Skip to main content

Main menu

  • HOME
  • CONTENT
    • Early Release
    • Featured
    • Current Issue
    • Issue Archive
    • Collections
    • Podcast
  • ALERTS
  • FOR AUTHORS
    • Information for Authors
    • Fees
    • Journal Clubs
    • eLetters
    • Submit
  • EDITORIAL BOARD
  • ABOUT
    • Overview
    • Advertise
    • For the Media
    • Rights and Permissions
    • Privacy Policy
    • Feedback
  • SUBSCRIBE

User menu

  • Log in
  • My Cart

Search

  • Advanced search
Journal of Neuroscience
  • Log in
  • My Cart
Journal of Neuroscience

Advanced Search

Submit a Manuscript
  • HOME
  • CONTENT
    • Early Release
    • Featured
    • Current Issue
    • Issue Archive
    • Collections
    • Podcast
  • ALERTS
  • FOR AUTHORS
    • Information for Authors
    • Fees
    • Journal Clubs
    • eLetters
    • Submit
  • EDITORIAL BOARD
  • ABOUT
    • Overview
    • Advertise
    • For the Media
    • Rights and Permissions
    • Privacy Policy
    • Feedback
  • SUBSCRIBE
PreviousNext
Research Articles, Systems/Circuits

Dynamics and Mechanisms of Contrast-Dependent Modulation of Spatial-Frequency Tuning in the Early Visual Cortex

Hiroki Tanaka and Ryohei Sawada
Journal of Neuroscience 14 September 2022, 42 (37) 7047-7059; DOI: https://doi.org/10.1523/JNEUROSCI.2086-21.2022
Hiroki Tanaka
1Faculty of Information Science and Engineering, Kyoto Sangyo University, Motoyama, Kamigamo, Kita-ku, Kyoto, 603-8555, Japan
2Graduate School of Frontier Biosciences, Osaka University, Yamadaoka, Suita, Osaka, 565-0871, Japan
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
Ryohei Sawada
1Faculty of Information Science and Engineering, Kyoto Sangyo University, Motoyama, Kamigamo, Kita-ku, Kyoto, 603-8555, Japan
  • Find this author on Google Scholar
  • Find this author on PubMed
  • Search for this author on this site
  • Article
  • Figures & Data
  • Info & Metrics
  • eLetters
  • PDF
Loading

Abstract

The spatial-frequency (SF) tuning of neurons in the early visual cortex is adjusted for stimulus contrast. As the contrast increases, SF tuning is modulated so that the transmission of fine features is facilitated. A variety of mechanisms are involved in shaping SF tunings, but those responsible for the contrast-dependent modulations are unclear. To address this, we measured the time course of SF tunings of area 17 neurons in male cats under different contrasts with a reverse correlation. After response onset, the optimal SF continuously shifted to a higher SF over time, with a larger shift for higher contrast. At high contrast, whereas neurons with a large shift of optimal SF exhibited a large bandwidth decrease, those with a negligible shift increased the bandwidth over time. Between these two extremes, the degree of SF shift and bandwidth change continuously varied. At low contrast, bandwidth generally decreased over time. These dynamic effects enhanced the processing of high-frequency range under a high-contrast condition and allowed time-average SF tuning curves to show contrast-dependent modulation, like that of steady-state SF tuning curves reported previously. Combinations of two mechanisms, one that decreases bandwidth and shifts optimal SF, and another that increases bandwidth without shifting optimal SF, would explain the full range of SF tuning dynamics. Our results indicate that one of the essential roles of tuning dynamics of area 17 neurons, which have been observed for various visual features, is to adjust tunings depending on contrast.

SIGNIFICANCE STATEMENT The spatial scales of features transmitted by cortical neurons are adjusted depending on stimulus contrast. However, the underlying mechanism is not fully understood. We measured the time course of spatial frequency tunings of cat area 17 neurons under different contrast conditions and observed a variety of dynamic effects that contributed to spatial-scale adjustment, allowing neurons to adjust their spatial frequency tuning range depending on contrast. Our results suggest that one of the essential roles of tuning dynamics of area 17 neurons, which have been observed for various visual features, is to adjust tunings depending on contrast.

  • coarse-to-fine processing
  • contrast modulation
  • spatial frequency tuning
  • tuning dynamics
  • visual cortex

Introduction

Spatial frequency (SF) is a fundamental visual feature systematically represented in the striate cortex (Bonhoeffer et al., 1995; Issa et al., 2000; Nauhaus et al., 2012, 2016; Ribot et al., 2013), with neurons narrowly tuned to it (Campbell et al., 1969; Movshon et al., 1978; De Valois et al., 1982). Since many receptive field (RF) properties, such as spatial or temporal integration, depend on stimulus contrast to optimize information transmission (Shapley and Victor, 1979; Sceniak et al., 1999; Bair and Movshon, 2004), the study of contrast dependence of SF tuning and its underlying mechanisms is critical to understanding how the cortex achieves optimal SF transmission. By measuring responses to drifting sinusoidal gratings, SF tunings of monkey striate (V1) neurons were shown to depend on contrast (Sceniak et al., 2002). As contrast increased, the bandwidth of SF tuning curves increased, predominantly elevating the amplitude on their high-SF side, so that transmission of fine features is facilitated. Such contrast-dependent SF tunings were also observed in cat striate neuron data (Skottun et al., 1987).

To elucidate mechanisms that shape selectivity for a stimulus feature, it is generally useful to measure the precise time course of tunings in response to brief stimuli (Ringach et al., 1997). Using this method, previous studies have revealed that SF tunings of striate neurons are shaped by a variety of mechanisms. First, bandpass tuning curves are already observed at response onset, indicating that early thalamocortical feedforward processing contributes to the generation of SF tunings of striate neurons. Then, however, the optimal SF is often shifted toward high SF, along with a decrease in tuning width over time (Bredfeldt and Ringach, 2002; Mazer et al., 2002; Frazor et al., 2004; Nishimoto et al., 2005; Ninomiya et al., 2012; Purushothaman et al., 2014). This processing may be because of delayed feedforward excitatory processing (Frazor et al., 2004; Allen and Freeman, 2006) or intracortical inhibition (Bredfeldt and Ringach, 2002). Alternatively, a portion of neurons increases the tuning width over time (Tanaka and Ohzawa, 2020), possibly because of another mechanism for dynamically shaping RF size (Suder et al., 2002; Malone et al., 2007).

It remains to be elucidated which of the aforementioned mechanisms plays an important role in contrast-dependent modulation of SF tunings. To address this, we here measure the time courses of SF tunings of cat striate, or area 17, neurons under different contrasts, with a subspace reverse correlation analysis in which a rapidly flashed sequence of sinusoidal gratings of different SFs is presented. If differences in SF tunings between different contrasts are formed at response onset, the early feedforward processing may be sufficient to explain contrast-dependent SF tunings. Alternatively, contrast-dependent modulation may develop subsequently. In this case, determining which tuning dynamics depends on contrast is critical for specifying the mechanisms.

We found that the optimal SF shifted to a higher SF over time, with a larger shift for higher contrast. At low contrast, neurons generally decreased the bandwidth over time. At high contrast, we found a range of bandwidth dynamics that depended on the degree of the optimal SF shift. These dynamic effects always enhanced the processing of high SF range under high contrast. They could be explained by the combinations of two mechanisms: one that decreases bandwidth and shifts optimal SF, and another that increases bandwidth without shifting optimal SF.

Tuning dynamics have been shown for SF, orientation, size, and disparity tunings. They have been implicated in coarse-to-fine processing (Marr and Poggio, 1979; Menz and Freeman, 2003; Malone et al., 2007) or increase in feature selectivity (Ringach et al., 1997). However, the crucial functions of tuning dynamics remain controversial (Frazor et al., 2004). The steady-state tuning curves for these features, to some extent, depend on contrast (Sceniak et al., 1999; Alitto and Usrey, 2004). Our results indicate that one of the essential roles of tuning dynamics of striate neurons is to adjust tunings depending on contrast.

Materials and Methods

Surgery and unit recordings

All animal care and experimental protocols conformed to the guidelines established by the National Institute of Health and were approved by the Osaka University Animal Care and Use Committee. The experiments were conducted with anesthetized and paralyzed male cats. Details of the surgery and unit recordings were provided in our previous paper (Tanaka and Ohzawa, 2020).

Two adult male cats (2.5–4.0 kg) were used. Initial surgery for tracheotomy and vascular catheterization was performed under anesthesia with 2%-3.5% isoflurane. Anesthesia was then switched to thiopental sodium (Ravonal, 1.0–1.5 mg·kg−1·h−1, dissolved in a Ringer's solution, 1 ml·kg−1·h−1) to maintain anesthesia for the rest of the surgery and subsequent recording sessions. Paralysis was induced and maintained with gallamine triethiodide (Flaxedil, 10 mg·kg−1·h−1, initial dose 10–20 mg, continuous infusion). Artificial respiration was maintained with a mixture of nitrous oxide (70%) and oxygen at 20-30 strokes/min. The respiration rate and stroke volume were adjusted to maintain an end-tidal CO2 between 3.5% and 4.3%. A rectangular hole was made in the skull over the representation of area 17, and the dura was dissected to allow for penetration by the microelectrode array. Pupils were dilated with atropine sulfate (1%), and the nictitating membranes were retracted with phenylephrine hydrochloride (Neosynesin, 5%). Contact lenses of appropriate power with 4 mm artificial pupils were placed over the corneas.

We used four-shaft silicon microelectrode arrays with a total of 32 recording probes (A4 × 8_200_400_177, NeuroNexus) to record neuronal activity in area 17. Electrode signals were amplified (PBX2, gain 1000, Plexon), bandpass filtered (0.1–3 kHz), and fed into a custom-made data acquisition system, where they were A-D converted at a sampling rate of 20 kHz and saved as data files. Accurate spike sorting using a commercial software package (Offline Sorter, Plexon) and data analyses were then conducted offline with these data. In parallel, the filtered signals were processed online with reduced accuracy with a custom-built spike sorter to monitor neuronal responses to visual stimuli in real time. Details of offline spike sorting were provided in our previous study (Tanaka and Ohzawa, 2020). Herein, we briefly describe this method. Spike segments that exceeded a threshold were extracted and plotted as data points in the PC1−PC2 space of the principal component analysis. The threshold was set at 2.5-5 SD of the raw signal amplitude. If a subset of the data points appeared to form an isolated cluster, we manually drew a contour enclosing these points. Templates of the spike waves for the clusters were then calculated by averaging all the spike segments inside the contours. Next, we performed a template-matching procedure to resort the spike segments into the clusters. Clusters were taken as the activity of single units if their averaged spike amplitude was >3 times the background noise level.

