Binary and analog variation of synapses between cortical pyramidal neurons

  1. Sven Dorkenwald  Is a corresponding author
  2. Nicholas L Turner
  3. Thomas Macrina
  4. Kisuk Lee
  5. Ran Lu
  6. Jingpeng Wu
  7. Agnes L Bodor
  8. Adam A Bleckert
  9. Derrick Brittain
  10. Nico Kemnitz
  11. William M Silversmith
  12. Dodam Ih
  13. Jonathan Zung
  14. Aleksandar Zlateski
  15. Ignacio Tartavull
  16. Szi-Chieh Yu
  17. Sergiy Popovych
  18. William Wong
  19. Manuel Castro
  20. Chris S Jordan
  21. Alyssa M Wilson
  22. Emmanouil Froudarakis
  23. JoAnn Buchanan
  24. Marc M Takeno
  25. Russel Torres
  26. Gayathri Mahalingam
  27. Forrest Collman
  28. Casey M Schneider-Mizell
  29. Daniel J Bumbarger
  30. Yang Li
  31. Lynne Becker
  32. Shelby Suckow
  33. Jacob Reimer
  34. Andreas S Tolias
  35. Nuno Macarico da Costa
  36. R Clay Reid
  37. H Sebastian Seung  Is a corresponding author
  1. Princeton Neuroscience Institute, Princeton University, United States
  2. Computer Science Department, Princeton University, United States
  3. Brain & Cognitive Sciences Department, Massachusetts Institute of Technology, United States
  4. Allen Institute for Brain Science, United States
  5. Department of Neuroscience, Baylor College of Medicine, United States
  6. Center for Neuroscience and Artificial Intelligence, Baylor College of Medicine, United States
  7. Department of Electrical and Computer Engineering, Rice University, United States
4 figures, 3 tables and 1 additional file

Figures

Figure 1 with 3 supplements
Reconstructing cortical circuits in spite of serial section electron microscopy (ssEM) image defects.

(a) Ideally, imaging serial sections followed by computational alignment would create an image stack that reflects the original state of the tissue (left). In practice, image stacks end up with …

Figure 1—figure supplement 1
Reconstruction of connections between layer 2/3 (L2/3) pyramidal cells.

250×140×90 µm3 3D image stack from L2/3 of mouse primary visual cortex.

Figure 1—figure supplement 2
Examples of reconstructed neurites near image defects.

(a) Illustration of a possible pair of neurites that pass through both missing section (cyan) and misalignment (orange) defects. (b) Illustration of a naive segmentation of the pair in (a). (c) Same …

Figure 1—figure supplement 3
Quantitative evidence for the effectiveness of training data augmentation.

Robustness of three boundary detectors trained with no data augmentation (‘baseline’, blue), simulated missing section (‘missing section’, red), and simulated misalignment (‘misalignment’, yellow) …

Figure 2 with 2 supplements
Wiring diagram for cortical neurons including multisynaptic connections.

(a) Wiring diagram of 362 proofread layer 2/3 (L2/3) pyramidal cells (PyCs) as a directed graph. Two orthogonal views with nodes at 3D locations of cell bodies. Single (gray), dual (blue), and …

Figure 2—figure supplement 1
Renderings of all synapses from multisynaptic connections between layer 2/3 (L2/3) pyramidal cells.

Dendritic spines (yellow) and synaptic clefts (red) are rendered in 3D. Most are dual connections (160), but there are also triples (24), quadruples (3), and quintuples (2).

Figure 2—figure supplement 2
Examples of synapses between layer 2/3 (L2/3) pyramidal cells.

Scale bar: 500nm. In each pair of images, the left shows one section through a synapse, and the right adds the automatically detected cleft as an overlay. Note here that the clefts are associated …

Figure 3 with 7 supplements
Modeling spine head volume with a mixture of two log-normal distributions.

(a) Dendritic spine heads (yellow) and clefts (red) of dual connections between layer 2/3 pyramidal cells (L2/3) PyCs. The associated electron microscopy (EM) cutout shows a 2D slice through the …

Figure 3—figure supplement 1
Linear spine head volume distributions.

(a) Spine volumes belonging to dual connections between layer 2/3 pyramidal cells (L2/3 PyCs). (b) Dual connections between L2/3 PyCs, each summarized by the geometric mean of two spine volumes.

Figure 3—figure supplement 2
Arithmetic means.

(a) Dual connections between layer 2/3 pyramidal cells (L2/3 PyCs), each summarized by the arithmetic mean of two spine volumes, modeled by a mixture (red, solid) of two normal distributions (red, …

Figure 3—figure supplement 3
Fits versus raw data histograms.

Plots are analogous to Figure 3 and Figure 3—figure supplement 2. (a) Histogram of same spine volumes, logarithmic scale. A mixture (red, solid) of two log-normal distributions (red, dashed) is …

Figure 3—figure supplement 4
Modeling spine volume with a bimodal versus unimodal mixture of two normal distributions.

(a) Spine volumes belonging to dual connections between layer 2/3 (L2/3) pyramidal cells. A bimodal mixture (red, solid) of two normal distributions (red, dashed) is a better fit than a unimodal …

Figure 3—figure supplement 5
Modeling cleft size with a mixture of two normal distributions.

