Abstract
The posterior parietal cortex exhibits choice-selective activity during perceptual decision-making tasks1,2,3,4,5,6,7,8,9,10. However, it is not known how this selective activity arises from the underlying synaptic connectivity. Here we combined virtual-reality behaviour, two-photon calcium imaging, high-throughput electron microscopy and circuit modelling to analyse how synaptic connectivity between neurons in the posterior parietal cortex relates to their selective activity. We found that excitatory pyramidal neurons preferentially target inhibitory interneurons with the same selectivity. In turn, inhibitory interneurons preferentially target pyramidal neurons with opposite selectivity, forming an opponent inhibition motif. This motif was present even between neurons with activity peaks in different task epochs. We developed neural-circuit models of the computations performed by these motifs, and found that opponent inhibition between neural populations with opposite selectivity amplifies selective inputs, thereby improving the encoding of trial-type information. The models also predict that opponent inhibition between neurons with activity peaks in different task epochs contributes to creating choice-specific sequential activity. These results provide evidence for how synaptic connectivity in cortical circuits supports a learned decision-making task.
This is a preview of subscription content, access via your institution
Access options
Access Nature and 54 other Nature Portfolio journals
Get Nature+, our best-value online-access subscription
$29.99 / 30 days
cancel any time
Subscribe to this journal
Receive 51 print issues and online access
$199.00 per year
only $3.90 per issue
Buy this article
- Purchase on Springer Link
- Instant access to full article PDF
Prices may be subject to local taxes which are calculated during checkout
Similar content being viewed by others
Data availability
Directions for accessing the EM dataset, reconstructed neurons and calcium imaging data are available at GitHub (https://github.com/htem/PPC_inhibitoryMotifs).
Code availability
Software and analysis code is available at Zenodo (https://zenodo.org/doi/10.5281/zenodo.10310186)75. Code used to perform all information-theoretic analyses was published previously66 and is available online (https://doi.org/10.1186/1471-2202-10-81). Code for the network model analyses is available at Zenodo (https://doi.org/10.5281/zenodo.10200999)76.
Change history
01 March 2024
In the version of this article initially published, the third software package listed in the Reporting Summary did not match the version listed in the Methods and has now been updated.
References
Assad, J. A. & Maunsell, J. H. R. Neuronal correlates of inferred motion in primate posterior parietal cortex. Nature 373, 518–521 (1995).
Shadlen, M. N. & Newsome, W. T. Motion perception: seeing and deciding. Proc. Natl Acad. Sci. USA 93, 628–633 (1996).
Harvey, C. D., Coen, P. & Tank, D. W. Choice-specific sequences in parietal cortex during a virtual-navigation decision task. Nature 484, 62–68 (2012).
Raposo, D., Kaufman, M. T. & Churchland, A. K. A category-free neural population supports evolving demands during decision-making. Nat. Neurosci. 17, 1784–1792 (2014).
Goard, M. J. et al. Distinct roles of visual, parietal, and frontal motor cortices in memory-guided sensorimotor decisions. eLife 5, 471–477 (2016).
Funamizu, A., Kuhn, B. & Doya, K. Neural substrate of dynamic Bayesian inference in the cerebral cortex. Nat. Neurosci. 19, 1682–1689 (2016).
Hwang, E. J., Dahlen, J. E., Mukundan, M. & Komiyama, T. History-based action selection bias in posterior parietal cortex. Nat. Commun. 8, 1242 (2017).
Song, Y. H. et al. A neural circuit for auditory dominance over visual perception. Neuron 93, 940–954 (2017).
Zhong, L. et al. Causal contributions of parietal cortex to perceptual decision-making during stimulus categorization. Nat. Neurosci. 22, 963–973 (2019).
Najafi, F. et al. Excitatory and inhibitory subnetworks are equally selective during decision-making and emerge simultaneously during learning. Neuron 105, 165–179 (2020).
Shadlen, M. N., Britten, K. H., Newsome, W. T. & Movshon, J. A. A computational analysis of the relationship between neuronal and behavioral responses to visual motion. J. Neurosci. 16, 1486–1510 (1996).
Wang, X. J. Decision making in recurrent neuronal circuits. Neuron 60, 215–234 (2008).
Wang, X. J. Probabilistic decision making by slow reverberation in cortical circuits. Neuron 36, 955–968 (2002).
Wong, K. F. & Wang, X. J. A recurrent network mechanism of time integration in perceptual decisions. J. Neurosci. 26, 1314–1328 (2006).
Roach, J. P., Churchland, A. K. & Engel, T. A. Choice selective inhibition drives stability and competition in decision circuits. Nat. Commun. 141, 147 (2023).
Denk, W. & Horstmann, H. Serial block-face scanning electron microscopy to reconstruct three-dimensional tissue nanostructure. PLoS Biol. 2, e329 (2004).
Xu, C. S. et al. Enhanced FIB-SEM systems for large-volume 3D imaging. eLife 6, e25916 (2017).
Eberle, A. L. et al. High-resolution, high-throughput imaging with a multibeam scanning electron microscope. J. Microsc. 259, 114–120 (2015).
Phelps, J. S. et al. Reconstruction of motor control circuits in adult Drosophila using automated transmission electron microscopy. Cell 184, 759–774 (2021).
Shapson-Coe, A. et al. A connectomic study of a petascale fragment of human cerebral cortex. Preprint at bioRxiv https://doi.org/10.1101/2021.05.29.446289 (2021).