Visual stimulation

Visual stimuli were displayed on a color CRT monitor (76 Hz, 1600 × 1024 pixels, 46.6 cm wide × 29.9 cm high, 47 cd·m−2, GDM-FW900, Sony). In each recording session, the luminance nonlinearity of the monitor was measured with a photometer (Minolta CS-100, Konica Minolta Photograph Imaging) and linearized using γ-correction look-up tables. Cats viewed the monitor screen through a custom-built haploscope, which allowed visual stimuli to be presented to the left and right eyes separately. A black separator was placed between the left and right visual fields to prevent stimulation of the unintended eye. The distance between the monitor screen and the eyes was set at 57 cm, subtending a visual field of 23.3° × 29.9° for each eye.

In total, we conducted three penetrations of the microelectrode arrays in two cats (two penetrations in one cat and one penetration in the other cat). After penetration, we adjusted the electrode depth to enable the recording from as many probes as possible. We waited ∼30 min before starting the recording sessions because the amplitude of the neural signals sometimes changed considerably shortly after electrode penetration. We first determined the approximate position, size, and basic tuning parameters of the classical RFs of one or two cells on each of the four electrode shafts, using small bars or small circular patches of sinusoidal grating under manual mouse control.

Next, we conducted the main recording session to accurately determine the time courses of orientation and SF tuning properties of all recorded neurons under a variety of contrasts, using a subspace reverse correlation technique (Fig. 1A) (Ringach et al., 1997). We repetitively presented rapid random sequences of sinusoidal gratings with circular patches that typically included 13 spatial frequencies, 18 orientations (0°–180° in 10° steps), four phases (0°, 90°, 180°, and 270°), and one blank stimulus. The contrast of the sequence was kept constant for each presentation of the sequence, which typically lasted for 37 s. The same stimulus sequences at three or four contrast levels between 3% and 50% were presented 10-15 times in an interleaved manner within a single recording session. We conducted three sessions in this study (one session per one electrode penetration) with different contrasts used for different sessions (first session: 6.25%, 12.5%, 25%, 50%; second session: 6.25%, 12.5%, 50%; third session: 3.125%, 6.25%, 12.5%, 50%). The flash duration of each grating in the sequence was set at 26.3 ms (two video frames); but when sufficient responses were not observed from some online-monitored cells, we reset the duration to 39.5 ms (three video frames) and restarted the recording session from the beginning. Therefore, the flash duration was always the same for the sequences of all contrast levels within a session. Two video frames were used in one session and three video frames were used in two sessions.

The size of the grating patch was adjusted to cover the RFs of all recorded cells. The range of stimulus SFs was adjusted to cover the entire frequency range that evoked responses from all recorded cells, and the distribution of the stimulus SFs was regular on a logarithmic scale (ranges: 0.1-2.0 or 2.5 cycles per degree [cpd]). We presented the stimuli monocularly. Therefore, the dominant eye was not always chosen for all cells. However, the majority of area 17 cells produce at least some response to each eye, and previous evidence indicates that SF tuning dynamics of area 17 cells are generally consistent between the two eyes (Ninomiya et al., 2012).

Data analysis

The evoked responses and the stimulus sequence of each contrast level were cross-correlated to obtain two-dimensional SF and orientation selectivity maps (SFOR maps, Fig. 1B) at correlation delays from 0 to 150 ms in 5 ms steps (phase was not discriminated). One-dimensional SF tuning curves were then calculated by integrating each time slice of the SFOR maps over a ±10° range of peak orientations. The SF tuning curves were further smoothed with a one-dimensional uniform filter (width 3) for noise reduction. We thus obtained the time course of SF tuning curves for each contrast condition (Fig. 1C). For a subset of neurons, we computed SF tuning curves at a suboptimal orientation, by integrating SFOR maps between 10° and 30° off the peak orientation.

The SF tuning curve was then fitted with the following empirical function (mixture of bandpass Gaussian and low-pass Gaussian) on a logarithmic scale as follows: G(f)=a⋅e−(logf−µ)22σ12 + b⋅e−(logf)22σ22 + c, Where f is SF, μ, and σ1 are the center and SD of the bandpass Gaussian, respectively, and σ2 is the SD of the low-pass Gaussian; a, b, and c are the constants. The second term sometimes improves fit by capturing amplitude elevation confined to the lower frequency side, as clearly observed in the function fitting shown in Figure 2A, C. The function fitting was conducted using the Levenberg–Marquardt algorithm (MATLAB, lsqcurvefit function) in a program written in MATLAB (The MathWorks).

For each contrast condition, neurons were judged whether they had a reliable time course of SF tuning curves for the contrast, as follows: For each contrast, we first estimated the noise level by computing the response variances of the SFOR maps for noncausal delays from −100 to 0 ms in 5 ms steps, and then determined consecutive time points between 0 and 150 ms at which the response variance exceeded the mean of the noise level by 5 SDs, and the function fitting of G(f) to tuning curves was successful (r2 >0.5). If the length of time points was ≥5 points (25 ms), the neuron was considered to have a reliable time course of SF tunings for that contrast condition.

The SF corresponding to the peak position of the fitted curve G(f) was taken as the optimal SF. The frequencies at half height of G(f) were defined as the low cutoff and high cutoff SFs. Here, the height was defined as the amplitude between the minimum and maximum values of the fitted curves G(f) on the range of stimulus SF. To extract the time courses of these parameters for each contrast condition, we defined peak time, Tpeak, as the delay to the greatest response amplitude at the optimal SF for the contrast condition, and development and decay times, Tdev and Tdec, as the delays in which responses first exceeded and subsequently fell below half of the response amplitude of the peak time, respectively. The time between Tdev and Tdec was defined as response duration. We computed the optimal, low cutoff, and high cutoff SFs for each time slice within the response duration. To quantify the tuning width, we calculated the full bandwidth, defined as a log-ratio (base 2) of high cutoff SF to low cutoff SF. For several analyses, we also used half bandwidth, defined as the log-ratio (base 2) of high cutoff SF to optimal SF, as described in the relevant sections. It is possible that the smoothing of the data before the function fitting caused our estimation of bandwidth to be larger than the true value. However, this effect was likely limited because the average full bandwidth obtained in this study (1.62 octaves at peak delay) was similar to that reported in previous studies (e.g., 1.57 octaves for complex cells) (Movshon et al., 1978). The effect was common to different contrasts and different delays; therefore, the differences in bandwidth between contrasts or delays are unlikely to be an artifact of the filtering.

To quantify the S/N ratio of the SF tuning curves, we computed a ratio of the height (defined above) of the fitted function to the square root of the residual variance as follows: 1N−n∑k=1N(R(fk)−G(fk))2,where fk, N, R(fk), and n are the SF at the kth sampling point, the total number of SF points, the response amplitude at fk, and the number of fit parameters, respectively.

To quantify the degree of dynamic change in the optimal SF, we calculated the SF shift as follows: SF shift=log2SFdecSFdev(octave) where SFdev and SFdec are the optimal SFs at Tdev and Tdec, respectively. To compute the half-width of a 95% CI of the SF shift, we calculated a square root of the sum of the squares of that for log2SFdev and log2SFdec. For this calculation, we could not use the CI values of the fit parameters of G(f)because the optimal SF is not simply related to one of these parameters. Therefore, to specifically measure CIs alone, we additionally fitted the SF tuning curve with a single bandpass-Gaussian function (G(f)without the second term), whose center parameter corresponds to the peak position of the fitted function (=optimal SF). We then obtained the CIs for the logarithms of the optimal SFs (the center parameter) using MALAB curvefit functions (lsqcurvefit and nlparci functions) and computed the CI of the SF shift. This procedure may be acceptable because the SF shift values obtained by the two fitting procedures were highly similar, with mean absolute differences of 0.037 octaves (SD: 0.045) and 0.049 octaves (SD: 0.094) for low- and high-contrast conditions, respectively (n = 59).

To place cells along a simple/complex cell continuum using responses obtained with a reverse correlation method, we used modulation index (MI) (Nishimoto et al., 2005; Ringach and Malone, 2007) defined as follows: MI=2×|R0−R180| + |R90−R270|R0+R90 + R180+R270 where Rθ is the peak-time response to the flashed grating of phase θ at the optimal orientation and SF. This index and a more conventional index (F1/F0 ratio), computed from responses to drifting grating (Skottun et al., 1991), are highly correlated (Nishimoto et al., 2005).

Experimental design and statistical analysis

The key values of the derived measures are often represented as mean ± SD. To compare the means of derived measures between different response times and contrasts, we used paired t test or two-sample t test (functions ttest or ttest2), and to compare the variance, we used two-sample F test (functions vartest2) using MATLAB. We provided n, p value, and mean ± SD for these tests in relevant portions of the text. Pearson correlation coefficient was represented as r.

Histology and reconstruction of the recording site

For each penetration, we passed currents of 7-10 µA for 10 s at several selected sites for histologic examination after the recordings for that penetration. We then retracted the electrode and, for one cat, reinserted the electrode at another cortical location for another recording session. After all recordings were completed, the animals were then administered an overdose of pentobarbital sodium (Nembutal or Somnopentyl, 100 mg/kg) and perfused transcardially with formalin (4% in buffered saline). The visual cortex was then sliced parasagittally at a thickness of 60 µm. Every other section was Nissl-stained with thionine and the remainder with cytochrome oxidase. From these sections, we assessed the laminar locations of the recording sites. Additional details were provided in our previous study (Tanaka et al., 2014).

Results

Cell population

Using four-shaft silicon microelectrode arrays with a total of 32 recording probes (A4 × 8_200_400_177, NeuroNexus), we conducted three recording sessions with different penetrations from cat area 17, in which we recorded the activities of neurons in response to a rapidly flashed, random sequence of sinusoidal gratings consisting of various SFs, orientations, and phases under several contrast conditions (Fig. 1A). The same stimulus sequences at three or four different contrast levels between 3% and 50% were presented 10-15 times in an interleaved manner within a single experimental session. The stimulus contrast was kept constant for each presentation of the stimulus sequence. We used different contrasts for different sessions (see Materials and Methods). Using the cross-correlation of the evoked response with the stimulus sequence, we calculated SFOR maps for the correlation delays from 0 to 150 ms in 5 ms steps for each contrast (Fig. 1B). Then, one-dimensional SF tuning curves were derived by integrating each time slice of the SFOR maps over a ±10° range of peak orientations and then smoothing it using a one-dimensional uniform filter (width 3). These curves were fitted with the following empirical function: G(f)=a⋅e−(logf−µ)22σ12 + b⋅e−(logf)22σ22 + c, Where f is SF, μ, and σ1 are the center and SD of the bandpass Gaussian, respectively, and σ2 is the SD of the low-pass Gaussian; a, b, and c are the constants (Fig. 1C).

Figure 1.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure 1.

