Towards a more informative representation of the fetal-neonatal brain connectome using variational autoencoder

Abstract

Recent advances in functional magnetic resonance imaging (fMRI) have helped elucidate previously inaccessible trajectories of early-life prenatal and neonatal brain development. To date, the interpretation of fetal-neonatal fMRI data has relied on linear analytic models, akin to adult neuroimaging data. However, unlike the adult brain, the fetal and newborn brain develops extraordinarily rapidly, far outpacing any other brain development period across the lifespan. Consequently, conventional linear computational models may not adequately capture these accelerated and complex neurodevelopmental trajectories during this critical period of brain development along the prenatal-neonatal continuum. To obtain a nuanced understanding of fetal-neonatal brain development, including non-linear growth, for the first time, we developed quantitative, systems-wide representations of brain activity in a large sample (>500) of fetuses, preterm, and full-term neonates using an unsupervised deep generative model called Variational Autoencoder (VAE), a model previously shown to be superior to linear models in representing complex resting state data in healthy adults. Here, we demonstrated that non-linear brain features, i.e., latent variables, derived with the VAE pretrained on rsfMRI of human adults, carried important individual neural signatures, leading to improved representation of prenatal-neonatal brain maturational patterns and more accurate and stable age prediction in the neonate cohort compared to linear models. Using the VAE decoder, we also revealed distinct functional brain networks spanning the sensory and default mode networks. Using the VAE, we are able to reliably capture and quantify complex, non-linear fetal-neonatal functional neural connectivity. This will lay the critical foundation for detailed mapping of healthy and aberrant functional brain signatures that have their origins in fetal life.

Data availability

Data from the Children's National cohort (or DBI dataset) are accessible here: https://doi.org/10.5061/dryad.cvdncjt6n. The Developing Human Connectome Project dataset (dHCP dataset) are here: http://www.developingconnectome.org. The source code, model and documentation for the VAE described in this paper are publicly available at https://github.com/libilab/rsfMRI-VAE.

The following data sets were generated
The following previously published data sets were used

Article and author information

Author details

  1. Jung-Hoon Kim

    Developing Brain Institute, Children's National Hospital, Washington, United States
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3032-8827
  2. Josepheen De Asis-Cruz

    Developing Brain Institute, Children's National Hospital, Washington, United States
    Competing interests
    The authors declare that no competing interests exist.
  3. Dhineshvikram Krishnamurthy

    Developing Brain Institute, Children's National Hospital, Washington, United States
    Competing interests
    The authors declare that no competing interests exist.
  4. Catherine Limperopoulos

    Developing Brain Institute, Children's National Hospital, Washington, United States
    For correspondence
    climpero@childrensnational.org
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-1735-0069

Funding

National Heart, Lung, and Blood Institute (R01 HL116585-01)

  • Catherine Limperopoulos

Canadian Institute of Health Research (MOP-81116)

  • Catherine Limperopoulos

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Ethics

Human subjects: All experiments were conducted under the regulations and guidelines approved by the Institutional Review Board (IRB) of Children's National (Study ID: Pro00013618); written informed consent was obtained from each pregnant woman who participated in the study.

Copyright

© 2023, Kim et al.

This article is distributed under the terms of the Creative Commons Attribution License permitting unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 739
    views
  • 128
    downloads
  • 9
    citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Jung-Hoon Kim
  2. Josepheen De Asis-Cruz
  3. Dhineshvikram Krishnamurthy
  4. Catherine Limperopoulos
(2023)
Towards a more informative representation of the fetal-neonatal brain connectome using variational autoencoder
eLife 12:e80878.
https://doi.org/10.7554/eLife.80878

Share this article

https://doi.org/10.7554/eLife.80878

Further reading

    1. Neuroscience
    Yi-Yun Ho, Qiuwei Yang ... Melissa R Warden
    Research Article

    The infralimbic cortex (IL) is essential for flexible behavioral responses to threatening environmental events. Reactive behaviors such as freezing or flight are adaptive in some contexts, but in others a strategic avoidance behavior may be more advantageous. IL has been implicated in avoidance, but the contribution of distinct IL neural subtypes with differing molecular identities and wiring patterns is poorly understood. Here, we study IL parvalbumin (PV) interneurons in mice as they engage in active avoidance behavior, a behavior in which mice must suppress freezing in order to move to safety. We find that activity in inhibitory PV neurons increases during movement to avoid the shock in this behavioral paradigm, and that PV activity during movement emerges after mice have experienced a single shock, prior to learning avoidance. PV neural activity does not change during movement toward cued rewards or during general locomotion in the open field, behavioral paradigms where freezing does not need to be suppressed to enable movement. Optogenetic suppression of PV neurons increases the duration of freezing and delays the onset of avoidance behavior, but does not affect movement toward rewards or general locomotion. These data provide evidence that IL PV neurons support strategic avoidance behavior by suppressing freezing.

    1. Neuroscience
    David C Williams, Amanda Chu ... Michael A McDannald
    Research Advance Updated

    Recognizing and responding to threat cues is essential to survival. Freezing is a predominant threat behavior in rats. We have recently shown that a threat cue can organize diverse behaviors beyond freezing, including locomotion (Chu et al., 2024). However, that experimental design was complex, required many sessions, and had rats receive many foot shock presentations. Moreover, the findings were descriptive. Here, we gave female and male Long Evans rats cue light illumination paired or unpaired with foot shock (eight total) in a conditioned suppression setting using a range of shock intensities (0.15, 0.25, 0.35, or 0.50 mA). We found that conditioned suppression was only observed at higher foot shock intensities (0.35 mA and 0.50 mA). We constructed comprehensive temporal ethograms by scoring 22,272 frames across 12 behavior categories in 200-ms intervals around cue light illumination. The 0.50 mA and 0.35 mA shock-paired visual cues suppressed reward seeking, rearing, and scaling, as well as light-directed rearing and light-directed scaling. These shock-paired visual cues further elicited locomotion and freezing. Linear discriminant analyses showed that ethogram data could accurately classify rats into paired and unpaired groups. Using complete ethogram data produced superior classification compared to behavior subsets, including an immobility subset featuring freezing. The results demonstrate diverse threat behaviors – in a short and simple procedure – containing sufficient information to distinguish the visual fear conditioning status of individual rats.