Fixational eye movements depend on task and target

Human fixational eye movements are so small and precise that they require high-speed, accurate tools to fully reveal their properties and functional roles. Where the fixated image lands on the retina and how it moves for different levels of visually demanding tasks is the subject of the current study. An Adaptive Optics Scanning Laser Ophthalmoscope (AOSLO) was used to image, track and present Maltese cross, disk, concentric circles, Vernier and tumbling-E letter fixation targets to healthy subjects. During these different passive (static) or active (discriminating) fixation tasks under natural eye motion, the landing position of the target on the retina was tracked in space and time over the retinal image directly. We computed both the eye motion and the exact trajectory of the fixated target’s motion over the retina. We confirmed that compared to passive fixation, active tasks elicited a partial inhibition of microsaccades, leading to longer drifts periods compensated by larger corrective saccades. Consequently the fixation stability during active tasks was larger overall than during passive tasks. The preferred retinal locus of fixation was the same for each task and did not coincide with the location of the peak cone density.


Introduction
When fixating our gaze on an object, our eyes are never truly at rest. Even while staring at a small object, like the bottom row of for two reasons. First, it will continue to build our knowledge on the remarkable ability for fine foveal vision in humans. A second, 51 more practical, reason is to learn what target and/or fixation task might minimize overall FEM in clinical settings where motion and its 52 consequent blurred or distorted retinal images can be detrimental, such as with fundus photography, OCT scans or microperimetry. 53 The current study aims to compare and contrast FEM during active fixation tasks -those that contain temporal variation and 54 require subject input -and passive fixation tasks, where the subject is simply instructed to maintain fixation on a target. An Adaptive 55 Optics Scanning Laser Ophthalmoscope (AOSLO) is used as an eye tracker to acquire high spatial (<1 arcmin) and temporal (960 Hz) 56 resolution eye traces. Since the AOSLO can also obtain an unambiguous record of the motion of the target that is projected onto the 57 retinal surface, we compare how the preferred retinal locus for fixation (PRL) relates to the location of peak cone density (PCD) for 58 each type of fixation target. combined to construct videos of the retina with 512x512 pixel sampling resolution at a frame rate of 30Hz (the speed of the slow vertical 76 scanner). For this experiment, the imaging wavelength was 680nm, with 940nm used for wavefront sensing. The field size of the video 77 was 0.9 x 0.9 • . Using an average power of 50-70 µW , the raster scan field appeared as a bright red square to the subject. Fixation targets 78 were presented to the subject within the red field by turning off the scanning laser using an acousto-optic modulator (Brimrose Corp, 79 MD) at the appropriate time points during the raster scan. To the subject, these targets appeared as black-on-red decrements. The stimuli 80 were very sharp and had high contrast owing to the use of adaptive optics on the input scanning beam. Importantly, these decrements 81 are also encoded directly into the video, which allows for an unambiguous measurement of the motion of the image of the fixation target

87
The experiment consisted of 5 different conditions: Maltese cross, disk, concentric circles, Vernier acuity, and a tumbling E (M, D, 88 C, V, and E respectively). The Maltese cross condition (M) was chosen as it has been suggested to provide a better fixation target than 89 the simple dot that is commonly used in fixation tasks. The disk condition (D) consisted of an annulus within the center of the raster that 90 the subjects were instructed to fixate. Both of these conditions were simple passive fixation tasks where subjects were instructed to hold 91 their gaze on the target. The concentric circles condition (C) consisted of concentric rings moving in a constricting radial motion. There 92 were 6 rings ranging in size from 10 to 1 arcmin that were presented over the course 18 frames (3 frames per ring size) and replayed 93 every 30 frames for a frequency of 1 Hz. The aim of the concentric rings was to provide a simple fixation task that was more visually 94 engaging than a static target. The Vernier hyperacuity condition (V) required subjects to judge the relative displacement of two tiny 95 horizontal bars which appeared at random intervals (seven 6-arcsecond steps). The tumbling E condition (E) consisted of a tumbling 96 E task where the subjects were asked to report the orientation of a letter E as it rotated randomly. The size of the E varied in seven 97 steps, from 20/6 to 20/20 Snellen acuity. For both V and E tasks the stimulus was presented for 0.5 sec (15 consecutive frames) and 98 there were random time intervals between presentations -evenly spread over 0.5 to 1.5 sec -where nothing was presented. The random 99 time intervals were used so that subjects could not anticipate the next trial and were therefore compelled to maintain fixation the entire 100 time. The V and E condition can be differentiated from the others as they both required active subject judgment and response, as well 101 as providing temporal variation. These conditions were further categorized into passive tasks (M, D) and active tasks (E, V) for further 102 analysis depending on whether they required subject response and varied in time. The concentric circles provided a mixed task as it had 103 temporal variations but did not require subject response. See Table 1 for the differences between the conditions. The different fixation 104 targets were presented in a pseudo-random order to eliminate any training or fatigue effects. Furthermore, subjects were given consistent instructions from a script to avoid known changes in behavior due to instruction (  the mean luminance fell below a threshold that was defined on a per-subject basis dependent on the average brightness of the respective 119 movies.