An example of time courses of SF tuning curves at different contrasts, using subspace reverse correlation method. Data are obtained from an area 17 cell of a cat. A, The configuration of a single recording session of the present experiment. We recorded the activities of neurons in response to a sequence of sinusoidal gratings of different SFs, orientations, and phases under different contrasts. The same stimulus sequences at three or four different contrast levels between 3% and 50% were presented 10-15 times in an interleaved manner within a single experimental session. The contrast was kept constant for each presentation of a stimulus sequence that typically lasts 37 s. The flash duration of each grating in the sequence (26.3 or 39.5 ms) was the same for the sequences of all contrasts within the same recording session. B, SF and orientation maps reconstructed by reverse correlation for low (top row) and high (bottom row) contrast stimuli. Data at delays of 40, 60, and 80 ms are shown. Abscissa and ordinate indicate SF and orientation (OR), respectively. Brighter colors represent larger responses. C, Panels represent SF tuning curves (dotted lines) and fitted functions (solid lines) for delays between 30 and 100 ms in 5 ms steps. Fitted functions are only shown for delays between Tdev and Tdec. Left and right columns correspond to low and high contrast, respectively. Vertical axis of each graph indicates response amplitude (unit: the number of spikes per flash presentation). In panels at Tdev, Tpeak, and Tdec, black bars and circles at the right margin represent baseline and baseline + 3 SD, respectively, where baseline is the response level at the highest frequency and SD represents the standard deviation of the data points around the fitted curves. D, E, Time courses of optimal SF (D) and full bandwidth (E). Thick green line and the thin magenta line indicate low and high contrast as shown in the inset, respectively. Three dots on each line indicate data at Tdev, Tpeak, and Tdec, respectively. deg, Degree; oct, octave.

The contrasts commonly used in the three sessions were 6.25%, 12.5%, and 50%. We limited our analysis to the tuning curves for these contrasts. We recorded a total of 103 neurons producing ≥250 spikes for contrast 50% condition. Among these neurons, 59 neurons showed a reliable time course of SF tuning curves for both 50% contrast and lower contrasts (12.5% and/or 6.25%). For these neurons, we compared SF tunings at low- and high-contrast levels. The low-contrast level refers to the lowest contrast under which neurons showed a reliable time course of tuning curves (6.25% and 12.5% for 40 and 19 neurons, respectively). The high-contrast level was always 50%. These 59 neurons produced ≥250 spikes and exhibited tuning curves nicely fitted with G(f) at both the contrast levels. The mean r2 value for a fit during the response time across the entire dataset was 0.97±0.017 (n = 59). For the neurons shown in Figures 1, 2A, 2B, and 2C, r2 was 0.95, 0.97, 0.99, and 0.99, respectively.

Figure 2.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure 2.

SF tuning dynamics at low and high contrast for another three cells. A–C, Data for each neuron. First and second columns, Normalized SF tuning curves and fitted functions at Tdev, Tpeak, and Tdec are shown for low (first column) and high contrast (second column). Normalized responses 0 and 1 correspond to the minimum and maximum values of the fitted functions within the stimulus SF range, respectively. Horizontal line in each panel indicates a half-height level. Third column, Green and magenta trajectories graph temporal change of optimal SF for low and high contrast, respectively. Three dots on each line indicate data at Tdev, Tpeak, and Tdec. Fourth column, Trajectories show temporal change of full bandwidth for low and high contrasts.

The amplitude of the fitted tuning curves for lower contrast was generally smaller than that for high contrast (Fig. 1C). However, the tuning curves for both contrasts showed similarly high S/N ratios, ruling out the possibility that the differences in the tuning dynamics between contrasts were artificially caused by an insufficient S/N ratio of the tuning curve at low-contrast condition. For the neuron shown in Figure 1C, the amplitude of the fitted functions for the low contrast at Tdev, Tpeak, and Tdec was 13, 12, and 9.2 times, respectively, larger than the square root of the residual variance (for definition, see Materials and Methods) that represents the SD of the data points around the fitted curve, while the corresponding values for high contrast were 8.8, 7.8, and 9.8. Overall, the mean S/N ratios of the amplitude to the SD during response were 15 ± 6.5 and 19 ± 8.3 for low and high contrast, respectively (n = 59). Even tuning curves fitted to the raw response amplitude showed sufficient signals with the mean S/N ratios of 5.4 ± 2.0 and 7.3 ± 2.9 for low and high contrast, respectively (n = 59).

Temporal shift of optimal SF depends on contrast

To elucidate the mechanisms underlying the contrast dependence of SF tunings, we assessed how and when differences in tuning arise during a response. Several previous studies reported that SF tuning curves were dynamically shifted toward a high SF end. Even if differences in SF tunings between the contrasts are not present at response onset, differences may arise as a result of this shift, depending on the contrast, which is what we examined, first. Figure 1 illustrates an example from one neuron in area 17. The two columns in Figure 1C show temporal sequences of SF tuning curves (dotted lines) for the delays between 30 and 100 ms in 5 ms steps for low (left column) and high contrast (right column). Fitted functions are also depicted with solid lines for delays between development and decay times (Tdev and Tdec), the times in which responses first exceeded and then fell below half the height of the maximum response, respectively. In this study, the period between Tdev and Tdec was defined as the response time. Inspection of the SF tuning curves indicates that this neuron increases the optimal SF, an SF at the peak position of the fitted function G(f), during a response even in the low-contrast condition. To quantitatively measure these temporal shifts of SF tunings and compare them between the contrasts, in Figure 1D, optimal SFs are plotted against delay time with green and magenta lines for low- and high-contrast conditions, respectively. As indicated by both curves rising to the right, the optimal SF continuously increased over time for both contrast conditions. However, the degree of shift was much larger for high contrast. At high contrast, the optimal SF of this neuron increased from 0.48 cpd (cycles per degree) to 1.0 cpd in the response time. At low contrast, the optimal SF increased from 0.48 to 0.67 cpd. To quantify the degree of temporal shift, we calculated the SF shift as follows: SF shift=log2SFdecSFdev(octave) where SFdev and SFdec are the optimal SFs at Tdev and Tdec, respectively. Positive values indicate that the optimal SF increases over time. The SF shift increased with contrast: 0.47 and 1.0 octaves for low and high contrasts, respectively.

We found that neurons typically exhibited a positive SF shift over a wide range of contrasts, but the degree of shift was highly dependent on contrast, as is the neuron illustrated in Figure 1. Another such example is shown in Figure 2A. Normalized tuning curves at Tdev, Tpeak, and Tdec for low and high contrast are shown in the left-hand portion, and the time courses of optimal SFs during the response time for each contrast are depicted in the third column from the left. While this neuron exhibited a very small SF shift of 0.10 octaves for low contrast, it showed a large SF shift of 0.90 octaves for high contrast. It should also be noted that not all neurons showed a large SF shift, even at high contrast. A considerable proportion of neurons exhibited a modest shift like the neuron shown in Figure 2B (0.52 and 0.15 octaves for high and low contrast, respectively) or a small shift like the neuron depicted in Figure 2C (0.29 and 0.10 octaves for high and low contrast, respectively).

Figure 3 presents a population summary of the temporal shift of the optimal SF at different contrasts. For the 59 neurons analyzed in this study, the distribution of optimal SF at peak delay (SFpeak) at high contrast ranges from 0.3 to 1.5 cpd with a median value of 0.63 cpd (Fig. 3A), a distribution that is similar to those of previous studies (Movshon et al., 1978; Nishimoto et al., 2005). Figure 3B displays the distributions of the SF shift for the two contrast conditions. The distribution for low contrast (top panel) is shifted to the positive end (p < 10−4, t test, n = 59). Therefore, a general tendency of the temporal shift of SF tuning curves toward a high SF end is observed from a considerably low-contrast level. However, the amount of shift highly depends on the contrast, with a much larger shift for higher contrast. The mean values of SF shift are 0.21 ± 0.38 (mean ± SD) octaves at low contrast and 0.39 ± 0.37 octaves at high contrast (p < 10−4, paired t test, n = 59). It should also be noted that there is considerable variability across neurons in terms of the amount of SF shift from −0.5 to 1.5, consistent with previous studies (Bredfeldt and Ringach, 2002; Nishimoto et al., 2005; Tanaka and Ohzawa, 2020). Even at a high contrast of 50%, 34% of the neurons (20 of 59) exhibited a small SF shift of < 0.2.

Figure 3.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure 3.

Population summary of the dynamic shift of the SF tuning curves (n = 59). A, Distribution of optimal SF at peak delay (SFpeak) at high contrast. B, The distribution of SF shifts at low and high contrasts are shown in the top and bottom rows, respectively. Triangles at the top portion represent the mean values of the distributions (0.21 and 0.39 octaves for low and high contrast). The half-width of the 95% CI of SF shift of individual neurons was, on average, 0.22 ± 0.11 and 0.17 ± 0.075 octaves for low and high contrast (n = 59). C, Top, SF shifts computed from SF tuning curves measured at optimal orientation (OR) under high contrast are plotted against those at suboptimal orientation (20 degrees apart from peak orientation) under high contrast. We here analyzed 35 neurons according to the criteria described in the text. Bottom, SF shifts are compared between two contrast conditions for the 35 neurons. D, The population average optimal SFs at Tdev, Tpeak, and Tdec, that is, SFdev, SFpeak, and SFdec, respectively, are plotted against the population average values of Tdev, Tpeak, and Tdec (values relative to Tdev) for each contrast condition. Green squares and magenta circles represent data for low and high contrast, respectively. Before taking the average across neurons, optimal SF values were normalized by SFdev at low-contrast condition for each neuron. E–G, The optimal SF for high contrast of each neuron is plotted against that for low contrast at Tdev (E), Tpeak (F), and Tdec (G). oct, Octave.

We examined whether the smaller SF shifts for low contrast arise from smaller responses at this contrast, rather than the low-contrast stimuli themselves. We calculated an SF shift for the SF tuning curves for high-contrast gratings whose orientation was 20° apart from the optimal orientation and compared it with that for the optimal orientation (Fig. 3C, top). Here, we analyzed 35 neurons in which the amplitude of SF tuning curves (at optimal delay) for the suboptimal orientation was between 20% and 60% of that for the optimal orientation (mean 43%). There was no tendency for a larger SF shift for tuning curves with a larger response (optimal orientation) (p = 0.82, paired t test, n = 35). Next, in Figure 3C, bottom, we compared the SF shifts of these neurons between the two contrast conditions as before (using SF tuning curves for the optimal orientation). The amplitude of the low-contrast SF tuning curves was, on average, 38% of that of the high-contrast tuning curves. There was a tendency for a larger SF shift for the high-contrast condition (p = 0.003). These results indicate that the differences in SF shift between the two contrast conditions cannot be explained by the differences in the magnitude of response.

