Emergent topological structure in spontaneous brain-organoid activity

Introduction to the show: ident: Genomics Radio. Generated commentary on the latest computational biology and genomics papers.

Ines: We're looking at the paper: "Emergent topological structure in spontaneous brain-organoid activity".

Marcus: Here's the gist of the paper. Emergent topological structure in spontaneous brain-organoid activity finds that persistent homology resolves structured topology in neural recordings at the scale experiments actually deliver Neural activity is widely held to organize on low-dimensional structure embedded in a high-dimensional state space Persistent homology reads such structure directly from the pattern of pairwise correlations, without assuming in advance which variables are relevant Applied to neural data, persistent homology has revealed the circular and toroidal geometry of spatial coding Brain organoids, self-organizing three-dimensional cultures of human or mouse cortical tissue

12, 2607.16517: are an increasingly important model system Persistent homology characterizes the global organization of that same correlation graph, the loops and voids that local measures of this kind miss Data Acquisition and Network Construction The analysis uses spontaneous extracellular recordings from human cerebral organoids (Lancaster protocol; datasets labeled On) and mouse cortical organoids (Paşca protocol; datasets labeled MOn) Activity was recorded on high-density CMOS MEAs at 20 kHz for three minutes per organoid, resolving spike times to sub-millisecond precision The resulting datasets span N = 26 to N = 234 units (Table 1) Correlations are computed within short lag windows of 0, 10, and 20 ms (Sec.

Ines: 3) The activity is strongly bursting, which the null model of Sec.

Marcus: 3 controls for Detection of Loop Structure Persistent homology finds loop topology over and above what a rate- and population-preserving null produces, at the unit counts experiments actually deliver The integrated first Betti number

4: , Rβ1(ρ) dρ, summarizes how much loop structure a network carries across the whole filtration Loop structure in the organoid data exceeds the surrogate null in 14 of the 18 datasets at p ≤ 0.05 The loops occur at low edge density: β1(ρ) peaks for ρ ≲ 0.15 and has decayed to near zero by ρ ≈ 0.2 The topology is therefore a property of the strong-correlation backbone, not of the dense graph Robustness and Core Structure If the loop structure were diffuse, supported redundantly by many interchangeable units, it would degrade gracefully as units are removed Targeted removal disrupts it far more, but quantifying that disruption requires measuring the bottleneck distance between the original and postremoval H1 persistence diagrams The loops are thus carried by an identifiable, non-redundant core of units rather than spread interchangeably across the population Scaling of Topological Features Richness here means how many homological dimensions a network resolves above the null, that is, how many of H0, H1, H2 (and higher) it populates The count within a dimension tells the same story Connected components (H0) resolve at every size Loops (H1) need enough units to close a cycle and resolve above the null in 14 of 18 datasets, not in strict order of size Enclosed voids (H2) come online later, clearing the null only among the larger networks Emergence of Higher-Order Structure Loops are the first rung of higher-order structure; enclosed voids (H2) are the next, and they require both more units and richer co-firing to form Significant H2 appears only for N ≥ 119 (dotted line) The voids that drive it are few and low-persistence, much shorter-lived than the loops in the same data The method begins to resolve second-order structure where the network is large enough to support it Independence from Recording Geometry The structural layout therefore imposes no topology of its own, and in O5 the loops are functional rather than spatial The loop structure we report is not an artifact of recording geometry With only two embedded datasets this is illustrative rather than definitive, but it indicates that the loop structure we report is not an artifact of recording geometry Persistent homology resolves real, structured loop topology in spontaneous organoid activity at the modest node counts that recordings provide H1 structure is statistically resolvable from roughly one hundred units upward, and β1 trends upward with N across the datasets This shows persistent homology to be a usable instrument for the recordings the field already has This points to a clear next step: applying the same pipeline to optical recordings with single-cell resolution and, ultimately, to three-dimensional recording at high temporal resolution A directed filtration would add the firing order the undirected construction sets aside The results extend the same topological signatures to self-organizing organoid tissue and, for method development, demonstrate that they are recoverable from datasets of order 102 units Acknowledgements We thank Ginestra Bianconi and Michael Freedman for helpful insight into multiple aspects of simplicial complexes and topological data analysis We acknowledge early experimental support and discussions with Raymond Griffard and Tjitse van der Molen, and theoretical support via discussions with Om Biyani and Bismah Rizwan This work was performed in part with support by the Google Academic Research Award program (LDC, KSK, DB) and the U.S.