(a) Spine volume versus cleft size for all layer 2/3 pyramidal cell (L2/3 PyC)-L2/3 PyC synapses. (b) Histogram of same spine volumes, logarithmic scale. A mixture (red, solid) of two normal …

Figure 3—figure supplement 6
Synapse size by connection type.

(a) Bottom: Spine head volume distribution for single connections (gray) along with synapses for all connections (black). Top: parameter estimates for the component means of single connections, as …

Figure 3—figure supplement 7
Relation of dendritic spine volume to spine apparatus.

(a) Examples of spine apparatus (SA) in electron microscopy (EM) images. Scale bar: 300 nm. (b) Spine volume distributions of synapses with no endoplasmic reticulum (ER) (blue), smooth ER (yellow), …

Figure 4 with 6 supplements
Latent state correlations between spines at dual connections.

(a) Scatter plot of spine volumes (black, lexicographic ordering) for dual connections. Data points are mirrored across the diagonal (gray). The joint distribution is fit by a mixture model (orange) …

Figure 4—figure supplement 1
Fits versus raw distributions.

Plots are analogous to Figure 4 and Figure 4—figure supplement 4. (a) Synapse pairs from all dual connections. (b) Dual connections of synapse pairs less than the median distance apart. (c) Dual …

Figure 4—figure supplement 2
Latent state correlations between clefts at dual connections.

(a) The latent states of synapses in a dual connection are positively correlated with each other. The latent states are more likely to be the same (SS or LL) rather than different (SL or LS), as …

Figure 4—figure supplement 3
Residuals spine head volume after subtracting binary components.

We assigned the synapse pairs from the 160 dual synaptic connections to their most likely state (SS, SL, LS, LL) and subtracted the mean of the binary components. (a) shows the residual components. …

Figure 4—figure supplement 4
Synapses in a dual connection: near versus far pairs.

Spine volumes (left) and cleft sizes (right). (a) Dual connections of synapse pairs less than 46.5 μm apart (phi = 0.534). (b) Dual connections of synapse pairs more than 46.5 μm apart (phi = …

Figure 4—figure supplement 5
Dual connection correlations are not a result of axon or dendrite biases.

Synapses are shuffled between dual connections to measure the correlations between synapses that have the same axon (red) or the same dendrite (blue) against a fully random baseline (black). …

Figure 4—figure supplement 6
Removing constraints on the synaptic population eliminates bimodality and reduces correlations.

Spine volumes (top) and cleft sizes (bottom). (a) Distribution of synapse sizes in dual connections received by layer 2/3 (L2/3) pyramidal cells, including those from orphan axons (566 synapses). (b)…

Tables

Table 1
Overview of number of data points obtained in this study.
Number of L2/3 PyCs in dataset417
Number of L2/3 PyCs selected for proofreading362
Number of proofread L2/3 PyCs connecting to any other L2/3 PyCs334
Number of inhibitory cells in dataset34
Number of synapses (automated) in the dataset3,239,275
Number of outgoing synapses (automated) in the dataset from proofread L2/3 PyCs10,788
Number of synapses between L2/3 PyCs1960
Number of connections between L2/3 PyCs1735
Number of connections between L2/3 PyCs with one synapse1546
Number of connections between L2/3 PyCs with two synapses160
Number of connections between L2/3 PyCs with three synapses24
Number of connections between L2/3 PyCs with four synapses3
Number of connections between L2/3 PyCs with five synapses2
Table 2
Overview of results from log-normal mixture fits for different synapse subpopulations.
Subset of L2/3 L2/3 PyC synapsesSLN
Mean(log10 µm3)Std(log10 µm3)WeightMean (log10 µm3)Std(log10 µm3)Weight
All synapses–1.420.240.77–0.770.220.231960
Single synapses–1.410.240.81–0.760.210.191546
Dual synapses–1.440.230.64–0.770.210.36320
Triple synapses–1.490.170.36–0.860.300.6472
All synapses with weights refitted to single synapses(–1.42)(0.24)0.80(–0.77)(0.248)0.201960 and 1546
All synapses with weights refitted to dual synapses(–1.42)(0.24)0.66(–0.77)(0.248)0.341960 and 320
All synapses with weights refitted to triple synapses(–1.42)(0.24)0.52(–0.77)(0.248)0.481960 and 72
Geometric mean of dual synapses–1.440.160.58–0.870.180.42160
Arithmetic mean of dual synapses–1.430.160.53–0.850.180.47160
Table 3
Overview of results from hidden Markov model (HMM) log-normal component fits for different dual synaptic connection subpopulations.
Subset of L2/3 L2/3 PyC dual synaptic connectionsSLWeightsPearson’s phiN
Mean(log10 µm3)Std(log10 µm3)Mean (log10 µm3)Std(log10 µm3)SSSL+LSLL
All connections–1.4700.216–0.8330.2440.4900.1770.3330.637160
Dist <median dist–1.5060.212–0.8610.2430.4270.2320.3420.53480
Dist >median dist–1.4490.207–0.8180.2510.5290.1230.3480.74580

Additional files

Download links