To further characterize the temporal shift of SF tuning curves, for each contrast, we computed mean optimal SFs at three delay times (Tdev, Tpeak, Tdec) on average across the whole dataset (n = 59). Before taking the average, for each neuron, the optimal SF values at these delays were normalized by SFdev (optimal SF at Tdev) at low-contrast condition. The averaged and normalized optimal SFs are plotted against the population-average values of Tdev,Tpeak, and Tdec in Figure 3D (green squares and magenta circles for low and high contrasts, respectively), where the delay is indicated relative to Tdev (56 and 41 ms for low and high contrast). For both contrasts, the optimal SF continuously increased over time throughout the response time. The rate of optimal SF change for the early response period (Tdev ∼ Tpeak) was similar to that for the late period (Tpeak ∼ Tdec) for each contrast condition, although the increase rate differed considerably between the contrasts.

Figure 3D also demonstrates that the mean value of the optimal SF is invariant to the contrast at Tdev. Differences in the optimal SF between the contrasts grew with delays. To examine whether this was consistent across the data population, we compared SFdev, SFpeak, and SFdec at the two contrasts cell by cell (n = 59). SFdev, the optimal SF at Tdev, was consistently similar between the two contrasts (mean ± SD: 0.59 ± 0.20 cpd and 0.59 ± 0.22 cpd for low and high contrast, respectively; p = 0.94, paired t test, Fig. 3E). SFpeak was consistently slightly higher for high contrast (0.62 ± 0.22 cpd and 0.67 ± 0.25 cpd for low and high contrast, respectively; p < 10−4, Fig. 3F). SFdec was consistently higher for high contrast (0.69 ± 0.27 cpd and 0.79 ± 0.33 cpd for low and high contrast; p < 10−4, Fig. 3G). Therefore, a difference in the optimal SF was not present at response onset and developed later. We will subsequently describe that these temporal shifts allow the average SF tuning curves during a response to exhibit a larger amplitude at a high SF range for higher contrast than for lower contrast, explaining the contrast dependence of conventional SF tunings. Therefore, the shift process plays an essential role in the contrast-dependent modulation of SF tunings.

Finally, nearly 90% of neurons (52 of 59) had Tdev within 15 ms around the population average at high contrast (26-56 ms). There was no correlation between Tdev and the SF shift at high contrast (r = −0.01, p = 0.35, n = 59). This suggests that the mechanisms underlying the shifts of optimal SFs allowed neurons to initiate the optimal SF shift with this range of Tdev. As we describe in detail in the Discussion, previous studies indicate intracortical suppression as a key mechanism generating the shift of optimal SF (Bredfeldt and Ringach, 2002; Ninomiya et al., 2012). Consistent with the range of the onset times of the SF shift, Ninomiya et al. (2012, their Fig. 10) indicated that neurons receive inhibitory inputs with various latencies between 30 and 60 ms, with a slight delay from the excitatory inputs, which seems to be effective in initiating a shift in optimal SF.

Bandwidth dynamics and two different underlying mechanisms

Previous research using drifting gratings has demonstrated that the tuning width of steady-state SF tuning curves increased as the contrast increased, predominantly elevating the amplitude on the high-SF side (Sceniak et al., 2002). We here examine how bandwidth and high-cutoff SFs at each time slice changed depending on contrast.

We first illustrate simple models of tuning-curve dynamics that may contribute to the contrast-dependent modulation of conventional SF tuning curves (Fig. 4), taking the following observations into account. In the previous section, we showed that the degree of shift of optimal SF was diverse across neurons. Moreover, we previously observed that, at high contrast, neurons with a larger shift of optimal SF tend to show a larger bandwidth decrease (Tanaka and Ohzawa, 2020). Therefore, we illustrate a descriptive model of neurons with a large shift and small shift of optimal SF separately, depicting tuning curves of these models in Figure 4A and Figure 4B, respectively. For both model neurons, we assume that bandwidth is invariant to contrast at response onset (top row). In the neuron with a large optimal-SF shift (Fig. 4A), SF tuning curve at high contrast (magenta) becomes more narrowly tuned to a high-frequency region with its bandwidth decreasing over time. This allows the neuron to selectively enhance processing at a high-frequency range at high contrast. In the neuron with a small optimal-SF shift (Fig. 4B), the bandwidth of tuning curve at high contrast may increase over time to sufficiently elevate high-cutoff frequency (magenta). At low contrast, where the shift in optimal SF is, on average, half of that at high contrast (Fig. 3B), the bandwidth may decrease over time to become more narrowly tuned to a low-frequency region (green, Fig. 4A,B). Based on these dynamics, for both the model neurons, the average SF tuning curves during a response (Fig. 4A,B, bottom), which approximate steady-state tuning curves (Nishimoto et al., 2005), would have a larger bandwidth and a higher high-cutoff SF for high contrast than for low contrast.

Figure 4.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure 4.

A, B, Tuning curves of hypothetical model neurons. The top three rows represent SF tuning curves at three different delays. Bottom panels, Average tuning curves during a response. Magenta and green represent high and low contrast. A, Panels represent tuning curves of a neuron exhibiting a large shift of optimal SF and a bandwidth decrease at high contrast. B, Panels represent tuning curves of a neuron with a small shift of optimal SF and a bandwidth increase at high contrast. The tuning curves for low contrast (green) are the same in A and B. For both models shown in A and B, the bandwidth of the temporally averaged SF tuning curves is larger at high contrast than at low contrast (bottom row). High cutoff SF is also higher at high contrast.

Hereafter, we will examine these possibilities. Currently, it is unknown how bandwidth dynamics depend on contrast. Further, the relationship between bandwidth change and optimal-SF shift has not been characterized in detail. Consequently, it is unclear whether the correlation of optimal-SF shift with bandwidth dynamics could enhance high-frequency processing at high contrast and contribute to contrast modulation of SF tunings.

Figure 1E and the rightmost panels of Figure 2A–C show the time courses of the full bandwidth of the example neurons, where full bandwidth is defined as log2high cutoff SFlow cutoff SF . High cutoff SF and low cutoff SF are the half-height locations of the functions G(f) fitted to the data. For low contrast (green), the neuron in Figure 1 does not change the bandwidth significantly. However, all the neurons in Figure 2 decreased the bandwidth over time. As these examples indicate, we found that the bandwidth generally decreased over time at low contrast. For high contrast (magenta), the bandwidth of the neurons shown in Figures 1 and 2A, which showed a large shift of optimal SF, decreased monotonically over time. In contrast, the bandwidth of the neurons shown in Figure 2B, C, which showed a small shift of optimal SF, increased over time. These examples are consistent with the models shown in Figure 4.

Although the above results suggest that bandwidth dynamics depend on the amount of shift of optimal SF, we first examined the mean time course of bandwidth for each contrast on average across the whole dataset. For this analysis, we used half bandwidth (log-ratio of high-cutoff SF to optimal SF) as a measure of tuning width because, in 7 of 59 cells, at least 1 of 6 SF tuning curves (Tdev, Tpeak, Tdec× 2 contrasts) was lowpass-like so that full bandwidth could not always be defined. By using half bandwidth, we could define the bandwidth of all neurons selected for the present analysis (n = 59). We denoted the half bandwidth at Tdev, Tpeak, and Tdec as BWdev, BWpeak, and BWdec, respectively. In Figure 5A, for each contrast, the population-average BWdev, BWpeak, and BWdec are plotted against the population-average values of Tdev, Tpeak, and Tdec. At low contrast (open squares), the bandwidth appeared to decrease monotonically (mean ± SD, BWdev: 0.82 ± 0.20, BWpeak: 0.78 ± 0.16, BWdec : 0.74 ± 0.19). At high contrast (solid circles), the average bandwidth showed little change (BWdev: 0.80 ± 0.19, BWpeak: 0.81 ± 0.16, BWdec : 0.79 ± 0.23). Similar results were obtained when we analyzed the full bandwidth for a smaller population in which full bandwidth was completely measured (mean for Tdev, Tpeak, Tdec: 1.63, 1.52, 1.54 for low contrast; 1.58, 1.62, 1.60 for high contrast; n = 49, data not shown).

Figure 5.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure 5.

Population summary for bandwidth dynamics at low and high contrast (n = 59). A, Population average BWdev, BWpeak, and BWdec (half bandwidth at Tdev, Tpeak, and Tdec) are plotted against population average values of these delays. Open squares and solid circles represent data for low and high contrast, respectively. B, C, (BWdec−BWdev) of each cell at low (B) and high (C) contrast is plotted against the SF shift. D, Differences in BWdev between contrasts (high contrast – low contrast) are plotted against the SF shift. E, A corresponding plot for BWdec. F, Differences in high-cutoff SF between contrasts (high contrast – low contrast) at Tdev are plotted against the SF shift. G, A corresponding plot for high cutoff SF at Tdec. B–G, The SF shift plotted on the horizontal axis is always the one for a high-contrast condition. Histograms for data along the vertical axis are shown on the right portion. Triangles represent the mean value of the histograms. B–E, Regression lines based on a least-square method are shown in the scatter plots. oct, Octave; # of cells, number of cells.

We next analyzed the bandwidth dynamics cell by cell for each contrast condition, considering its dependency on the optimal SF shift. The bandwidth difference (BWdec−BWdev) at low contrast is plotted against the SF shift in Figure 5B. The majority of data points are in the lower half region, resulting in a mean (BWdec−BWdev) significantly< 0 (see right histogram, −0.08 ± 0.25, p = 0.017, paired t test, n = 59). This indicates that bandwidth generally decreases over time at low contrast.

Figure 5C shows the bandwidth difference (BWdec−BWdev) at high contrast plotted against the SF shift. Although we found no overall differences between BWdev and BWdec (right histogram, −0.01 ± 0.28, p = 0.74, paired t test), there was a negative correlation between the SF shift and (BWdec−BWdev) (r = −0.71, p < 10−4). Such a correlation was not observed in the low-contrast condition (r = −0.01, p = 0.88, Fig. 5B). At the high-contrast condition (Fig. 5C), based on a least-square regression line fitted to the data, the estimated (BWdec−BWdev) at an SF shift of 0 was significantly >0 (0.2, t = 5.18, p < 0.001, t test, df = 56), indicating that bandwidth tends to increase over time for the neurons exhibiting a small SF shift. On the other hand, the estimated bandwidth difference at an SF shift of 0.8 was significantly< 0 (−0.22, t = −5.9, p < 0.001), showing that neurons with a large SF shift tended to decrease bandwidth over time.

These results are generally consistent with the hypothesis described in Figure 4, in which there are two types of neurons: some showing bandwidth decreases and a large optimal-SF shift, and others showing bandwidth increases and a negligible optimal-SF shift. However, neurons in Figure 5C were not discontinuously segregated into these two types but formed a continuum in which the two neurons illustrated in Figure 4 represent two extremes. In the middle portion, this continuum contains neurons with an intermediate SF shift (0.3-0.5 octaves) and a relatively small bandwidth change. Like the extreme neurons, these neurons are also expected to have a larger high-cutoff SF for higher contrast because they should exhibit a larger shift of tuning curves for higher contrast while retaining their bandwidth.