Kasthuri, N. et al. Saturated reconstruction of a volume of neocortex. Cell 162, 648–661 (2015).
Bock, D. D. et al. Network anatomy and in vivo physiology of visual cortical neurons. Nature 471, 177–182 (2011).
Lee, W.-C. A. et al. Anatomy and function of an excitatory network in the visual cortex. Nature 532, 370–374 (2016).
Motta, A. et al. Dense connectomic reconstruction in layer 4 of the somatosensory cortex. Science 366, eaay3134 (2019).
Dorkenwald, S. et al. Binary and analog variation of synapses between cortical pyramidal neurons. eLife 11, e76120 (2022).
Turner, N. L. et al. Reconstruction of neocortex: organelles, compartments, cells, circuits, and activity. Cell 185, 1082–1100 (2022).
MICrONS Consortium et al. Functional connectomics spanning multiple areas of mouse visual cortex. Preprint at bioRxiv https://doi.org/10.1101/2021.07.28.454025 (2021).
Hofer, S. B. et al. Differential connectivity and response dynamics of excitatory and inhibitory neurons in visual cortex. Nat. Neurosci. 14, 1045–1052 (2011).
Runyan, C. A. et al. Response features of parvalbumin-expressing interneurons suggest precise roles for subtypes of inhibition in visual cortex. Neuron 67, 847–857 (2010).
Ma, W. P. et al. Visual representations by cortical somatostatin inhibitory neurons—selective but with weak and delayed responses. J. Neurosci. 30, 14371–14379 (2010).
Driscoll, L. N., Pettit, N. L., Minderer, M., Chettih, S. N. & Harvey, C. D. Dynamic reorganization of neuronal activity patterns in parietal cortex. Cell 170, 986–999 (2017).
Runyan, C. A., Piasini, E., Panzeri, S. & Harvey, C. D. Distinct timescales of population coding across cortex. Nature 548, 92–96 (2017).
Holler, S., Köstinger, G., Martin, K. A. C., Schuhknecht, G. F. P. & Stratford, K. J. Structure and function of a neocortical synapse. Nature 591, 111–116 (2021).
Stepanyants, A., Hof, P. R. & Chklovskii, D. B. Geometry and structural plasticity of synaptic connectivity. Neuron 34, 275–288 (2002).
Holtmaat, A. & Svoboda, K. Experience-dependent structural synaptic plasticity in the mammalian brain. Nat. Rev. Neurosci. 109, 647–658 (2009).
Cannon, S. C., Robinson, D. A. & Shamma, S. A proposed neural network for the integrator of the oculomotor system. Biol. Cybern. 49, 127–136 (1983).
Seung, H. S. in The Handbook of Brain Theory and Neural Networks 2nd edn (ed. Arbib, M. A.) 94–97 (MIT, 2003).
Lim, S. & Goldman, M. S. Balanced cortical microcircuitry for maintaining information in working memory. Nat. Neurosci. 16, 1306–1314 (2013).
Rajan, K., Harvey, C. D. D. & Tank, D. W. W. Recurrent network models of sequence generation and memory. Neuron 90, 128–142 (2016).
Song, H. F., Yang, G. R. & Wang, X. J. Training excitatory-inhibitory recurrent neural networks for cognitive tasks: a simple and flexible framework. PLoS Comput. Biol. 12, e1004792 (2016).
Kim, R. & Sejnowski, T. J. Strong inhibitory signaling underlies stable temporal dynamics and working memory in spiking neural networks. Nat. Neurosci. 24, 129–139 (2021).
Orlandi, J. G., Abdolrahmani, M., Aoki, R., Lyamzin, D. R. & Benucci, A. Distributed context-dependent choice information in mouse posterior cortex. Nat. Commun. 14, 192 (2023).
Koyama, M. et al. A circuit motif in the zebrafish hindbrain for a two alternative behavioral choice to turn left or right. eLife 5, e16808 (2016).
Jovanic, T. et al. Competitive disinhibition mediates behavioral choice and sequences in Drosophila. Cell 167, 858–870 (2016).
Wilson, D. E., Scholl, B. & Fitzpatrick, D. Differential tuning of excitation and inhibition shapes direction selectivity in ferret visual cortex. Nature 560, 97–101 (2018).
Wilson, D. E. et al. GABAergic neurons in ferret visual cortex participate in functionally specific networks. Neuron 93, 1058–1065 (2017).
Keller, A. J. & Martin, K. A. C. Local circuits for contrast normalization and adaptation investigated with two-photon imaging in cat primary visual cortex. J. Neurosci. 35, 10078–10087 (2015).
Cardin, J. A., Palmer, L. A. & Contreras, D. Stimulus feature selectivity in excitatory and inhibitory neurons in primary visual cortex. J. Neurosci. 27, 10333–10344 (2007).
Tseng, S. Y., Chettih, S. N., Arlt, C., Barroso-Luque, R. & Harvey, C. D. Shared and specialized coding across posterior cortical areas for dynamic navigation decisions. Neuron 110, 2484–2502 (2022).
Zagha, E., Ge, X. & McCormick, D. A. Competing neural ensembles in motor cortex gate goal-directed motor output. Neuron 88, 565–577 (2015).
Bartol, T. M. et al. Nanoconnectomic upper bound on the variability of synaptic plasticity. eLife 4, e10778 (2015).
Cossell, L. et al. Functional organization of excitatory synaptic strength in primary visual cortex. Nature 518, 399–403 (2015).
