access to from the ABCD study. The manuscript is well written overall.

    My main concerns relate to core assumptions and techniques that underlie the premise of the study. First, while there is comorbidity between AD and MDD, a causal relationship between the two (in either direction) is not established. Though MDD often predates AD, this is to be expected given MDD's high lifetime prevalence (15-20% of the general population) and typical age of onset before age 65. Because AD typically presents late in life (>65 years of age), MDD will, by definition, usually predate AD. While new onset, late life MDD is often the first presenting symptom of AD/Parkinson's disease and other neurodegenerative conditions, it is also not clear that this is the same disorder as idiopathic MDD.

    To this point, two genetic tools can help us determine the biological relationship between MDD/AD, genetic correlation and Mendelian Randomization. Using the data from the MDD PRS used in this analysis, the Supplementary Table 3 from the Howard et al. 2019 paper (https://doi.org/10.1038/s41593-018-0326-7) reveals a genetic correlation of -0.041 between the two. This indicates essentially no strong relationship between the MDD/AD (perhaps even a slightly inverse relationship). Mendelian Randomization studies in addition to the Howard et al paper (https://doi.org/10.1212/WNL.0000000000010463) find no causal role for MDD towards AD and vice versa. Thus, their comorbidity is likely mediated by additional factors. Additionally, while stress contributes to AD pathophysiology, AD is strongly genetic and, given its late onset, it is unclear how genetic risk for AD would meaningfully impact the psychological resilience of a 9 to 10-year-old.

    My second concern is regarding the statement "adolescents at genetic risk for AD/MDD" when describing the sample. Per Howard et al 2019 out-of-sample prediction testing, the MDD PRS used by the authors explains between 1.5-3.2% of the phenotypic variance in MDD when used on a sample such as ABCD. MDD PRS is in its infancy and cannot reliably be used to identify individuals at high risk of MDD given that even individuals in the top 10th percentile of MDD PRS have an odds ratio for depression of only ~2.4. We would expect 90 or so individuals in this cohort to fall into this group leaving significant concerns about statistical power and the potential for false positive discoveries. While the AD PRS is significantly further along compared to MDD because of AD's simpler genetic architecture, the same concerns apply as, outside of APOE, the AD PRS does not capture the majority of phenotypic variance in AD.

    The authors state that they wish to examine the effects of perinatal adversity directly/indirectly on biological aging and then assess the potential effects of biological aging on resilience. The authors use of pubertal age as a measure of accelerated aging is understandable given the data available, though not ideal. There are well validated measures of biological age such as Horvath's epigenetic clock. While advanced pubertal age is technically a form of accelerated aging, the majority of pubertal age as a phenotype is not likely to be explained by perinatal adversity. Rather, a combination of unmeasured variables including genetic variation, dietary factors, environmental exposures (endocrine disrupting chemicals), and obesity that play a substantial role in determining pubertal age. Childhood stress has been shown to have relatively small effects on pubertal age (d = -0.1) (10.1037/bul0000270).

    Lastly, the authors employ the use of an as of yet unpublished technique to map neurotransmitters density to structural data from neuroimaging studies. While this technique is certainly interesting, its face validity is not clear given that many of the receptor-disease associations reported in the original preprint do not line up with what we know about the biology of these disorders from strong human genetics data or current FDA approved treatments. Moreover, the authors mention "Excitation/Inhibition" imbalance but the technique used appears to only include glutamate data from one receptor type, mGluR5. This may not be an adequate measure of E/I imbalance, despite there being a statistically significant finding.

    Measuring both transcriptional output from GWAS loci and gene expression correlates from MRI data is a noisy and challenging prospect. Indeed, recent research has shown poor correlation between gene expression and neurotransmitter receptor density.(https://doi.org/10.1016/j.neuroimage.2022.119671).

    Thus, fundamental aspects of this manuscript including the use of MDD PRS to identify "at risk" individuals, the unclear link between AD and adolescent psychological resilience, the use of prepubertal age as a measure of biological age, and the limited conclusions that can be drawn from the gene expression and receptor density technique limits confidence in the results as presented.

  4. eLife

    Author Response:

    We appreciate the Reviewers’ feedback. The manuscript was extensively revised and ultimately accepted for publication (Petrican and Fornito, 2023, Developmental Cognitive Neuroscience). The revisions address the Reviewers’ key concerns, including the theoretical basis of the link between MDD and AD, the rationale for studying this link in adolescence, clear references to significant genetic associations between the two, detailed assessment of CCA and PLS model generalisability and reliability, quantification of resilience, residualization of confounders, and corrections for multiple comparisons. We also note that the details concerning the receptor density maps we use in our analysis have now been published (Hansen et al., 2022, Nature Neuroscience; Markello et al., 2022, Nature Methods).

  5. Version published to 10.1101/2022.07.19.22277815v1 on medRxiv