On 2017 Apr 07, Janelia Neural Circuit Computation Journal Club commented:

Highlight/Summary Wilson and colleagues measured the orientation selectivity of dendritic spines on layer 2/3 pyramidal neurons in ferret visual cortex using two-photon calcium imaging. They are trying to address the question of how tuned output is produced by neurons that receive diverse types of input, a core question in neural computation. They suggest that higher orientation selectivity of individual neurons was correlated with greater clustering of similarly tuned spines (and not a narrower distribution spine preference or residence in a low-rate-of-change location of the orientation preference map). They conclude that dendritic nonlinearities play a critical role in shaping orientation selectivity.

Impact / strengths This is the first study of the functional organization of synaptic input onto the dendrites in a cortical region with columnar architecture. Previous work in mouse visual cortex (Jia et al., 2010) -- which lacks a columnar organization of orientation -- found no evidence of spatial clustering of input orientation preference on the dendrites of layer 2/3 neurons. On a technical level, this work is among only a handful of studies that have used genetically-encoded calcium indicators to estimate the functional selectivity of input to individual spines (cf Chen et al 2013).

By combining two-photon imaging of the dendritic arbor with intrinsic imaging of the orientation preference map, an impressive amount of information was gathered for each neuron. Although the number of neurons was modest (N=9), the number of identifiable spines imaged per neuron (N~300) is respectable for addressing questions of clustering. Signal-to-noise ratio at individual spines was also relatively high, comparable to previous studies (Chen et al., 2013).

Weaknesses The conclusions of this paper are critically dependent on the estimation of calcium influx through synaptic NMDA-Rs and contributions of local-dendritic calcium signals, as well as their interactions. Dendritic electrical events, including local synaptic potentials, local dendritic spikes, and depolarization produced by more global back-propagation of action potentials, cause dendritic calcium influx through voltage gated calcium channels (VGCCs) and also modulate NMDA-Rs. Depolarization interacts non-linearly with NMDA-R Ca influx. The field has not yet converged on methods to disambiguate global and local contributions to spine calcium signals. This is a critical issue for the key claim of the paper (‘clustering of similarly tuned spines’). For example, do the apparent clusters of similarly tuned spines drive the cell (the interpretation that is offered)? Or does the postsynaptic response at an optimal stimulus for the cell potentiate NMDA-R Ca influx in spines to produce an apparent cluster?

Wilson and colleagues estimate the spine signal by subtracting a linear fit of the spine fluorescence against the dendritic fluorescence. They then apply four inclusion criteria (pg 1004 and Methods). After this procedure, spines with orientation preference similar to the parent soma have similar tuning to spines with orthogonal orientation preference; this they take as an indication that the subtraction was effective. Given that some of the inclusion criteria are related to tuning curves (e.g., tuning must be well-described by a Gaussian fit) it was not clear if the inclusion criteria themselves biased this analysis. We would have welcomed analyses of the robustness of the results to changes in the parameters of the subtraction procedures and inclusion criteria.

A related concern is about the spatial scale of the dendritic hotspots. The size of the hotspots is not much smaller than the field-of-view of the images. The inclusion criteria (e.g., a Gaussian fit must explain 70% of the variance) have a possibly strong but unevaluated effect on the results.

On 2017 Apr 07, Janelia Neural Circuit Computation Journal Club commented:

Highlight/Summary Wilson and colleagues measured the orientation selectivity of dendritic spines on layer 2/3 pyramidal neurons in ferret visual cortex using two-photon calcium imaging. They are trying to address the question of how tuned output is produced by neurons that receive diverse types of input, a core question in neural computation. They suggest that higher orientation selectivity of individual neurons was correlated with greater clustering of similarly tuned spines (and not a narrower distribution spine preference or residence in a low-rate-of-change location of the orientation preference map). They conclude that dendritic nonlinearities play a critical role in shaping orientation selectivity.

Impact / strengths This is the first study of the functional organization of synaptic input onto the dendrites in a cortical region with columnar architecture. Previous work in mouse visual cortex (Jia et al., 2010) -- which lacks a columnar organization of orientation -- found no evidence of spatial clustering of input orientation preference on the dendrites of layer 2/3 neurons. On a technical level, this work is among only a handful of studies that have used genetically-encoded calcium indicators to estimate the functional selectivity of input to individual spines (cf Chen et al 2013).

By combining two-photon imaging of the dendritic arbor with intrinsic imaging of the orientation preference map, an impressive amount of information was gathered for each neuron. Although the number of neurons was modest (N=9), the number of identifiable spines imaged per neuron (N~300) is respectable for addressing questions of clustering. Signal-to-noise ratio at individual spines was also relatively high, comparable to previous studies (Chen et al., 2013).

Weaknesses The conclusions of this paper are critically dependent on the estimation of calcium influx through synaptic NMDA-Rs and contributions of local-dendritic calcium signals, as well as their interactions. Dendritic electrical events, including local synaptic potentials, local dendritic spikes, and depolarization produced by more global back-propagation of action potentials, cause dendritic calcium influx through voltage gated calcium channels (VGCCs) and also modulate NMDA-Rs. Depolarization interacts non-linearly with NMDA-R Ca influx. The field has not yet converged on methods to disambiguate global and local contributions to spine calcium signals. This is a critical issue for the key claim of the paper (‘clustering of similarly tuned spines’). For example, do the apparent clusters of similarly tuned spines drive the cell (the interpretation that is offered)? Or does the postsynaptic response at an optimal stimulus for the cell potentiate NMDA-R Ca influx in spines to produce an apparent cluster?

Wilson and colleagues estimate the spine signal by subtracting a linear fit of the spine fluorescence against the dendritic fluorescence. They then apply four inclusion criteria (pg 1004 and Methods). After this procedure, spines with orientation preference similar to the parent soma have similar tuning to spines with orthogonal orientation preference; this they take as an indication that the subtraction was effective. Given that some of the inclusion criteria are related to tuning curves (e.g., tuning must be well-described by a Gaussian fit) it was not clear if the inclusion criteria themselves biased this analysis. We would have welcomed analyses of the robustness of the results to changes in the parameters of the subtraction procedures and inclusion criteria.

A related concern is about the spatial scale of the dendritic hotspots. The size of the hotspots is not much smaller than the field-of-view of the images. The inclusion criteria (e.g., a Gaussian fit must explain 70% of the variance) have a possibly strong but unevaluated effect on the results.