Morcos, A. S. & Harvey, C. D. History-dependent variability in population dynamics during evidence accumulation in cortex. Nat. Neurosci. 19, 1672–1681 (2016).
Krumin, M., Lee, J. J., Harris, K. D. & Carandini, M. Decision and navigation in mouse parietal cortex. eLife 7, e42583 (2018).
Mimica, B., Dunn, B. A., Tombaz, T., Srikanth Bojja, V. P. T. N. & Whitlock, J. R. Efficient cortical coding of 3D posture in freely behaving rats. Science 362, 584–589 (2018).
Minderer, M., Brown, K. D. & Harvey, C. D. The spatial structure of neural encoding in mouse posterior cortex during navigation. Neuron 102, 232–248 (2019).
Hahnloser, R. H. R., Kozhevnikov, A. A. & Fee, M. S. An ultra-sparse code underliesthe generation of neural sequences in a songbird. Nature 419, 65–70 (2002).
Koay, S. A., Charles, A. S., Thiberge, S. Y., Brody, C. D. & Tank, D. W. Sequential and efficient neural-population coding of complex task information. Neuron 110, 328–349 (2022).
Aronov, D. & Tank, D. W. Engagement of neural circuits underlying 2D spatial navigation in a rodent virtual reality system. Neuron 84, 442–456 (2014).
Pachitariu, M. et al. Suite2p: beyond 10,000 neurons with standard two-photon microscopy. Preprint at bioRxiv https://doi.org/10.1101/061507 (2016).
Pnevmatikakis, E. A. et al. A structured matrix factorization framework for large scale calcium imaging data analysis. Preprint at https://doi.org/arxiv.org/abs/1409.2903v1 (2014).
Chan, K. Y. et al. Engineered AAVs for efficient noninvasive gene delivery to the central and peripheral nervous systems. Nat. Neurosci. 20, 1172–1179 (2017).
Dimidschstein, J. et al. A viral strategy for targeting and manipulating interneurons across vertebrate species. Nat. Neurosci. 19, 1743–1749 (2016).
Marshel, J. H. et al. Cortical layer-specific critical dynamics triggering perception. Science 365, eaaw5202 (2019).
Shannon, C. E. A mathematical theory of communication. Bell Syst. Tech. J. 27, 379–423 (1948).
Magri, C., Whittingstall, K., Singh, V., Logothetis, N. K. & Panzeri, S. A toolbox for the fast information analysis of multiple-site LFP, EEG and spike train recordings. BMC Neurosci. 10, 81 (2009).
Panzeri, S. & Treves, A. Analytical estimates of limited sampling biases in different information measures. Network 7, 87–107 (1996).
Bogovic, J. A., Hanslovsky, P., Wong, A. & Saalfeld, S. Robust registration of calcium images by learned contrast synthesis. In Proc. 2016 IEEE 13th International Symposium on Biomedical Imaging 1123–1126 (ISBI, 2016).
Saalfeld, S., Cardona, A., Hartenstein, V. & Tomancak, P. CATMAID: collaborative annotation toolkit for massive amounts of image data. Bioinformatics 25, 1984–1986 (2009).
Schneider-Mizell, C. M. et al. Quantitative neuroanatomy for connectomics in Drosophila. eLife 5, 1133–1145 (2016).
Peters, A., Palay, S. L. & Webster, H. deF. The Fine Structure of the Nervous System: Neurons and Their Supporting Cells (Oxford Univ. Press, 1991).
Loewenstein, Y., Kuras, A. & Rumpel, S. Multiplicative dynamics underlie the emergence of the log-normal distribution of spine sizes in the neocortex in vivo. J. Neurosci. 31, 9481–9488 (2011).
Press, W. H., Teukolsky, S. A., Vetterling, W. T. & Flannery, B. P. Numerical Recipes: The Art of Scientific Computing (Cambridge Univ. Press, 2007).
Perich, M. G. et al. Inferring brain-wide interactions using data-constrained recurrent neural network models. Preprint at bioRxiv https://doi.org/10.1101/2020.12.18.423348 (2021).
Kuan, A. T. Data analysis code for ‘synaptic wiring motifs in posterior parietal cortex support decision-making’. Zenodo https://doi.org/10.5281/zenodo.10310186 (2023).
Bondanelli, G. Circuit modeling code for ‘synaptic wiring motifs in posterior parietal cortex support decision-making’. Zenodo https://doi.org/10.5281/zenodo.10200999 (2023).
Wang, Q. et al. The Allen Mouse Brain Common Coordinate Framework: a 3D reference atlas. Cell 181, 936–953 (2020).
