Post

My most recent and biggest work so far is finally out as a preprint! 🧠🧵 *Linking human brain functional connectivity to underlying neurotransmission* doi.org/10.64898/202... 1/n TL;DR below ↓ #neuroskyence #neuroimaging #MedSky #PsySciSky

TL;DR: We develop a framework linking fMRI & MEG connectivity to underlying neurotransmitter distributions. It is robust, reproducible, applicable at the individual level, sensitive to pharmacological manipulation & clinical alterations. A step toward biologically informed connectome analysis. 2/n

Resting-state (rs)fMRI connectivity has been measured for 30 years. But its neurobiological basis remains poorly understood - BOLD is indirect, unspecific, and shaped by neurovascular coupling. Which neurotransmitter systems drive the networks we observe remains largely an open question. 3/n

We introduce the NEOFC framework: if a receptor contributes to regional signals, regions with higher density should show more synchronization. We quantify this by masking an individual's connectome by receptor/transporter density percentiles, producing a curve and a subject-level "AUC" score. 4/n

With this approach, we evaluate 25 PET atlases for their associations with individual connectome architecture. We test for statistical significance with spatial nulls of the PET maps. But first, we run a positive control confirming that we can trace underlying canonical resting-state networks. 5/n

Next, we find that, across 6 independent cohorts, rsfMRI connectivity aligns with neurobiology. Importantly, the strongest effect by far emerges for the noradrenaline transporter (NET), significant in every individual subject! AUC scores showed good-to-excellent 1-day-test-retest reliability! 6/n

During our exploration, we realized: Our approach is intrinsically sensitive to positive associations (high connectivity in regions with high PET density: "AUC+"), but to test for negative associations, we can invert the PET maps. A range of transmitter systems showed strong "AUC–" effects! 7/n

Before going into interpretation, we look at reproducibility and external validation. Do the AUC profiles across neurotransmitter systems replicate in independent samples? And do we find the same associations in a different imaging modality? 8/n

Indeed, they do! AUC profiles correlate 0.8 – 0.9 between rsfMRI datasets, ICC approaches 0.9. MEG mirrors MRI results, but with a twist: We find high correspondence in the beta band (scatter plots). But especially gamma bands show pronounced AUC+ effects not present in MRI. 9/n

Interpretation? fMRI and MEG may reflect the same signal: Assume regions interact at high frequencies such as found in MEG gamma (AUC+). The same neural sources in fMRI are observed through heavy filtering, which may reduce interregional info and leave a local-appearing BOLD pattern. ... 10/n

... Such local patterns may then show as negative associations (fMRI AUC–). Only ultra-slow fluctuations, producing networks with high spatial coherence – such as those supported by long-range projecting neurons – may actually "survive" in fMRI (→ NET AUC+). 11/n

Back to a core finding: The NET effect is driven by a sensorimotor-insular network. In two independent samples, NET AUC+ is inversely associated with autonomic arousal – heart rate variability and pupil unrest – indicating a physiologically relevant signal. Correlation: 0.6 for pupil unrest! 12/n

Application: Pharmaco-fMRI data for 3/4 substances show AUC score modulation largely in line with expected receptor targets (risperidone, ketamine, midazolam). For ketamine and midazolam administered intravenously during scans, NEOFC scores change dynamically during infusion (lower right). 13/n

Limitations: Correlative design, dependence on PET map quality, intercorrelated neurotransmitter maps, and small pharmacological cohorts. Effects nonetheless replicate across 6 cohorts, two imaging modalities, multiple parcellations, and are present at the individual subject level! 15/n

Here, we apply our approach to resting-state fMRI and MEG data on one side, and PET maps on the other. A strength of the approach is its flexibility. I'm excited to see it extended to other biological priors, EEG (think about clinical applicability), and task fMRI (e.g., a reward paradigm)! 16/n

Finally, some resources: Code: github.com/LeonDLotter/... Method as a Python tool: github.com/LeonDLotter/... (Full working example coming soon) Again, the preprint: doi.org/10.64898/202... In the future, we plan an integration into NiSpace: github.com/LeonDLotter/NiSpace 17/n