These dynamic effects may be parsimoniously explained by two underlying mechanisms as follows. One mechanism causes a bandwidth decrease and a positive shift of tuning curves. Another mechanism increases bandwidth at high contrast without a shift of optimal SF. The continuous distribution of dynamic effects across neurons indicates that these two mechanisms do not affect distinct groups of neurons separately, but rather affect a considerable portion of neurons in an overlapping manner. As a result, the neurons equally affected by the two mechanisms show an intermediate shift of optimal SF and little bandwidth change because the effects of the two mechanisms on bandwidth cancel each other out. Various relative effects of the two mechanisms essentially explain the full range of the relationship between optimal-SF shift and bandwidth changes in the data population. We will consider the cellular basis of these mechanisms in the Discussion.

We next analyzed how bandwidth difference between contrast arises dynamically cell by cell, taking the types of SF shifts into account. In Figure 5D, differences in BWdev between the two contrasts (high contrast – low contrast) are plotted against the SF shift (at high contrast). The mean difference in BWdev was −0.02 (±0.18), a value not significantly different from 0 (right histogram, p = 0.35, paired t test, n = 59). There was no significant correlation between the bandwidth difference and the SF shift (r = 0.16, p = 0.22). A corresponding plot for BWdec is shown in Figure 5E. The mean difference in BWdec between the contrast was 0.047 ± 0.29, a value not significantly different from 0 (right histogram, p = 0.22). However, the bandwidth difference was negatively correlated with the SF shift (r = −0.56, p < 10−3, Fig. 5E). The regression line fitted to the data indicated that the estimated bandwidth difference at SF shift of 0 was significantly >0 (0.22, t = −5.6, p < 10−3, t test, df = 56). These results demonstrate that, mainly for neurons with a small SF shift, the bandwidth at high contrast later became larger than that at low contrast, consistent with the models shown in the third row of Figure 4 (delay 3).

We then examined whether the diverse dynamic effects always enhance high SF processing at high contrast. As shown in Figure 5F, the high-cutoff SF at Tdev was not significantly different between the contrasts (right histogram, mean difference –0.026, p = 0.40, paired t test, 1.04 ± 0.32 and 1.01 ± 0.34 cpd for low and high contrast). In contrast, the high cutoff SF at Tdec is, for the overall population, statistically higher for the high contrast than low contrast (Fig. 5G, right histogram, mean difference 0.17, p < 10−4, 1.17 ± 0.44 and 1.33 ± 0.47 cpd for low and high contrast). Moreover, this was true for neurons with an SF shift of < 0.2 (p = 0.013, n = 20, paired t test), for those with an SF shift of 0.2-0.6 (p = 0.014, n = 26), and for those with an SF shift > 0.6 (p<10−4, n = 13). Therefore, neurons with various dynamics always increased a high-frequency processing range as the contrast increased.

Together, the results presented in the current and previous sections indicate that there is a range of dynamic effects whose extreme behaviors are depicted in Figure 4, and which always dynamically enhance high-frequency processing at high contrast. They may mediate the contrast-dependent modulation of the steady-state SF tunings. We further examine this issue in the next section.

How dynamic effects contribute to contrast-dependent modulation of steady-state SF tuning

So far, we demonstrated the dynamic effects that modulate the SF tuning curves depending on contrast, enhancing high-frequency processing at high contrast. The dynamic effects likely contribute to the contrast-dependent modulation of steady-state SF tuning curves observed for drifting gratings. However, to what degree do these effects contribute? To examine this, we calculated predicted steady-state SF tuning curves by summing the sequence of tuning curves in all the time slices during the response time (Tdev ∼ Tdec). Such temporally summed tuning curves of data from a reverse correlation method correlate well with steady-state tuning curves in response to drifting gratings (Nishimoto et al., 2005; Xing et al., 2011). As was conducted for each time slice, we fitted an empirical function G(f) to the summed tuning curves (r2 = 0.98 ± 0.021 and 0.98 ± 0.016 for low and high contrast, mean ± SD, n = 59). We quantitatively compared the summed SF tuning curves between low and high contrast and examined whether the difference was comparable to those reported for drifting gratings. Figure 6A shows the data for the neuron shown in Figure 1 (r2 for G(f) fitting: 0.98 and 0.94 for low and high contrasts). The top and bottom panels depict the summed tuning curves for low and high contrast, respectively. The bandwidth for high contrast (2.1 octaves) was larger than that for low contrast (1.8 octaves). In addition, the optimal SF (black triangles) and high cutoff SF (white triangles) for high contrast (0.67 and 1.3 cpd) were higher than those for low contrast (0.57 and 1.0 cpd). Low cutoff SF (gray triangles) did not vary appreciably (0.31 and 0.30 cpd for low and high contrast). Such differences between the contrasts are in good agreement with those reported for drifting gratings, as mentioned above.

Figure 6.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure 6.

Tuning curves obtained by summing SF tuning curves in all the time slices between Tdev and Tdec (temporally summed SF tuning curves) show contrast-dependent modulation consistent with that obtained using drifting grating stimuli. A, Data of one example neuron (the same neuron as that shown in Fig. 1). Data for low and high contrast are shown in the top and bottom rows. Black, gray, and white triangles represent optimal, low cutoff, and high cutoff SFs, respectively. B–F, Comparison of temporally summed SF tuning curves between the contrasts (n = 59). B, Half bandwidth of temporally summed SF tuning curves of each neuron is compared between low and high contrasts. C–F, Similar comparisons for full bandwidth (C), optimal SF (D), high cutoff SF (E), and low cutoff SF (F), respectively.

For the population summary, we compared the bandwidth, optimal SF, high cutoff frequency, and low cutoff frequency of the summed tuning curves between low and high contrast. As shown in Figure 6B, half bandwidth was, on average, 5% larger for high contrast (mean ± SD: 0.78 ± 0.15 and 0.81 ± 0.17 octaves for low and high contrast, p = 0.033, paired t test, n = 59). We also compared the full bandwidth between contrasts. For the summed tuning curves, full bandwidth at both low and high contrasts was assessed for all tested neurons (Fig. 6C). Full bandwidth was on average 7% larger for high contrast (1.58 ± 0.31 and 1.67 ± 0.37 octaves for low and high contrast, p = 0.009, paired t test, n = 59). Because there were no systematic differences in the distribution of BWdev between the two contrasts (Fig. 5D), the general tendency for larger bandwidth of summed tuning curves for high contrast across the data population should be because of the dynamic effects during a response described so far.

The optimal SF of the summed tuning curves was, on average, 7% larger for high contrast (0.62 ± 0.22 and 0.67 ± 0.25 cpd for low and high contrast, p < 10−4, paired t test, Fig. 6D). Since there were no systematic differences in SFdev between the contrast (Fig. 3E), the larger optimal SF for the summed tuning curves at high contrast should be because of a larger dynamic shift of tuning curves at this contrast. Actually, the SF shift values at high contrast were highly correlated with the difference in the optimal SF of the summed tuning curves between the contrasts (r = 0.37, p = 0.0035).

In the last section, we showed that the high cutoff SF of the tuning curve in the time slice at Tdev did not differ significantly between the contrasts (Fig. 5F), while that at Tdec was consistently higher at high contrast for a wide range of SF shifts (Fig. 5G). With such effects of dynamics, in Figure 4, we expected that the high cutoff SF of the summed tuning curves is higher for higher contrast. Indeed, this holds true with a high cutoff SF on average, 10% larger for higher contrast (1.06 ± 0.33 vs 1.16 ± 0.37 cpd for low and high contrast, p < 10−4, paired t test, Fig. 6E). The significant effects were separately observed for neurons with an SF shift of < 0.2 (p < 10−3), those with an SF shift of 0.2-0.6 (p = 0.021), and those with an SF shift of > 0.6 (p < 10−3), showing that various types of dynamic effects always enhance processing at a high SF range for higher contrast.

In Figure 6F, the low cutoff SF of the summed tuning curves at high contrast is plotted against that at low contrast. The difference in low cutoff SF between contrasts is on average 3% (p = 0.037, paired t test, 0.37 ± 0.15 and 0.38 ± 0.18 for low and high contrast), a value much smaller than the corresponding value for high cutoff SF. This result is consistent with the observation of Sceniak et al. (2002) that the difference in low cutoff SF between contrasts is much smaller than that of high cutoff SF.

Together, the predicted steady-state SF tuning curves obtained by summing the temporal sequences of tuning curves show contrast dependence, which is consistent with that found for drifting gratings. This indicates that the dynamic shifts described in the last sections essentially explain the contrast-dependent modulation of the conventional SF tuning curves.

The range of SF dynamics in relation to simple/complex classification and laminar structure

We found a continuous range of SF dynamics likely mediated by two different mechanisms. Herein, we address how these mechanisms work in relation to the known cortical cell types and laminar structure. Simple and complex cells are the two major cell classes in the striate cortex. We investigated whether there are any differences in the range of SF dynamics between these cell types. To place cells along the simple/complex continuum based on data obtained by reverse correlation analysis, we computed a modulation index (MI, see Materials and Methods), which nicely corresponds to the more standard index, F1/F0 ratio computed from responses to drifting grating stimuli (Nishimoto et al., 2005; Ringach and Malone, 2007). Using the commonly used criteria to dichotomize cells (0-1: complex cell; 1-2: simple cell), 10 and 49 cells were classified into simple and complex cells, respectively (17% and 83%).

We examined how SF dynamics depends on MI to assess the difference in tuning dynamics between simple and complex cells. We plotted the SF shift and half-bandwidth difference (BWdec−BWdev) at high-contrast condition against MI in Figure 7A and Figure 7B, respectively. We found that the entire range of SF shift and bandwidth difference observed in the whole dataset was covered by neurons with MI < 0.5 alone (n = 35). Therefore, the complex cell population contains the full range of dynamic effects. On the other hand, neurons with MI > 1 or those with MI > 0.75 (n = 16) infrequently showed a large SF shift of > 0.7. In addition, the variance of SF shift for MI > 0.75 was significantly smaller than that for MI < 0.5 (0.054 vs 0.18, p = 0.018, two-sample F test). These results indicate that the effects of the mechanism shifting the optimal SF would be weaker for simple cells than for complex cells. The proportion of neurons with a bandwidth increase of > 0.2 octaves was 20% (2 of 10) for MI > 1, 25% (4 of 16) for MI > 0.75, 20% (7 of 35) for MI < 0.5, indicating the effects of the bandwidth-increase mechanism may be similar for simple and complex cells. Although these results indicate that the two mechanisms differentially affect simple and complex cells, both simple and complex cells enhance processing at high-frequency range at high contrast, as Figure 7C indicates that the difference in high-cutoff SF between contrasts (high contrast – low contrast) at Tdec was >0 for not only neurons with MI < 0.5 (p < 10−4, t test), but also for those with MI > 0.75 (p = 0.028) or neurons with MI > 1 (p = 0.029) .