Acknowledgements
We thank M. Eroles, M. Larson, K. Ng, L. Hulshof, T. Xie, O. Sato, T. Khanna, E. DiPietro, E. Phalen, L. Guo, S. Kushner, L. Decoursey, S. Xie, C. Zharyy, E. Zboinski, L. Sadilina, M. Narwani, R. Xu, K. Molloy, D. Patel, B. Reicher, L. Walsh, Z. Diaks, Z. Wing Fan, N. Byrne, J. Shin, T. Pedersen and A. Buckner for neuron tracing; X. Chen for providing image annotation software; W. Copeland, J. Kiperman, B. Sanders and T. Yusuf for assisting with EM imaging; W. Lou, S. Gerhard, J. Phelps and R. Zheng for assistance with image alignment; A. Buckner, R. Xu and S. Ah Kim for assistance with mouse training; N. Pettit for assistance with calcium imaging; S. Kira, A. Emanuel and A. Wang for providing mice; G. Hood for providing alignment software (AlignTK); P. Schlegel for providing analysis software (pymaid and navis); K. Hayworth for help with X-ray microCT processing; and M. Andermann, R. Born, T. Nguyen, X. Chen and D. Hidalgo for comments on the manuscript. AAV8-CaMKIIa-ChRmine-mCherry-Kv2.1 was a gift from the Deisseroth laboratory. pAAV-mDlx-NLS-mRuby2 was a gift from V. Gradinaru (Addgene, 99130-AAV9). This work was supported by the NIH (R01NS108410 to C.D.H., S.P. and W.-C.A.L.; DP1MH125776 and R01NS089521 to C.D.H.; RF1MH114047 to W.-C.A.L.; F32MH118698 to D.E.W.; and K99EB032217 to A.T.K.), the Bertarelli Program in Translational Neuroscience and Neuroengineering, Edward R. and Anne G. Lefler Center, Stanley and Theodora Feldberg Fund to W.-C.A.L. Portions of this research were conducted on the Orchestra High Performance Compute Cluster at Harvard Medical School partially provided through NIH NCRR (S10RR028832).
Author information
Authors and Affiliations
Contributions
A.T.K., G.B., L.N.D., S.P., C.D.H. and W.-C.A.L. conceptualized the project and designed experiments. L.N.D. and D.E.W. performed mouse behaviour and calcium imaging experiments. W.-C.A.L., D.G.C.H. and A.T.K. prepared tissue samples for EM. D.G.C.H. and A.T.K. performed GridTape sectioning. B.J.G. and A.T.K. developed automated EM techniques and performed EM imaging. A.T.K. and L.A.T. performed EM image processing and alignment. M.K. and A.T.K. performed co-registration between calcium imaging and EM datasets. J.H. and A.T.K. performed and managed neuron tracing and data annotation. A.T.K. performed analysis of circuit connectivity. A.T.K., G.B. and S.P. analysed functional data. G.B. performed circuit modelling. A.T.K., G.B., S.P., C.D.H. and W.-C.A.L. wrote the paper, and all of the authors assisted in reviewing and revising the manuscript.
Corresponding authors
Ethics declarations
Competing interests
Harvard University filed a patent application regarding GridTape (WO2017184621A1) on behalf of the inventors, including W.-C.A.L, D.G.C.H. and B.J.G., and negotiated licensing agreements with interested partners. The other authors declare no competing interests.
Peer review
Peer review information
Nature thanks the anonymous reviewers for their contribution to the peer review of this work.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data figures and tables
Extended Data Fig. 1 Trial-type selectivity analysis and image co-registration.
(a) Performance on the navigational two-alternative forced-choice memory task plotted over different daily sessions. Functional data from the last 4 sessions (“late”), which were closest in time to when the brain was preserved for EM, were used for all analyses with the exception of those investigating how structure-function relationships evolve over time (Extended Data Fig. 5j, k, Extended Data Fig. 6d), which also include “early” and “middle” sessions. Early, middle, and late sessions were all after the initial training of the animal and comprise expert level task performance. (b) Approximate location of cranial window and calcium imaging ROI relative to cortical regions (adapted from the Allen Mouse Brain Common Coordinate Framework77). Although this figure depicts the right hemisphere, the experimental data was collected from the left hemisphere. (c) Calcium imaging ROI: 4 500 µm x 500 µm planes in layer 2/3 (separated by 25 µm in z) were imaged at 5.3 Hz volume rate. (d) Diagram indicating temporal epochs of trial. Different trials were synchronized to temporal landmarks (trial start, cue off, trial end) and 5 temporal trial epochs (cue early, cue late, delay, turn, inter-trial interval (ITI)) were defined relative to those landmarks (Methods). Each epoch lasts 2.2-2.4 s. (e) Trial-averaged activity for left (blue) and right (red) trials for several example neurons. Neurons typically have some non-zero activity away from the temporal peak on preferred trial type. (f) Proportion of time points with significant selectivity that share the same trial-type preference as the max selectivity, showing that the left-right direction of max selectivity is generally representative of selectivity throughout the trial, especially for strongly selective neurons (n = 134 neurons active in the last 4 sessions). (g) Max selectivity for functionally characterized neurons contained within the EM volume (n = 134 neurons active in the last 4 sessions). (h) Cumulative histograms of mutual information with trial type, demonstrating that it was not statistically different between inhibitory (purple solid line) and excitatory (green solid line) neurons (n = 124 excitatory, 16 inhibitory neurons, p = 0.20, K-S test). Shading indicates 95% confidence intervals generated via bootstrap. (i) Corresponding slices from co-registered two-photon calcium imaging (2p, top) and EM (bottom) datasets. Cyan arrows – blood vessels used as landmarks. (j) Example cellular ROIs (green) co-registered to calcium imaging (top) and EM data (bottom).
Extended Data Fig. 2 Trial-averaged neuron activity.
Each row shows trial-averaged activity of a functionally-characterized neuron. Row labels are neuron IDs corresponding to the CATMAID EM database, Extended Data Figs. 3 and 4, and Supplementary Tables 1 and 2. Left and right columns show activity on left and right trials, respectively. Temporal epochs shown correspond to Extended Data Fig. 1d. Activity was averaged over the last 4 behavioural sessions, then normalized by the maximum activity rate over both trial types for each neuron separately. Neurons were sorted first by selectivity (indicated by coloured bars on the y-axis: left (max selectivity < −0.05) – blue, right (max selectivity > 0.05) – red, non-selective (−0.05 ≤ max selectivity ≤ 0.05) – grey), then by timepoint of maximum mutual information with trial type (early to late).
