Functional specialization of mouse higher visual cortical areas

Abstract

The mouse is emerging as an important model for understanding how sensory neocortex extracts cues to guide behavior, yet little is known about how these cues are processed beyond primary cortical areas. Here, we used two-photon calcium imaging in awake mice to compare visual responses in primary visual cortex (V1) and in two downstream target areas, AL and PM. Neighboring V1 neurons had diverse stimulus preferences spanning five octaves in spatial and temporal frequency. By contrast, AL and PM neurons responded best to distinct ranges of stimulus parameters. Most strikingly, AL neurons preferred fast-moving stimuli while PM neurons preferred slow-moving stimuli. By contrast, neurons in V1, AL, and PM demonstrated similar selectivity for stimulus orientation but not for stimulus direction. Based on these findings, we predict that area AL helps guide behaviors involving fast-moving stimuli (e.g., optic flow), while area PM helps guide behaviors involving slow-moving objects.

Figure 1. Widefield Functional Imaging in Awake Mice Suggests Differences across Visual Areas

(A) To target expression of calcium indicators to higher visual areas, we first mapped changes in the intrinsic autofluorescence signal in awake mice during presentation of a local patch (40° diameter, smooth edges) containing upward-drifting sinusoidal gratings, at one of two spatial locations in the upper visual field (14° elevation; 60° or 96° azimuth; darker regions in A, bottom panel, reflect areal responses to stimulation at 96°). Retinotopic maps were generated using blue/ red pseudocolor images of responses to the two stimulus positions (A, upper panel). These maps reliably delineated primary visual cortex (V1) and several higher visual areas (LM: lateromedial; AL: anterolateral; RL: rostrolateral; A: anterior; AM: anteromedial; PM: posteromedial; cf. Wang and Burkhalter, 2007). See Figure S1 for further details. (B) Differences in intrinsic autofluorescence signal among lateral visual areas (top panels, cf. beige rectangle in A) and among medial visual areas (bottom panels, cf. orange rectangle in A) to stimuli varying in spatial frequency (left panels) and temporal frequency (right panels). Data were averaged across three imaging sessions (same mouse as in A). Note that responses were similar in areas V1, LM, and AM, while responses were distinct in areas AL and PM. (C) Targeted viral expression of GCaMP3 calcium indicator in areas AL and PM (purple and green dashed regions). (D) Average calcium responses (lighter spots) using widefield GCaMP3 imaging confirmed the presence of clear differences in spatial and temporal frequency tuning in 600-µm-wide regions of interest encompassing area AL (top panel) versus area PM (bottom panel). Gaussian smoothing: σ = 12 µm. ΔF/F: percent change in fluorescence. Yellow diagonal lines are iso-speed lines. Scale bars in (A) and (C), 1 mm.

Figure 2. Cellular Imaging of Spatial and Temporal Frequency Responses

(A) Maximum-intensity projections of baseline two-photon fluorescence volumes (GCaMP3), recorded in area AL (left panel) and PM (right panel) of the same mouse. Scale bars, 50 µm. Middle panel: epifluorescence image of targeted GCaMP3 expression in areas AL and PM (green halos), superimposed on the brightfield image. Scale bar, 1 mm. (B and C) Average fluorescence time courses of single neurons (yellow circles in A) from areas AL (B) and PM (C) during presentation of stimuli at various spatial and temporal frequencies (top panels). ΔF/F: percent change in fluorescence. Shaded regions are ± SEM. The average change in fluorescence during presentation of each stimulus (blue bars, top panels) was used to generate a response map (bottom left panels) that was then fit to a two-dimensional Gaussian (bottom middle panels). Contours plots of the model fit at 60% of peak (red, halfwidth of σ) and 88% of peak (blue, halfwidth of σ/2), respectively, are overlaid on the same model fit, rendered with higher sampling (bottom right panels; Experimental Procedures). See also Figure S2.

Figure 3. AL and PM Neurons Prefer Nearly Nonoverlapping Ranges of Speeds Spanned by V1