Figure 7.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure 7.

Relationship between SF dynamics and simple/complex cell class (n = 59). MI is used to place cells on simple and complex continuum. Complex cells are placed toward left while simple cells are placed toward right. The SF shift at high contrast (A), (BWdec−BWdev), the differences in half bandwidth between Tdev and Tdec, at high contrast (B), and the differences in high-cutoff SF between contrasts (high contrast – low contrast) at Tdec (C) are shown on the vertical axis, respectively.

We next examined the effects of the two mechanisms of SF dynamics for each cortical layer. We previously demonstrated a wide range of SF dynamics for all six layers (Tanaka and Ohzawa, 2020). Here we focus on whether their generation can be explained by the interactions of the two mechanisms. Each panel of Figure 8 graphs the half-bandwidth change (BWdec−BWdev) plotted against the optimal SF shift under a high-contrast condition for each layer (n = 10, 13, 21, and 15 for layers 2/3, 4, 5, and 6, respectively). As we found for the entire data population (Fig. 5C), there was a significant or marginally significant negative correlation between the SF shift and the bandwidth change for each layer (r = −0.68, −0.70, −0.77, −0.49; p = 0.032, 0.007, 0.001, 0.063) for layers 2/3, 4 5, and 6, respectively). The SF shift of the neurons for layer 2/3 and layer 4 ranged from near 0 to ∼0.8, and these neurons showed positive and negative bandwidth changes. The results indicate that the two underlying mechanisms for SF dynamics form a range of SF dynamics as early as in the input layer, layer 4. The SF shift of the neurons of layers 5 and 6 ranged from ∼0 to 1.5, suggesting that the two mechanisms work ubiquitously across the six layers and form various types of dynamic effects there.

Figure 8.
  • Download figure
  • Open in new tab
  • Download powerpoint
Figure 8.

Each panel plots (BWdec−BWdev) the temporal difference in half bandwidth between Tdev and Tdec, against the SF shift under high-contrast condition for each layer. Regression lines were drawn based on a least-square method. A negative correlation between (BWdec−BWdev) and SF shift is observed for all layers. The number of cells for layers 2/3, 4, 5, and 6 is 10, 13, 21, and 15, respectively.

In addition, an F test revealed that the range (variance) of the SF shift in layer 5 was significantly larger than that in layer 4 (p < 0.01). We also previously showed with a larger data sample (n = 42 and 64 for layer 4 and layer 5) that the variance of SF shifts was significantly different between layers 4 and 5 (p < 0.03), but the range of SF shifts covered by neurons in layer 4 corresponded to ∼80% of that in layer 5 (Tanaka and Ohzawa, 2020). Therefore, a wide range of SF dynamics is already present as early as in layer 4, although additional enhancement of SF dynamics likely occurs outside layer 4.

Discussion

We measured the time course of SF tunings of cat area 17 neurons under different contrast conditions and found a range of dynamic effects that enhance the processing of high-frequency ranges under a high-contrast condition. These dynamic effects are likely to be mediated by combinations of two different mechanisms. Below, we compare our results with those of previous studies and discuss the circuit basis of the mechanisms and functional implications of the present results.

SF tuning dynamics explain the contrast dependence of SF tunings

Whereas neurons with bandwidth increases and decreases were observed in a similar frequency in this study, Bredfeldt and Ringach (2002) reported that, in monkeys, the majority of neurons decrease bandwidth over time. The average shift of optimal SF reported in their study was 0.62, ∼1.5 times larger than the current results. It may be possible that, in monkeys, the effect of the bandwidth-decrease mechanism is stronger than that of the bandwidth-increase mechanism.

We showed that the dynamic SF tuning effects generally explain the contrast-dependent tuning modulation for drifting gratings found in monkey V1 neurons (Sceniak et al., 2002). However, the differences in SF tuning curves between contrast for drifting grating were more prominent than those of the integrated tuning curves shown in the present study. For example, the average SF bandwidth of monkey V1 neurons increased by 124% with contrast, while the corresponding percentage is 107% in this study. Several factors may be responsible for this difference. First, while Sceniak et al. (2002) determined contrast levels of low- and high-contrast stimuli for each neuron so that they lie at sufficiently low- and high-sloping portions of the contrast response function, we set high contrast at 50% for all neurons. It is possible that, for some neurons, 50% contrast is not sufficient to evoke strong dynamics of SF tuning. Next, Sceniak et al. (2002) showed that a threshold effect partially explains bandwidth differences between the contrasts, which is not simply measurable in the present reverse-correlation analysis.

Nevertheless, we found that the contrast-dependent modulation of SF tunings for temporally integrated responses in reverse correlation qualitatively agrees well with that measured using drifting gratings. Furthermore, the contrast-dependent shift of tuning curves toward a high SF end cannot be explained by a threshold effect. Therefore, the mechanisms of SF dynamics we reported in this study essentially explain the contrast-dependent modulation of SF tuning curves as previously observed for drifting gratings.

Neural circuits of contrast-dependent SF tuning modulation

We indicated that the dynamic effects are mediated by two mechanisms: one that decreases bandwidth and shifts optimal SF, and the other that increases bandwidth without shifting optimal SF. The exact neuronal circuitry underlying these mechanisms remains unclear. We here describe a hypothesis regarding this, first for the bandwidth increase. Previous studies have shown that the size of the RF is smaller at higher contrast (Kapadia et al., 1999; Sceniak et al., 1999; Nienborg et al., 2013) and shrinks over time (Suder et al., 2002; Malone et al., 2007). An SF bandwidth increase during a response is likely related to this RF shrinkage because, for linear RF, the bandwidth of SF tuning is inversely related to the size of RF. While RF size is shaped by feedforward, lateral (horizontal), and feedback circuits (Nurminen et al., 2018), a marked reduction of lateral connectivity at high contrast has been demonstrated (Nauhaus et al., 2009). Earlier, Sceniak et al. (1999) suggested that the RF shrinkage at high contrast is because of this reduction, and further discussed that a decline of EPSP of lateral connections at high contrast is possibly caused by depression of excitatory synapses (Thomson and Deuchars, 1997) or increased shunting effects of postsynaptic neurons elicited by increased cortical activity at high contrast (Bernander et al., 1991). In both synaptic mechanisms, RF shrinkage could occur rapidly, likely accompanied by a dynamic bandwidth increase.

Previous evidence indicates that the shift of optimal SF and associated bandwidth decrease are mostly accompanied by tuned suppressive components of SF tuning curves, indicating that intracortical tuned inhibition plays an essential role in the bandwidth-decrease dynamics (Bredfeldt and Ringach, 2002; Ninomiya et al., 2012). Because the magnitude of intracortical inhibition depends on stimulus contrast, becoming dominant over excitation as contrast increases (Rubin et al., 2015; Adesnik, 2017), this tuned suppression may explain the larger optimal-SF shift for higher contrast. On the other hand, our results demonstrated that optimal-SF shift is more prominent in complex cells than in simple cells. This would be explained by a larger SF-tuned suppression for complex cells. Alternatively, this may indicate a different mechanism for the optimal-SF shift mediated by excitatory intracortical feedforward processing, where complex cells receive inputs from other cortical neurons tuned to a variety of SFs with different latencies. For this mechanism to mediate a contrast-dependent shift, high-frequency input should be less sensitive to contrast than low-frequency input. Further studies using optogenetic circuit manipulation (Atallah et al., 2012; Wilson et al., 2012) may help to determine which circuit mechanisms play more essential roles.

Different from the cortical origin of the SF dynamics described above, Allen and Freeman (2006) showed that neurons of cat LGN display a substantial temporal shift of optimal SF. Additionally, using drifting gratings, the high cutoff SF of LGN neurons generally increases with contrast (Nolt et al., 2004). Assuming that SF tuning dynamics of LGN neurons depend on contrast, the contrast-dependent modulation of high cutoff SF may be caused by SF tuning dynamics of LGN neurons. This may in turn contribute to the contrast-dependent SF dynamics of striate cells. However, this contribution is probably limited, given that the shift in optimal SF of cortical neurons was often shaped by tuned suppressive components. A larger optimal-SF shift of complex cells also suggests an essential contribution of intracortical process, indicating that subcortical contribution is limited. Together with the results of laminar analysis, this evidence indicates that intracortical processing, particularly within layer 4, likely plays an essential role.

A new view of the function of tuning dynamics

The contrast-dependent modulation of SF tuning may be generally understood in terms of optimal information transmission, together with the effect of RF size change (Sceniak et al., 1999). Under high contrast, neurons increase the spatial resolution by narrowing the RF, sacrificing sensitivity. For a higher spatial resolution, it is also reasonable to tune to higher SFs to better transmit information of fine details. Under low contrast, neurons need to increase sensitivity by using a larger RF size, sacrificing resolution. In such a case, it may be reasonable to be more tuned to lower SFs abundantly contained in features of a larger size.

It is generally considered that contrast-dependent response modulation occurs in two phases: a slow phase with a time scale of seconds (Ohzawa et al., 1982; Carandini and Ferster, 1997) and a fast phase with a time scale of <0.2 s (Victor, 1987; Baccus and Meister, 2002; Mante et al., 2005). Fast modulation has a functional sense because in daily life stimulus contrast within neurons' RF may change at every fixation, which occurs every 300 ms. Here, we did not directly measure the time required for the contrast-dependent modulation of SF tunings to work after changing the contrast level. However, our demonstration that this modulation is based on dynamic effects suggests that the contrast-dependent modulation of SF tuning is at least partly mediated by a fast adaptive process. For example, the degree of SF shift mediated by intracortical suppression could immediately reflect the amplitude of transient inhibitory inputs in response to recent stimuli, which increases with stimulus contrast. However, we cannot rule out a slow adaptive effect because these inhibitory inputs may become even larger with a slow time scale. To isolate a fast adaptive effect, a reverse correlation experiment simultaneously changing stimulus contrast and SF is required.

The dynamics of SF tunings, as well as disparity and size tunings, have been implicated in coarse-to-fine algorithms, the processing where coarse and fine information is analyzed sequentially (Menz and Freeman, 2003; Malone et al., 2007; Neri, 2011), but this remains controversial (Frazor et al., 2004; Mareschal et al., 2006). The present results demonstrate that the tuning dynamics contribute to the contrast-dependent modulation of SF tunings. Like SF, contrast modulation of size tuning was shown under a steady-state condition (Sceniak et al., 1999). We suggest that tuning dynamics play critical roles in contrast-dependent tuning modulation for these features.