Extended Data Fig. 5 Co-selective Excitatory-to-Inhibitory Connectivity and Simultaneous-Selectivity Analysis.
(a) Number of synapses plotted as a function of axon/dendrite overlap for E-to-I neuron pairs, including both connected and non-connected pairs. Co-selective pairs (green) generally have more synapses per overlap than anti-selective (purple). Dotted lines indicate linear fits. (b) Normalized synapse frequency for E-to-I connections including non-connected pairs, calculated by pooling all co- or anti-selective pairs together (Methods). Co-selective pairs have higher normalized synapse frequency than anti-selective pairs (anti: 0.0315 ± 0.001 μm−1, n = 21 pairs, co: 0.0157 ± 0.001 μm−1, n = 21, p < 0.001, bootstrap test). Violin plots show distributions generated via bootstrap. (c) Maximum-selectivity similarity as a function of the maximum selectivity of the two neurons. Co-selective neuron pairs have positive (green) and anti-selective pairs have negative values (purple). (d) Correlation coefficient between normalized synapse frequency and selectivity similarity for E-to-I connections in PPC (as in Fig. 2h), calculated using different distance thresholds for determining axon/dendrite overlap (see Fig. 2b). Grey shading indicates correlations expected by chance (95% confidence interval of random shuffles). The correlation is significant over a broad range of values (~2–18) μm, and is maximized for thresholds close to value used (5 μm, dotted line). (e) Top: instantaneous selectivity for example right- (red) and left-selective (blue) neurons. Arrows indicate maximum selectivity for each neuron. Bottom: simultaneous-selectivity similarity quantifies the selectivity similarity between neurons at the same time point. Arrow indicates the maximum absolute value, which is used for x-axis values in (f) and (g). (f) Simultaneous-selectivity similarity is correlated to normalized synapse frequency (n = 21 pairs, r = 0.58, p = 0.006, Pearson correlation test). (g) Simultaneous-selectivity similarity is correlated to synapse size (n = 29 synapses, r = 0.50, p = 0.006, Pearson correlation test). (h) Number of neurons that change maximum selectivity from left to right or vice versa (“switch”), or keep the same right-left selectivity (“stay”), between early (8–10 days before sacrifice) and late (0–3 days before) sessions. Grey region indicates expected range if neurons switch at chance levels. (i) Number of neuron pairs that change maximum-selectivity similarity from co- to anti-selective or vice versa (“switch”), or keep the same similarity (“stay”) between early and late sessions. Grey region indicates expected range if neurons switch at chance levels. (j) Correlations between normalized synapse frequency and maximum-selectivity similarity for E-to-I connections, calculated using functional data from early (8–10 days before sacrifice), middle (4–7 days before), and late sessions (0–3 days before). Correlation coefficients are lower for earlier sessions (early: r = 0.31, p = 0.27, middle: r = 0.53, p = 0.019, late: r = 0.66, p = 0.001, Pearson correlation tests). (k) Correlations between PSD area and maximum-selectivity similarity for E-to-I synapses, calculated for early, middle, and late sessions. Correlation coefficients are lower for earlier sessions (early: r = 0.19, p = 0.42, middle: r = 0.46, p = 0.023, late: r = 0.56, p = 0.002, Pearson correlation tests). All statistics reported as mean ± standard error with two-tailed significance tests.
Extended Data Fig. 6 Anti-selective Inhibitory-to-Excitatory Connectivity and Structure-function Analysis Across Connection Types.
(a) Number of synapses plotted as a function of axon/dendrite overlap for I-to-E neuron pairs, including both connected and non-connected pairs. Anti-selective pairs (purple) generally have more synapses per overlap than co-selective pairs (green). Dotted lines indicate linear fits. (b) Normalized synapse frequency for I-to-E connections including non-connected neuron pairs, calculated by pooling co- or anti-selective pairs together (Methods). Anti-selective pairs have higher normalized synapse frequency than co-selective pairs (anti: 0.0109 ± 0.0004 µm−1, n = 99 pairs, co: 0.0096 ± 0.0003 µm−1, n = 112, p = 0.03, bootstrap test). Violin plots show distributions generated via bootstrap. (c) Correlation coefficient between normalized synapse frequency and maximum-selectivity similarity for I-to-E connections (as in Fig. 3c, calculated using different distance thresholds for determining axon/dendrite overlap (see Fig. 2b). Grey shading indicates correlations expected by chance (95% confidence interval of random shuffles). The correlation is significant over a range of values (~3–15 μm), and is maximized for distances thresholds close to value used (5 μm, dotted line). (d) Normalized synapse frequency plotted as a function of maximum-selectivity similarity for I-to-E connections in PPC, calculated using functional data from early (8–10 days before sacrifice), middle (4–7 days before), and late (0–3 days before) sessions. Correlation coefficients are less negative for earlier sessions (early: r = −0.06, p = 0.52; middle: r = −0.05, p = 0.64; late: r = −0.22, p = 0.021, Pearson correlation test). (e) Simultaneous-selectivity similarity is negatively correlated to normalized synapse frequency (r = −0.20, p = 0.041, Pearson correlation test). (f–m): Structure-function correlations for all connection types in PPC. For each panel, the scatterplot on the left side shows the correlation between maximum-selectivity similarity and normalized synapse frequency (top row) or synapse size (bottom row). Different connection types are shown in each column (from left to right: E-to-E, E-to-I, I-to-E, I-to-I). Red line indicates linear fit. (f) E-to-E synapse frequency: n = 14, c = −0.40, p = 0.15. (g) E-to-I synapse frequency: n = 21, c = 0.65, p = 0.001. (h) I-to-E synapse frequency: n = 103, c = −0.22, p = 0.023. (i) I-to-I synapse frequency: n = 10, c = −0.37, p = 0.30. (j) E-to-E synapse size: n = 14, c = 0.20, p = 0.50. (k) E-to-I synapse size: n = 29, c = 0.56, p = 0.002. (l) I-to-E synapse size: n = 156, c = 0.084, p = 0.29. (m) I-to-I synapse size: n = 14, c = 0.092, p = 0.75. All p-values are from two-tailed significance tests.
