Final Report Summary - HVGC (Imaging neural gain control in the human visual system)
OVERVIEW
This research project had an overall goal of understanding more about how contrast and colour are processed in the human visual system with particular attention paid to a process called ‘gain control’. To address this goal, we used a combination of functional magnetic resonance imaging (fMRI) and electrophysiology (EEG). As well as using brain measurement techniques to examine the way that gain control alterer neuronal responses, we also used a technique called ‘psychophysics’ to ask how gain control altered what people saw.
fMRI
Using fMRI we examined the computations involved in long-range gain control in early visual cortex. In these experiments we asked whether long-range suppression in striate cortex (V1) is tuned for elementary features such as orientation and luminance polarity. This work was particularly important because it spoke to recent theoretical observations suggesting that the salience of visual stimuli is determined by the magnitude of the response they elicit in V1. Our work provided evidence that elements with similar features do suppress each other as early as V1 and leant support to the ‘early’ saliency hypothesis.
Additional studies on this phenomenon indicated that the effect was robust. Self-suppression (a manifestation of feature-tuned gain control) was present in both V1 and higher areas like V2 and V4. Moreover, the magnitude of the release from suppression in V1 was strongly predictive of the saliency of the perceptual ‘popout’ features.
We also used fMRI to study the interaction of contrast adaptation (a gain control computation) and attention (also modeled as a gain control computation). Our data indicated a surprising result: adaptation in V1 did not appear to be modulated by attention. Although we observed profound effects of both attention and stimulus contrast on the fMRI BOLD signal, the two did not interact: attending to a particular location did not amplify the magnitude of the subsequent adaptation effect. More subtly, recent work by other groups has shown that attention is a complex gain control computation that can increase as well as decrease sensitivity. Additional experiments conducted as part of this project confirmed this basic result but also demonstrated an invariance to attention. This invariance was particularly surprising given the observation that attention did appear to alter the perceived contrast of a subsequent probe target presented in the same location. We are currently working on additional experiments to identify the cortical site at which the BOLD signal correlates with this perceived, attentionally-driven contrast modulation.
Animal EEG
Because contrast gain control appears to be a fundamental ‘canonical’ neuronal computation, our lab has spent considerable time over the past four years developing a rapid, portable electrophysiological assay of this phenomenon. In slightly different configurations, this test can be used for both human lab-based work (for example, to examine gain control mechanisms involved in attention or the effect of neurological disease on visual processing) and also for animal work. As a direct result of the CIG grant, we were awarded a Wellcome-funded Institutional C2D2 grant to pursue the application of these gain control measurements in the study of Parkinson’s Disease in an international study involving patients in North Africa as well as animal models in both the UK and Denmark.
The results of our animal studies have resulted in two high-impact papers. In our first study, we demonstrated that gain control is abnormal in a transgenic Drosophila model of human PD. This abnormality results in vastly increased visual responses in the flies at an early age leading to atrophy of the visual system in later life. In collaboration with a new industrial partner (Lundbeck A/S, Copenhagen) we showed that tool compounds that target the increased activity of the kinase domain in the mutant transgenic protein also restore normal function. This technique therefore has profound implications for both drug discovery and patient monitoring and the work has resulted in a subsequent investment by Lundbeck in the University of York to develop additional animal models and faster, more robust visual assays.
The second paper demonstrated that different Parkinson’s disease genes alter the pattern of visual sensitivity across a range of spatial and temporal frequencies. Using advanced machine learning techniques, we were able to classify the genotype of individual animal PD models with close to 80% accuracy based solely on their visual responses. This work has led to additional projects (including a PhD project funded by a Marie Curie ITN grant: ‘NextGenVis’) and we are in the process of extending its findings to human patients.
Human EEG
We have collected a dataset from a large cohort of PD patients in Tunisia using a newly-developed, portable, low-profile EEG system This is the first SSVEP dataset to acquire data from genotyped PD patients and it is also the first to collect large amounts of data from patients with the LRRK-G2019S genotype (the most common genetic form of PD). Analysis of this dataset is underway and a manuscript is in preparation.
SSVEP techniques have also been used to examine an outstanding question in human visual gain control: In 2012 it was observed that locomotion in rodents reduced the level of surround suppression in the mouse brain (a form of long-range gain control). We have examined whether the same effect is observed in humans using SSVEP and an exercise treadmill. Our data indicate that while the SSVEP technique can measure profound surround suppression in human vision, it is not modulated by locomotion. This provides important evidence of functional differences between basic contrast processing in human and rodent models of vision. As suggested in the original grant proposal, we have related these data to matched perceptual experiments that also demonstrate no change in surround suppression driven by locomotion. These data are of particular interest because the mouse is now being adopted as a genetically tractable model of visual processing. Significant low-level computational differences between mouse and human vision should therefore be noted by the vision community.
CONCLUSION
We have used both EEG and fMRI to examine the spatial and temporal characteristics of long-range gain control. We have also related these data to perception - the ultimate goal being to predict the appearance of novel, dynamic stimuli. In an extension to the original goals of the projects, we have also used a version of the SSVEP technique to examine vision in animal models of neurological disease. These experiments have been synergistic with our human work and provide important information about potential disease biomarkers that we are addressing in a new round of experiments.
Our data are of interest to vision scientists because they provide new information about the way that signals are processed in the visual brain. They are also of interest to clinicians and drug companies because they suggest new biomarkers for neurological disease and therefore new ways of monitoring those diseases and testing potential therapeutic agents. Finally, they are of interest to the general public because they show that some of the mechanisms that underlie vision have been preserved across the millions of years that separate humans and flies and that these mechanisms can be used to tell us about neurological diseases that afflict millions of people across the globe.