(A) Example response profiles of subsets of neurons from the same local volume in each cortical area. Simultaneously recorded neurons in V1 often preferred very different combinations of spatial and temporal frequency. By contrast, most neurons in AL preferred low spatial and high temporal frequency, while most neurons in PM preferred high spatial and low temporal frequency. Blue and red ellipses are contours of model fits at 88% and 60% of peak, respectively. (B) Coverage of the spatiotemporal frequency plane by V1, AL, and PM populations is illustrated by shaded contours (at 88% of peak; left panel) and by scatter plots of neurons’ preferred spatial and temporal frequencies (right panel). Yellow lines in (A) and (B) represent the line in the spatiotemporal frequency plane that best classified response preferences as belonging to AL or PM neurons (an iso-speed line at 41.9°/s). (C and D) Bar plots of median preferred and high cutoff (50% of peak response) spatial and temporal frequencies across areas. (E) Distributions of peak speeds (ratio of preferred temporal:spatial frequencies). Darker bars (and darker symbols at edges of scatter plots in right panels of B) indicate neurons with low-or high-pass tuning for spatial or temporal frequency. (F) Bar plots of median peak speed across areas. Error bars in (C), (D), and (F) are 95% confidence intervals on estimated median values (maximum-likelihood estimate for lognormal distribution). Numbers of cells: 87 in V1,107 in AL, and 46 in PM. Peak % ΔF/F for example neurons in (A), clockwise from top-left, V1:28,9, 12, 27, 11, 8; AL: 56, 14, 74, 22, 34, 41; PM: 7, 4, 8, 8, 15, 13. See also Figure S3.

Figure 4. Tuning for Speed in Individual Neurons

(A) Top panel: yellow iso-speed lines at fixed ratios of temporal:spatial frequency. Bottom panels: neurons with the same peak speed across spatial frequencies (bottom right) are “tuned for speed” (power law relationship between temporal and spatial frequency with exponent ξ ≈ 1 in model fit, see Experimental Procedures). (B) Top panels: example neuron in PM that is approximately tuned for speed (top right; gray bar at preferred speed; ξ =1.18) but not for temporal frequency (top left). Bottom panels: example neuron in AL that is not tuned for speed (bottom right; ξ= −0.07) but is tuned for temporal frequency (bottom left; gray bar at preferred temporal frequency). Insets in left panels: cell responses across spatial and temporal frequencies (cf. Figure 2). (C and D) Distributions (C) and average values (D, mean ± SEM) of ξ across areas. Only neurons for which ξ could be accurately estimated are included (i.e., neurons with peak spatial and temporal frequencies contained in our sampling range, and with spatial and temporal frequency bandwidths greater than the sampling resolution of 1 octave). (E) Scatter plot of tuning for speed (ξ) versus peak speed reveals an inverse relationship, both across and within areas (see Results; black line: least-squares fit across neurons in all areas). Numbers of included cells: 20 in V1, 46 in AL, and 13 in PM. See also Experimental Procedures and Figure S4.

Figure 5. Similarity in Orientation Selectivity, but Not Direction Selectivity, between Visual Areas

In a second protocol, stimuli were presented at one of 8 directions and 5–6 spatial frequencies. (A) Top row: polar plots of direction tuning at the neurons’ preferred spatial frequencies, for simultaneously recorded neurons in V1. Shaded regions are ± SEM. Values at lower right are maximum response strength (ΔF/F). Middle/bottom rows: example neurons in AL and PM of the same mouse. (B) Distributions of direction selectivity (left panels) and orientation selectivity (right panels) across areas. (C) Mean direction selectivity (top panel) and orientation selectivity (bottom panel) across areas. Values are ± SEM. Numbers of cells: 78 in V1, 40 in AL, and 43 in PM. See also Figure S5.

Figure 6. Areal Differences in Peak Speed Are Largely Independent of Locomotion

(A) Examples of neural responses across temporal frequencies (at the preferred spatial frequency) are shown for a neuron in PM (left panel) and in AL (right panel), illustrating shifts in both response strength (left) and in preferred temporal frequency and speed (right) with locomotion. (B and C) Increases in response strength were observed in all areas as demonstrated in scatter plots across behavioral conditions (B) and in population averages (C; mean ± SEM; see Results). (D and E) Small but consistent increases in peak speed were also observed in scatter plots (D) and population averages (E, top panel; mean ± SEM, in octaves; see Results). These increases in peak speed were mainly due to increases in preferred temporal frequency (E, middle panel) but not spatial frequency (E, lower panel). Numberof included neurons:35 in V1, 27in AL, and 8 in PM. See also Figures S6 and S2.

Figure 7. Model of Areal Specialization in Awake Mouse Visual Cortex

Our results reveal intercalated groups neurons in V1 that prefer a broad range of stimulus speeds, while neurons in PM respond to low speeds and neurons in AL respond to high speeds. We predict that this specialization in AL and PM may arise in part from selective input from subsets of V1 neurons whose speed sensitivity matches that of the target area. Differences in peak speed, anatomical location, and connectivity also suggest that area AL may contribute to behaviors involving high-speed stimuli (e.g., optic flow during navigation), while PM may contribute to behaviors involving monitoring of slow-moving objects (see Figure S4).