Extended Data Fig. 7 Connections Between Cells with Activity Peaks in Different Epochs and Noise Correlations.
(a) Example activity of connected neurons that have activity peaks during the same or different epochs. Top row indicates presynaptic excitatory neuron, bottom row indicates postsynaptic inhibitory neuron. Trial-averaged activity for right (red) and left (blue) trials are plotted for each neuron. Temporal epochs are indicated by colours as in Extended Data Fig. 1d. (b) Proportion of connected (left) and non-connected (right) E-to-I pairs that have activity peaks in the same (blue) or different (orange) temporal epochs. (c) Proportion of connected (left) and non-connected (right) I-to-E pairs that have activity peaks in the same (blue) or different (orange) temporal epochs. (d) Noise correlations were positively correlated to normalized synapse frequency for E-to-I connections (n = 19 connections, r = 0.45, p = 0.026, one-tailed Pearson correlation test). (e) Noise correlations were negatively correlated to normalized synapse frequency for I-to-E connections (n = 84, r = −0.26, p = 0.009, one-tailed Pearson correlation test).
Extended Data Fig. 8 Decision-making circuit dynamics in the linear rate model.
(a) Cartoon illustrating the distance between mean responses to left and right trial types (coloured dots) in the state-space of the excitatory neuron activity. This distance is enhanced by opponent inhibition between trial-type selective subnetworks. On left trial-types, both suppression of ER and amplification of EL (grey arrows) contribute to increased separation along the signal axis (and symmetrically for right trials), thus enhancing the encoding of trial-type in presence of readout noise. (b) Values of the steady-state activity of EL (top) and ER (bottom) units as a function of the I-to-E (x-axis) and E-to-I (line colours) connection selectivity. (c) Normalized distance between the mean activity corresponding to left and right trial-types (see panel (a)), as a function of I-to-E and E-to-I connection selectivity. (d) Decoding accuracy computed by linearly decoding the trial-type from the excitatory units in presence of readout noise, as a function of the E-to-I and I-to-E connection selectivity. The region with anti-selective I-to-E selectivity (purple arrow and box) corresponds to networks with opponent inhibition, and the black square denotes the parameter values used in simulations of Fig. 4b. The grey area corresponds to unstable network dynamics. (e) Decoding accuracy as a function of E-to-I (top) and I-to-E (bottom) selectivity, corresponding to two cuts of the phase plot (dashed lines in panel (d)). In panels (d) and (e), purple and green arrows indicate the directions where connection motifs increase respectively their anti- and co-selectivity (see Fig. 4a). (f) Dependence of the relative trial-type decoding accuracy (see Fig. 4c) as a function of the selectivity of the four connection types. Each subplot shows the dependence on E-to-I and I-to-E selectivity (as in Fig. 4c); different subplots correspond to different values of the E-to-E and I-to-I connection selectivity. The subplot highlighted by a magenta box corresponds to the one shown in Fig. 4c. In the linear rate model (Fig. 4a–d), the region corresponding to selective E-to-I and anti-selective I-to-E (left quadrant) becomes linearly unstable when the E-to-E and I-to-I selectivity (ΔEE and ΔII) satisfy \({(1-\Delta }_{{EE}}){(1+\Delta }_{{II}}) < 0\). (g–h) Absolute and relative trial-type decoding accuracy for time-dependent input noise, i.e. input noise that varies in time. Here the noise affects only the excitatory units, which are also the units that receive external trial-specific input. (i–j) Same as (g-h) for time-independent input noise, i.e. noise that affects the external input on the excitatory units on a single-trial basis, but is otherwise constant in time.
Extended Data Fig. 9 Decision-making circuit dynamics in models receiving external input onto both excitatory and inhibitory units.