Ines: National Science Foundation under Grants DGE-2125899 (MB, LDC) and PHY-2515059 (LDC) NM gratefully acknowledges funding from the Noyce Foundation References

1: Singh G, Memoli F, Ishkhanov T, Sapiro G, Carlsson G and Ringach D L 2008 Journal of Vision 8 11 URL https://doi.org/10.1167/8.8.11

2: Gardner R J, Hermansen E, Pachitariu M et al 2022 Nature 602 123–128

3: Acosta F, Sanborn S, Dao Duc K, Madhav M and Miolane N 2023 Quantifying extrinsic curvature in neural manifolds Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops pp 610–619

4: Giusti C, Pastalkova E, Curto C and Itskov V 2015 Proceedings of the National Academy of Sciences 112 13455–13460 URL https://doi.org/10.1073/pnas.1506407112

5: Sizemore A E, Phillips-Cremins J E, Ghrist R and Bassett D S 2019 Network Neuroscience 3 656–673 URL https://doi.org/10.1162/netn a 00073 <ref:2607.

Paper summary: Marcus: So, to summarize the core claim of "Emergent topological structure in spontaneous brain-organoid activity," they are showing that persistent homology reads structured topology directly from neural recordings at the modest node counts that these experiments actually deliver.

Ines: The thesis is pretty strong: they show that this analysis finds loop topology over and above what a rate- and population-preserving null model produces, which is what we actually get in the unit count experiments.

Yuki: So, what does this mean for us in terms of biological understanding? If these loops are real, what are they actually representing in the context of brain function or development?

Ines: The results show that the loops appear at low edge density. Specifically, beta one peaks when the density is less than zero point fifteen and then decays toward zero by zero point two. This tells us that this topology is really a property of the strong-correlation backbone, not just a dense graph.

Marcus: And they also discuss robustness here—if the loop structure were just diffuse, supported redundantly by many interchangeable units, it would degrade gracefully when you remove units. But targeted removal disrupts it far more if you measure that bottleneck distance between the original and postremoval H1 persistence diagrams.

Yuki: That implies these loops are carried by a non-redundant core of units rather than just being spread out across the entire population of cells in the organoid. That’s a significant biological detail.

Ines: And they scale this up to higher dimensions: connected components, loops, and enclosed voids. They find that loops only need enough units to close a cycle and resolve above the null in fourteen of eighteen datasets, not strictly ordered by size.

Conclusion: Marcus: So, looking at the title "Emergent topological structure in spontaneous brain-organoid activity," it really captures that this isn't just finding some abstract mathematical pattern; it’s finding real, structured loop topology in the spontaneous activity of these organoids.

Ines: They did this using persistent homology to resolve real, structured loop topology at the modest node counts that recordings actually provide. The authors are suggesting this is a usable instrument for the recordings we already have available.

Yuki: For someone listening who doesn't do topological data analysis every day, what does this actually change for understanding how these three-dimensional cultures are organizing themselves?

Ines: It means we can use these tools to detect functional geometry in spontaneous activity that might be invisible to standard rate or population measures. The finding suggests that the loop structure is functional rather than just a spatial artifact of how we set up the recording.

Marcus: And they point toward future work, suggesting applying the same pipeline to optical recordings with single-cell resolution, and eventually to three-dimensional recording at high temporal resolution. They also mention that a directed filtration could add information about firing order, which right now is left out of the undirected construction.

Yuki: So what’s the practical implication for population genetics or developmental biology? If we can see these topological signatures in organoids, it might give us new ways to characterize complex functional networks before we even look at the gene expression levels.

Ines: It points toward a clear next step: using persistent homology on different recording modalities and scales to see if these topological signatures hold up across different biological systems.