In addition, tuning dynamics for orientation processing were shown for monkeys (Ringach et al., 1997). These dynamics have been viewed as a process for narrowing orientation tuning. Because orientation selectivity (measured in terms of circular variance) also depends on contrast to some degree under a steady-state condition (Alitto and Usrey, 2004), our results suggest that, rather than simply sharpening orientation tuning, the orientation dynamics adjust the tuning in a contrast-dependent manner. Overall, our findings implicate that the dynamic tuning process plays an essential role in contrast-dependent tuning refinement for various visual features in the early visual processing.

Footnotes

  • This work was supported by Ministry of Education, Culture, Sports, Science and Technology/Japan Society for the Promotion of Science KAKENHI Grants JP 16K01965, JP 15H05921, and 22K06437; and Kyoto Sangyo University Research Grant M2001. We thank I. Ohzawa for a great deal of support and valuable discussions; and K. Sasaki, M. Fukui, Y. Asada, T. Arai, T. Nakazono, R. Mizoguchi, D. Kato, K. Kurihara, M. Yoshida, M. Baba, and K. Kohara for assistance with the experiments and discussions.

  • The authors declare no competing financial interests.

  • Correspondence should be addressed to Hiroki Tanaka at rtanaka{at}cc.kyoto-su.ac.jp

SfN exclusive license.

References

  1. ↵
    1. Adesnik H
    (2017) Synaptic mechanisms of feature coding in the visual cortex of awake mice. Neuron 95:1147–1159. doi:10.1016/j.neuron.2017.08.014 pmid:28858618
    OpenUrlCrossRefPubMed
  2. ↵
    1. Alitto HJ,
    2. Usrey WM
    (2004) Influence of contrast on orientation and temporal frequency tuning in ferret primary visual cortex. J Neurophysiol 91:2797–2808. doi:10.1152/jn.00943.2003 pmid:14762157
    OpenUrlCrossRefPubMed
  3. ↵
    1. Allen EA,
    2. Freeman RD
    (2006) Dynamic spatial processing originates in early visual pathways. J Neurosci 26:11763–11774. doi:10.1523/JNEUROSCI.3297-06.2006 pmid:17093097
    OpenUrlAbstract/FREE Full Text
  4. ↵
    1. Atallah B,
    2. Bruns W,
    3. Carandini M,
    4. Scanziani M
    (2012) Parvalbumin-expressing interneurons linearly transform cortical responses to visual stimuli. Neuron 73:159–170. doi:10.1016/j.neuron.2011.12.013 pmid:22243754
    OpenUrlCrossRefPubMed
  5. ↵
    1. Baccus S,
    2. Meister M
    (2002) Fast and slow contrast adaptation in retinal circuitry. Neuron 36:909–919. doi:10.1016/S0896-6273(02)01050-4
    OpenUrlCrossRefPubMed
  6. ↵
    1. Bair W,
    2. Movshon JA
    (2004) Adaptive temporal integration of motion in direction-selective neurons in macaque visual cortex. J Neurosci 24:7305–7323. doi:10.1523/JNEUROSCI.0554-04.2004 pmid:15317857
    OpenUrlAbstract/FREE Full Text
  7. ↵
    1. Bernander O,
    2. Douglas RJ,
    3. Martin KA,
    4. Koch C
    (1991) Synaptic background activity influences spatiotemporal integration in single pyramidal cells. Proc Natl Acad Sci USA 88:11569–11573. doi:10.1073/pnas.88.24.11569 pmid:1763072
    OpenUrlAbstract/FREE Full Text
  8. ↵
    1. Bonhoeffer T,
    2. Kim DS,
    3. Malonek D,
    4. Shoham D,
    5. Grinvald A
    (1995) Optical imaging of the layout of functional domains in area 17 and across the area 17/18 border in cat visual cortex. Eur J Neurosci 7:1973–1988. doi:10.1111/j.1460-9568.1995.tb00720.x pmid:8528473
    OpenUrlCrossRefPubMed
  9. ↵
    1. Bredfeldt CE,
    2. Ringach DL
    (2002) Dynamics of spatial frequency tuning in macaque V1. J Neurosci 22:1976–1984. pmid:11880528
    OpenUrlAbstract/FREE Full Text
  10. ↵
    1. Campbell FW,
    2. Cooper GF,
    3. Enroth-Cugell C
    (1969) The spatial selectivity of the visual cells of the cat. J Physiol 203:223–235. doi:10.1113/jphysiol.1969.sp008861 pmid:5821877
    OpenUrlCrossRefPubMed
  11. ↵
    1. Carandini M,
    2. Ferster D
    (1997) A tonic hyperpolarization underlying contrast adaptation in cat visual cortex. Science 276:949–952. doi:10.1126/science.276.5314.949 pmid:9139658
    OpenUrlAbstract/FREE Full Text
  12. ↵
    1. De Valois RL,
    2. Albrecht DG,
    3. Thorell LG
    (1982) Spatial frequency selectivity of cells in macaque visual cortex. Vision Res 22:545–559. doi:10.1016/0042-6989(82)90113-4 pmid:7112954
    OpenUrlCrossRefPubMed
  13. ↵
    1. Frazor RA,
    2. Albrecht DG,
    3. Geisler WS,
    4. Crane AM
    (2004) Visual cortex neurons of monkeys and cats: temporal dynamics of the spatial frequency response function. J Neurophysiol 91:2607–2627. doi:10.1152/jn.00858.2003 pmid:14960559
    OpenUrlCrossRefPubMed
  14. ↵
    1. Issa NP,
    2. Trepel C,
    3. Stryker MP
    (2000) Spatial frequency maps in cat visual cortex. J Neurosci 20:8504–8514. pmid:11069958
    OpenUrlAbstract/FREE Full Text
  15. ↵
    1. Kapadia MK,
    2. Westheimer G,
    3. Gilbert CD
    (1999) Dynamics of spatial summation in primary visual cortex of alert monkeys. Proc Natl Acad Sci USA 96:12073–12078. doi:10.1073/pnas.96.21.12073 pmid:10518578
    OpenUrlAbstract/FREE Full Text
  16. ↵
    1. Malone BJ,
    2. Kumar VR,
    3. Ringach DL
    (2007) Dynamics of receptive field size in primary visual cortex. J Neurophysiol 97:407–414. doi:10.1152/jn.00830.2006 pmid:17021020
    OpenUrlCrossRefPubMed
  17. ↵
    1. Mante V,
    2. Frazor R,
    3. Bonin V,
    4. Geisler WS,
    5. Carandini M
    (2005) Independence of luminance and contrast in natural scenes and in the early visual system. Nat Neurosci 8:1690–1697. doi:10.1038/nn1556 pmid:16286933
    OpenUrlCrossRefPubMed
  18. ↵
    1. Mareschal I,
    2. Dakin SC,
    3. Bex PJ
    (2006) Dynamic properties of orientation discrimination assessed by using classification images. Proc Natl Acad Sci USA 103:5131–5136. doi:10.1073/pnas.0507259103 pmid:16549798
    OpenUrlAbstract/FREE Full Text
  19. ↵
    1. Marr D,
    2. Poggio T
    (1979) A computational theory of human stereo vision. Proc R Soc Lond B Biol Sci 204:301–328.
    OpenUrlCrossRefPubMed
  20. ↵
    1. Mazer J,
    2. Vinje WE,
    3. McDermott J,
    4. Schiller PH,
    5. Gallant JL
    (2002) Spatial frequency and orientation tuning dynamics in area V1. Proc Natl Acad Sci USA 99:1645–1650. doi:10.1073/pnas.022638499 pmid:11818532
    OpenUrlAbstract/FREE Full Text
  21. ↵
    1. Menz MD,
    2. Freeman RD
    (2003) Stereoscopic depth processing in the visual cortex: a coarse-to-fine mechanism. Nat Neurosci 6:59–65. doi:10.1038/nn986 pmid:12469131
    OpenUrlCrossRefPubMed
  22. ↵
    1. Movshon JA,
    2. Thompson ID,
    3. Tolhurst DJ
    (1978) Spatial and temporal contrast sensitivity of neurones in areas 17 and 18 of the cat's visual cortex. J Physiol 283:101–120. doi:10.1113/jphysiol.1978.sp012490 pmid:722570
    OpenUrlCrossRefPubMed
  23. ↵
    1. Nauhaus I,
    2. Busse L,
    3. Carandini M,
    4. Ringach DL
    (2009) Stimulus contrast modulates functional connectivity in visual cortex. Nat Neurosci 12:70–76. doi:10.1038/nn.2232 pmid:19029885
    OpenUrlCrossRefPubMed
  24. ↵
    1. Nauhaus I,
    2. Nielsen KJ,
    3. Disney AA,
    4. Callaway EM
    (2012) Orthogonal micro-organization of orientation and spatial frequency in primate primary visual cortex. Nat Neurosci 15:1683–1690. doi:10.1038/nn.3255 pmid:23143516
    OpenUrlCrossRefPubMed
  25. ↵
    1. Nauhaus I,
    2. Nielsen KJ,
    3. Callaway EM,
    4. Nauhaus I,
    5. Nielsen KJ,
    6. Callaway EM
    (2016) Efficient receptive field tiling in primate V1. Neuron 91:893–904. doi:10.1016/j.neuron.2016.07.015 pmid:27499086
    OpenUrlCrossRefPubMed
  26. ↵
    1. Neri P
    (2011) Coarse to fine dynamics of monocular and binocular processing in human pattern vision. Proc Natl Acad Sci USA 108:10726–10731. doi:10.1073/pnas.1101246108 pmid:21670301
    OpenUrlAbstract/FREE Full Text
  27. ↵
    1. Nienborg H,
    2. Hasenstaub A,
    3. Nauhaus I,
    4. Taniguchi H,
    5. Huang ZJ,
    6. Callaway EM
    (2013) Contrast dependence and differential contributions from somatostatin- and parvalbumin-expressing neurons to spatial integration in mouse V1. J Neurosci 33:11145–11154. doi:10.1523/JNEUROSCI.5320-12.2013 pmid:23825418
    OpenUrlAbstract/FREE Full Text
  28. ↵
    1. Ninomiya T,
    2. Sanada TM,
    3. Ohzawa I
    (2012) Contributions of excitation and suppression in shaping spatial frequency selectivity of V1 neurons as revealed by binocular measurements. J Neurophysiol 107:2220–2231. doi:10.1152/jn.00832.2010 pmid:22236707
    OpenUrlCrossRefPubMed
  29. ↵
    1. Nishimoto S,
    2. Arai M,
    3. Ohzawa I
    (2005) Accuracy of subspace mapping of spatiotemporal frequency domain visual receptive fields. J Neurophysiol 93:3524–3536. doi:10.1152/jn.01169.2004 pmid:15647396
    OpenUrlCrossRefPubMed
  30. ↵
    1. Nolt MJ,
    2. Kumbhani RD,
    3. Palmer LA
    (2004) Contrast-dependent spatial summation in the lateral geniculate nucleus and retina of the cat. J Neurophysiol 92:1708–1717. doi:10.1152/jn.00176.2004 pmid:15128751
    OpenUrlCrossRefPubMed
  31. ↵
    1. Nurminen L,
    2. Merlin S,
    3. Bijanzadeh M,
    4. Federer F,
    5. Angelucci A
    (2018) Top-down feedback controls spatial summation and response amplitude in primate visual cortex. Nat Commun 9:2281. doi:10.1038/s41467-018-04500-5 pmid:29892057
    OpenUrlCrossRefPubMed
  32. ↵
    1. Ohzawa I,
    2. Sclar G,
    3. Freeman RD
    (1982) Contrast gain control in the cat visual cortex. Nature 298:266–268. doi:10.1038/298266a0 pmid:7088176
    OpenUrlCrossRefPubMed
  33. ↵
    1. Purushothaman G,
    2. Chen X,
    3. Yampolsky D,
    4. Casagrande VA
    (2014) Neural mechanisms of coarse-to-fine discrimination in the visual cortex. J Neurophysiol 112:2822–2833. doi:10.1152/jn.00612.2013 pmid:25210162
    OpenUrlCrossRefPubMed
  34. ↵
    1. Ribot J,
    2. Aushana Y,
    3. Bui-Quoc E,
    4. Milleret C
    (2013) Organization and origin of spatial frequency maps in cat visual cortex. J Neurosci 33:13326–13343. doi:10.1523/JNEUROSCI.4040-12.2013 pmid:23946391
    OpenUrlAbstract/FREE Full Text
  35. ↵
    1. Ringach DL,
    2. Malone BJ
    (2007) The operating point of the cortex: neurons as large deviation detectors. J Neurosci 27:7673–7683. doi:10.1523/JNEUROSCI.1048-07.2007 pmid:17634362
    OpenUrlAbstract/FREE Full Text
  36. ↵
    1. Ringach DL,
    2. Hawken MJ,
    3. Shapley R
    (1997) Dynamics of orientation tuning in macaque primary visual cortex. Nature 387:281–284. doi:10.1038/387281a0 pmid:9153392
    OpenUrlCrossRefPubMed
  37. ↵
    1. Rubin DB,
    2. Van Hooser SD,
    3. Miller KD
    (2015) The stabilized supralinear network: a unifying circuit motif underlying multi-input integration in sensory cortex. Neuron 85:402–417. doi:10.1016/j.neuron.2014.12.026 pmid:25611511
    OpenUrlCrossRefPubMed
  38. ↵
    1. Sceniak MP,
    2. Hawken MJ,
    3. Shapley R
    (2002) Contrast-dependent changes in spatial frequency tuning of macaque V1 neurons: effects of a changing receptive field size. J Neurophysiol 88:1363–1373. doi:10.1152/jn.2002.88.3.1363 pmid:12205157
    OpenUrlCrossRefPubMed
  39. ↵
    1. Sceniak MP,
    2. Ringach DL,
    3. Hawken MJ,
    4. Shapley R
    (1999) Contrast's effect on spatial summation by macaque V1 neurons. Nat Neurosci 2:733–739. doi:10.1038/11197 pmid:10412063
    OpenUrlCrossRefPubMed
  40. ↵
    1. Shapley R,
    2. Victor JD
    (1979) Nonlinear spatial summation and the contrast gain control of cat retinal ganglion cells. J Physiol 290:141–161. doi:10.1113/jphysiol.1979.sp012765 pmid:469742
    OpenUrlCrossRefPubMed
  41. ↵
    1. Skottun BC,
    2. Bradley A,
    3. Sclar G,
    4. Ohzawa I,
    5. Freeman RD
    (1987) The effects of contrast on visual orientation and spatial frequency discrimination: a comparison of single cells and behavior. J Neurophysiol 57:773–786. doi:10.1152/jn.1987.57.3.773 pmid:3559701
    OpenUrlCrossRefPubMed
  42. ↵
    1. Skottun BC,
    2. De Valois RL,
    3. Grosof DH,
    4. Movshon JA,
    5. Albrecht DG,
    6. Bonds AB
    (1991) Classifying simple and complex cells on the basis of response modulation. Vision Res 31:1079–1086. doi:10.1016/0042-6989(91)90033-2 pmid:1909826
    OpenUrlCrossRefPubMed
  43. ↵
    1. Suder K,
    2. Funke K,
    3. Zhao Y,
    4. Kerscher N,
    5. Wennekers T,
    6. Wörgötter F
    (2002) Spatial dynamics of receptive fields in cat primary visual cortex related to the temporal structure of thalamocortical feedforward activity: experiments and models. Exp Brain Res 144:430–444. doi:10.1007/s00221-002-1061-5 pmid:12037629
    OpenUrlCrossRefPubMed
  44. ↵
    1. Tanaka H,
    2. Ohzawa I
    (2020) Local organization of spatial frequency tuning dynamics in the cat visual areas 17 and 18. J Neurophysiol 124:178–191. doi:10.1152/jn.00222.2019 pmid:32519574
    OpenUrlCrossRefPubMed
  45. ↵
    1. Tanaka H,
    2. Tamura H,
    3. Ohzawa I
    (2014) Spatial range and laminar structures of neuronal correlations in the cat primary visual cortex. J Neurophysiol 112:705–718. doi:10.1152/jn.00652.2013 pmid:25252337
    OpenUrlCrossRefPubMed
  46. ↵
    1. Thomson AM,
    2. Deuchars J
    (1997) Synaptic interactions in neocortical local circuits: dual intracellular recordings in vitro. Cereb Cortex 7:510–522. doi:10.1093/cercor/7.6.510 pmid:9276176
    OpenUrlCrossRefPubMed
  47. ↵
    1. Victor JD
    (1987) The dynamics of the cat retinal X cell centre. J Physiol 386:219–246. doi:10.1113/jphysiol.1987.sp016531 pmid:3681707
    OpenUrlCrossRefPubMed
  48. ↵
    1. Wilson NR,
    2. Runyan CA,
    3. Wang FL,
    4. Sur M
    (2012) Division and subtraction by distinct cortical inhibitory networks in vivo. Nature 488:343–348. doi:10.1038/nature11347 pmid:22878717
    OpenUrlCrossRefPubMed
  49. ↵
    1. Xing D,
    2. Ringach DL,
    3. Hawken MJ,
    4. Shapley RM
    (2011) Untuned suppression makes a major contribution to the enhancement of orientation selectivity in macaque V1. J Neurosci 31:15972–15982. doi:10.1523/JNEUROSCI.2245-11.2011
    OpenUrlAbstract/FREE Full Text
Back to top