(a) Decoding accuracy of the linear rate model, computed by linearly decoding the trial-type from the excitatory units in presence of readout noise, as a function of the E-to-I and I-to-E connection selectivity. The upper left quadrant corresponds to networks with opponent inhibition. The grey area corresponds to unstable network dynamics. (b) Relative decoding accuracy of linear rate model, defined as the ratio of output to input decoding accuracy, as a function of E-to-I and I-to-E selectivity. Compared to the corresponding model with external inputs only onto E units (Fig. 4c), this model shows a larger improvement in decoding accuracy from opponent inhibition (higher relative decoding accuracy when opponent inhibition is present, and lower relative decoding accuracy when opponent inhibition is absent). (c) Correlation coefficient between connection strengths and max-selectivity similarity across trained RNNs (n = 164). The majority of networks exhibit opponent inhibition. Compared to RNNs trained with external inputs only onto E neurons (Fig. 4h), I-to-E connections are more anti-selective, whereas E-to-I connections are less co-selective. (d) Distance between left and right RNN activity (averaged across time) as a function of selectivity perturbations, normalized by its value in the unperturbed network (dashed lines). E-to-I (top) and I-to-E (bottom) connection weights were perturbed in a way that changes anti- (purple arrow) or co-selectivity (green arrows) without changing the average connection weight. Solid lines represent the median across networks, error bars correspond to the 25th and 75th percentile across networks (n = 164). Stronger opponent inhibition (stronger E-to-I co-selectivity or I-to-E anti-selectivity) amplified the separation between left and right population responses, similar to the effect of perturbations on RNNs with external inputs on E units only (Fig. 4i).
Extended Data Fig. 10 Functional role of connections between neurons with activity peaks in the same or different task epochs for sequential, trial-type selective population dynamics in the recurrent neural networks.
(a) Correlation coefficient between connection strengths and maximum-selectivity similarity across trained RNNs (as in Fig. 4h). Opponent inhibition motif (co-selective E-to-I and anti-selective I-to-E connections) is observed when separately considering connections between neurons with activity peaks in the same temporal epoch within the trial (i.e. ‘same-epoch’ connections; blue distributions, ‘same’) and connections between neurons peaking at different temporal epochs (i.e. ‘different-epoch’ connections; red distributions, ‘diff’). (b) Distance between left- and right RNN activity (averaged across time) as a function of selectivity perturbations of ‘same-epoch’ and ‘different-epoch’ connections separately, normalized by its value in the unperturbed networks (dashed lines). E-to-I (left) and I-to-E (right) connection weights were perturbed in a way that altered connection selectivity without changing the average connection weight. For each connection type (E-to-I or I-to-E) perturbations were induced separately on the ‘same-epoch’ and ‘diff-epoch’ connections. Solid lines represent the median across networks, error bars correspond to the 25th and 75th percentile across networks (as in Fig. 4i). (c) Contributions from off-peak neural activity to the distance between left and right RNN activity (averaged across time) as a function of ‘different-epoch’ E-to-I (left) and I-to-E (right) selectivity perturbations, normalized by its value in the unperturbed network (dashed lines; as in Fig. 4i). Solid lines represent the median across networks, error bars correspond to the 25th and 75th percentile across networks (n = 147, randomly initiated artificial neural networks). The contribution from off-peak activity to the distance was computed as follows. Prior to averaging the distance over time points, at each time point the distance was computed by selecting only neurons that were not at their activity peaks at that specific time point. (d) Same as (c) for the contribution of peak activity to the distance between trial-type selective population activity. Prior to averaging the distance over time points, at each time point the distance was computed by selecting only neurons that were at their activity peaks at that specific time point. (e) Total excitatory (left) and inhibitory (right) currents received by postsynaptic excitatory neurons through ‘same-epoch’ (blue) and ‘different-epoch’ connections (red), centred around the peak time of the postsynaptic neurons’ activity averaged over preferred and non-preferred trial types. Currents are computed as the mean across trained RNNs of single-cell currents, summed over presynaptic neurons and averaged over postsynaptic neurons and over preferred and non-preferred trial types. (f) Values of the sequentiality index of neural responses across all fit RNNs (grey), RNNs with reduced strength of ‘same-epoch’ (blue) and ‘different-epoch’ (red) connections. The sequentiality index was defined as the Spearman (rank) correlation between the order of activity peak times across neurons in PPC data and the fit RNN responses (see Supplementary Methods Section “Analysis of the function of ‘same-epoch’ and ‘different-epoch’ RNN connectivity”). (g) Median excitatory activity across RNNs subject to different types of connection perturbations. Left: all fit RNNs (without perturbations); centre left and centre right: RNNs with reduced strength of ‘same-epoch’ or ‘different-epoch’ connections respectively; right: RNNs with ‘different-epoch’ connections entirely removed prior to training (only ‘same’ connections are trained). (h) Peak activity values averaged across neurons and stimuli across all fit RNNs (grey), RNNs with reduced strength of ‘same-epoch’ (blue) and ‘different-epoch’ (red) connections (as in panel g). (i-m) Same as (h) for the off-peak activity, ratio of peak to off-peak activity, trial-type selectivity of the peak and off-peak activity and overall selectivity (see Supplementary Methods Section “Analysis of epoch-specific RNN connectivity”). In panels (h-m) and the selectivity was defined as the normalized difference between left/right trial- and time-averaged activity (i.e. \({({\boldsymbol{r}}}_{{\boldsymbol{L}}}{\boldsymbol{-}}{{\boldsymbol{r}}}_{{\boldsymbol{R}}})/{({\boldsymbol{r}}}_{{\boldsymbol{R}}}+{{\boldsymbol{r}}}_{{\boldsymbol{L}}})\) or \({({\boldsymbol{r}}}_{{\boldsymbol{R}}}{\boldsymbol{-}}{{\boldsymbol{r}}}_{{\boldsymbol{L}}})/{({\boldsymbol{r}}}_{{\boldsymbol{R}}}+{{\boldsymbol{r}}}_{{\boldsymbol{L}}})\) respectively for left and right neurons) and was averaged across all neurons belonging to each left/right selective sub-population comprising both excitatory and inhibitory neurons, then averaged over left and right sub-populations. In all panels, the peak activity was defined as the activity at times ranging from −10 to +10 time bins around the time at which the activity averaged across left and right trial types was maximum. In all panels the results are shown across (n = 147) RNNs.