This research project had an overall goal of understanding more about how contrast and colour are processed in the human visual system with particular attention paid to a process called ‘gain control’. To address this goal, we used a combination of functional magnetic resonance imaging (fMRI) and electrophysiology (EEG). As well as using brain measurement techniques to examine the way that gain control alterer neuronal responses, we also used a technique called ‘psychophysics’ to ask how gain control altered what people saw.
fMRI
Using fMRI we examined the computations involved in long-range gain control in early visual cortex. In these experiments we asked whether long-range suppression in striate cortex (V1) is tuned for elementary features such as orientation and luminance polarity. This work was particularly important because it spoke to recent theoretical observations suggesting that the salience of visual stimuli is determined by the magnitude of the response they elicit in V1. Our work provided evidence that elements with similar features do suppress each other as early as V1 and leant support to the ‘early’ saliency hypothesis.
Additional studies on this phenomenon indicated that the effect was robust. Self-suppression (a manifestation of feature-tuned gain control) was present in both V1 and higher areas like V2 and V4. Moreover, the magnitude of the release from suppression in V1 was strongly predictive of the saliency of the perceptual ‘popout’ features.
We also used fMRI to study the interaction of contrast adaptation (a gain control computation) and attention (also modeled as a gain control computation). Our data indicated a surprising result: adaptation in V1 did not appear to be modulated by attention. Although we observed profound effects of both attention and stimulus contrast on the fMRI BOLD signal, the two did not interact: attending to a particular location did not amplify the magnitude of the subsequent adaptation effect. More subtly, recent work by other groups has shown that attention is a complex gain control computation that can increase as well as decrease sensitivity. Additional experiments conducted as part of this project confirmed this basic result but also demonstrated an invariance to attention. This invariance was particularly surprising given the observation that attention did appear to alter the perceived contrast of a subsequent probe target presented in the same location. We are currently working on additional experiments to identify the cortical site at which the BOLD signal correlates with this perceived, attentionally-driven contrast modulation.
Animal EEG
Because contrast gain control appears to be a fundamental ‘canonical’ neuronal computation, our lab has spent considerable time over the past four years developing a rapid, portable electrophysiological assay of this phenomenon. In slightly different configurations, this test can be used for both human lab-based work (for example, to examine gain control mechanisms involved in attention or the effect of neurological disease on visual processing) and also for animal work. As a direct result of the CIG grant, we were awarded a Wellcome-funded Institutional C2D2 grant to pursue the application of these gain control measurements in the study of Parkinson’s Disease in an international study involving patients in North Africa as well as animal models in both the UK and Denmark.
The results of our animal studies have resulted in two high-impact papers. In our first study, we demonstrated that gain control is abnormal in a transgenic Drosophila model of human PD. This abnormality results in vastly increased visual responses in the flies at an early age leading to atrophy of the visual system in later life. In collaboration with a new industrial partner (Lundbeck A/S, Copenhagen) we showed that tool compounds that target the increased activity of the kinase domain in the mutant transgenic protein also restore normal function. This technique therefore has profound implications for both drug discovery and patient monitoring and the work has resulted in a subsequent investment by Lundbeck in the University of York to develop additional animal models and faster, more robust visual assays.
The second paper demonstrated that different Parkinson’s disease genes alter the pattern of visual sensitivity across a range of spatial and temporal frequencies. Using advanced machine learning techniques, we were able to classify the genotype of individual animal PD models with close to 80% accuracy based solely on their visual responses. This work has led to additional projects (including a PhD project funded by a Marie Curie ITN grant: ‘NextGenVis’) and we are in the process of extending its findings to human patients.
Human EEG
We have collected a dataset from a large cohort of PD patients in Tunisia using a newly-developed, portable, low-profile EEG system This is the first SSVEP dataset to acquire data from genotyped PD patients and it is also the first to collect large amounts of data from patients with the LRRK-G2019S genotype (the most common genetic form of PD). Analysis of this dataset is underway and a manuscript is in preparation.
SSVEP techniques have also been used to examine an outstanding question in human visual gain control: In 2012 it was observed that locomotion in rodents reduced the level of surround suppression in the mouse brain (a form of long-range gain control). We have examined whether the same effect is observed in humans using SSVEP and an exercise treadmill. Our data indicate that while the SSVEP technique can measure profound surround suppression in human vision, it is not modulated by locomotion. This provides important evidence of functional differences between basic contrast processing in human and rodent models of vision. As suggested in the original grant proposal, we have related these data to matched perceptual experiments that also demonstrate no change in surround suppression driven by locomotion. These data are of particular interest because the mouse is now being adopted as a genetically tractable model of visual processing. Significant low-level computational differences between mouse and human vision should therefore be noted by the vision community.
CONCLUSION
We have used both EEG and fMRI to examine the spatial and temporal characteristics of long-range gain control. We have also related these data to perception - the ultimate goal being to predict the appearance of novel, dynamic stimuli. In an extension to the original goals of the projects, we have also used a version of the SSVEP technique to examine vision in animal models of neurological disease. These experiments have been synergistic with our human work and provide important information about potential disease biomarkers that we are addressing in a new round of experiments.
Our data are of interest to vision scientists because they provide new information about the way that signals are processed in the visual brain. They are also of interest to clinicians and drug companies because they suggest new biomarkers for neurological disease and therefore new ways of monitoring those diseases and testing potential therapeutic agents. Finally, they are of interest to the general public because they show that some of the mechanisms that underlie vision have been preserved across the millions of years that separate humans and flies and that these mechanisms can be used to tell us about neurological diseases that afflict millions of people across the globe.