In this issue

The Journal of Neuroscience: 42 (37)
Journal of Neuroscience
Vol. 42, Issue 37
14 Sep 2022
  • Table of Contents
  • Table of Contents (PDF)
  • About the Cover
  • Index by author
  • Masthead (PDF)
Email

Thank you for sharing this Journal of Neuroscience article.

NOTE: We request your email address only to inform the recipient that it was you who recommended this article, and that it is not junk mail. We do not retain these email addresses.

Enter multiple addresses on separate lines or separate them with commas.
Dynamics and Mechanisms of Contrast-Dependent Modulation of Spatial-Frequency Tuning in the Early Visual Cortex
(Your Name) has forwarded a page to you from Journal of Neuroscience
(Your Name) thought you would be interested in this article in Journal of Neuroscience.
CAPTCHA
This question is for testing whether or not you are a human visitor and to prevent automated spam submissions.
Print
View Full Page PDF
Citation Tools
Dynamics and Mechanisms of Contrast-Dependent Modulation of Spatial-Frequency Tuning in the Early Visual Cortex
Hiroki Tanaka, Ryohei Sawada
Journal of Neuroscience 14 September 2022, 42 (37) 7047-7059; DOI: 10.1523/JNEUROSCI.2086-21.2022

Citation Manager Formats

  • BibTeX
  • Bookends
  • EasyBib
  • EndNote (tagged)
  • EndNote 8 (xml)
  • Medlars
  • Mendeley
  • Papers
  • RefWorks Tagged
  • Ref Manager
  • RIS
  • Zotero
Respond to this article
Request Permissions
Share
Dynamics and Mechanisms of Contrast-Dependent Modulation of Spatial-Frequency Tuning in the Early Visual Cortex
Hiroki Tanaka, Ryohei Sawada
Journal of Neuroscience 14 September 2022, 42 (37) 7047-7059; DOI: 10.1523/JNEUROSCI.2086-21.2022
Reddit logo Twitter logo Facebook logo Mendeley logo
  • Tweet Widget
  • Facebook Like
  • Google Plus One

Jump to section

  • Article
    • Abstract
    • Introduction
    • Materials and Methods
    • Results
    • Discussion
    • Footnotes
    • References
  • Figures & Data
  • Info & Metrics
  • eLetters
  • PDF

Keywords

  • coarse-to-fine processing
  • contrast modulation
  • spatial frequency tuning
  • tuning dynamics
  • visual cortex

Responses to this article

Respond to this article

Jump to comment:

No eLetters have been published for this article.

Related Articles

Cited By...

More in this TOC Section

Research Articles

  • Macrophages Promote Repair of Inner Hair Cell Ribbon Synapses following Noise-Induced Cochlear Synaptopathy
  • The Basolateral Amygdala Sends a Mixed (GABAergic and Glutamatergic) Projection to the Mediodorsal Thalamic Nucleus
  • Total Sleep Deprivation Increases Brain Age Prediction Reversibly in Multisite Samples of Young Healthy Adults
Show more Research Articles

Systems/Circuits

  • Dorsolateral Striatum is a Bottleneck for Responding to Task-Relevant Stimuli in a Learned Whisker Detection Task in Mice
  • The Basolateral Amygdala Sends a Mixed (GABAergic and Glutamatergic) Projection to the Mediodorsal Thalamic Nucleus
  • A Short Route for Reach Planning between Human V6A and the Motor Cortex
Show more Systems/Circuits
  • Home
  • Alerts
  • Visit Society for Neuroscience on Facebook
  • Follow Society for Neuroscience on Twitter
  • Follow Society for Neuroscience on LinkedIn
  • Visit Society for Neuroscience on Youtube
  • Follow our RSS feeds

Content

  • Early Release
  • Current Issue
  • Issue Archive
  • Collections

Information

  • For Authors
  • For Advertisers
  • For the Media
  • For Subscribers

About

  • About the Journal
  • Editorial Board
  • Privacy Policy
  • Contact
(JNeurosci logo)
(SfN logo)

Copyright © 2023 by the Society for Neuroscience.
JNeurosci Online ISSN: 1529-2401

The ideas and opinions expressed in JNeurosci do not necessarily reflect those of SfN or the JNeurosci Editorial Board. Publication of an advertisement or other product mention in JNeurosci should not be construed as an endorsement of the manufacturer’s claims. SfN does not assume any responsibility for any injury and/or damage to persons or property arising from or related to any use of any material contained in JNeurosci.