Extended Data Fig. 11 Error-Trial Selectivity Degradation in PPC activity and in models.
(a) Average activity (experimental calcium imaging data) in correct and error trials for example excitatory (top row) and inhibitory neurons (bottom row). In each plot, neuronal activity is plotted as a function of time. Trials are grouped based on whether the white (left column) or black cue (right column) was presented and coloured based on which direction the mouse turned. Arrows indicate activity changes on error trials that degrade selectivity: activity is generally decreased in preferred cue trials (Δpref, i.e. white/black for left/right – selective neurons) and increased on non-preferred cue trials (Δnon, i.e. black/white for left/right – selective neurons). (b) Error-trial selectivity degradation (Δpref +Δnon, see (a)), which quantifies how much selective activity is degraded on error versus correct trials (Methods), for E and I neurons. Each datapoint indicates one neuron. At the population level, both E and I neurons both had positive degradation (E: 0.74 ± [−0.52, 0.80], p < 0.001; I: 0.52 ± [−0.28,−0.79], p = 0.002, Wilcoxon signed-rank tests), and there was no significant difference between E and I (p = 0.99, Mann-Whitney U-test). (c) Median error trial selectivity degradation for E and I neurons in the EM mouse (same as (b), n = 1 mouse, 120 E neurons, 16 I neurons, 11 sessions, 95–1354 trials per neuron), additional mice (n = 3, 824 E neurons, 96 I neurons, 1-2 sessions, 149–395 trials per neuron), and the linear rate model. Error bars show 95% confidence interval of the median. For experimental animals, both E and I neurons exhibited significant error trial selectivity degradation (EM mouse: E: 0.82, I: 0.72, additional mice: E: 0.38, I: 0.48, p < 0.001 for all cases, permutation tests), and the there was no significant difference between E and I neurons (EM mouse: p = 0.45, additional mice; p = 0.31, permutation tests). The model shows qualitatively similar results (E: 1.12, I: 1.07). (d) Example imaging plane from an additional mouse showing labelling strategy to distinguish excitatory and inhibitory neurons. In addition to the green calcium indicator, inhibitory neurons were labelled with a nuclear-localized red fluorophore (magenta arrows) and excitatory neurons were labelled with a somatic membrane-localized red fluorophore. To identify neurons with enriched nuclear red fluorescence, we computed the ratio of nuclear to membrane fluorescence. Insets show average images of cells deemed excitatory (top right) and inhibitory (bottom right). Scale bars: Left: 50 μm, Right: 10 μm. (e) Error-trial selectivity degradation as a function of E-to-I (y-axis) and I-to-E (x-axis) selectivity for the linear rate model. Values for E and I units are both positive and similar in magnitude (ratio ≈ 1) only in networks exhibiting opponent inhibition (E-to-I selectivity > 0 and I-to-E selectivity <0). Grey square marker indicates the selectivity values corresponding to the bars in (c). Statistics reported as mean ± standard error with two-tailed significance tests.
Extended Data Fig. 12 Tests of methods for estimation of noise correlations and information.
(a,b) Illustration of the distributions of firing activity of two neurons of fixed trial type across multiple trials for two simulated example sessions. The correlation value is different across the two sessions (heterogeneity across sessions). (c) Distributions of the firing activity of the two neurons pooled across sessions 1 and 2 (pooled distributions). In presence of heterogeneity of correlation values across sessions 1 and 2, the correlation value computed from the pooled distribution is equal to the value averaged across session 1 and 2. (d,e) Illustration of the distributions of the firing activity of a neuron across multiple trials for two trial types in two example sessions. The trial-type information value of the neuron is different across the two sessions (heterogeneity across sessions), as a result of changes in the trial-type-specific mean firing and firing variance across trial types. (f) Distributions of firing activity of the neuron across the two trial types pooled across sessions 1 and 2 (pooled distributions). Because of the heterogeneity in the neuron’s firing parameters across sessions 1 and 2, the trial-type information of the pooled distribution is less than the average information on session 1 and 2. (g) Average correlation and RMSE of correlation computed without regressing out behaviour (grey) and regressing out behaviour (black). (h) Average correlation and RMSE of correlation computed using the average value across sessions (black) or pooling data across sessions (red) when no heterogeneity across sessions is present. (i) Same as (h) when realistic heterogeneity across sessions is present. (j) Average trial-type information and RMSE of trial-type information computed without using a bias correction (grey) and using the Panzeri-Treves (PT) bias correction (black). (k) Average trial-type information and RMSE of information computed using the average value across sessions (black) or pooling data across sessions (red) when no heterogeneity across sessions is present. (l) Same as (k) when realistic heterogeneity across sessions is present. In all panels, left and right plots respectively show the average value and the root mean squared error (RMSE) computed across 5000 simulations of the correlation (g-i) or information (j-l) values.
Supplementary information
Supplementary Information
Supplementary Methods and Supplementary Tables 1 and 2.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Kuan, A.T., Bondanelli, G., Driscoll, L.N. et al. Synaptic wiring motifs in posterior parietal cortex support decision-making. Nature 627, 367–373 (2024). https://doi.org/10.1038/s41586-024-07088-7
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/s41586-024-07088-7
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