120
The AOSLO records high-resolution videos of the retina for each trial and the fixation target is directly encoded into the video,  would suppress their microsaccades, drift away from the target, and then be forced to make relatively larger microsaccades to reorient 173 themselves during the active discrimination tasks. The active tasks also had overall larger fixation as indicated by the isoline area.

174
Overall, these data reveal that there was more motion in the active tasks vs the passive tasks, even though there were fewer saccades.  Figure 3: FEM measurements across the five conditions. In general, the eye tended to have longer and larger intersaccadic periods of drift in the active fixation tasks where subject response was required (purple bars). Whereas during the passive fixation tasks (green bars), subjects tended to make shorter but more frequent saccades. The circle task (orange bar) represents a mixed task. While it did not require subject response, it still had temporal variations that were significant enough to differentiate it from a simple fixation task.
Subject behavior in the circles tasks did not readily align with either the passive or the active tasks. Error bars represent S.E.M.
In recent years, fixation has been considered a more "active" process than previously thought (Rucci & Victor, 2015). In order to 176 tease apart the effects of different fixation targets, it is prudent to analyze behavior under active fixation conditions, in which the subject 177 is required to attend and respond to a changing target; compared to passive fixation tasks, where the subject is required to simply hold 178 their gaze on a stationary target. The circle was excluded from this analysis since it represents a mix of these two distinct categories, 179 and behavior during the circle task did not readily align with either the passive or active tasks. Since the two active tasks (Vernier Active and passive tasks were averaged together, as well as their error bars in order to simplify the analysis to just the active vs the passive tasks. Error bars represent S.E.M and the asterisks represent levels of significance (repeated-measures t-test, p<0.05, p<0.01, and p<0.001 respectively). Figure 5 shows the 68% isoline contours for each subject for each condition centered on the PRL, which is defined at the peak of 185 the fixation positions' PDF. The stimulus for each condition is drawn in the center of each graph for reference, but the stimulus will 186 sweep across the retina based on the extent of the eye motion. Although the overall fixation area was larger for the active tasks compared 187 to the passive tasks (two-sampled t-test, p<0.01). Extensive intersubject variability is readily apparent in Figure 5. The average standard 188 deviation of the ISOA between subjects for each condition (columns in Figure 5) was 43.53 arcmin 2 . Whereas the average standard 189 deviation between conditions for each subject (rows in Figure 5

218
Our study has shown that large and significant variations of microsaccade and drift kinematics exists between subjects and between 219 different tasks. We have confirmed that an individual's FEM behavior depends on the task involved. Specifically, we have found 220 that active tasks result in less frequent microsaccades and correspondingly longer and larger drift epochs which, in turn, cause the 221 microsaccades to be larger. This pattern of behavior leads to overall larger ISOAs during active fixation tasks. Passive tasks, by 222 comparison, are marked by shorter and more frequent microsaccades and smaller, briefer drift epochs, all leading to a smaller ISOA.

223
The increase in ISOA from passive to active tasks was 57% (+/-23% S.D.) on average. These results can be explained in part by previous 224 findings that suggest that FEM is an active behavior that is subconsciously mediated to serve different functions depending on the task at 225 hand (Rucci & Victor, 2015). During the active tasks subjects tended to suppress their microsaccades because the rapid transients from 226 these movements can be detrimental to fine-scale discrimination due to microsaccadic suppression, either from blurring of the retinal 227 image or central suppression.

228
The stability of the PRL was tested using a 2-dimensional Kolmonorov-Smirnov test and was found to remain the same regardless conditions or for more complex viewing experiences, such as during smooth pursuit or fixation within extended scenes.

232
Although the peak cone density on the retina offers the best location for photoreceptor spatial sampling (according to the Nyquist ). Interestingly, we found that these corrective saccades appear to only partially "refoveate" the stimuli (i.e. reposition the image on 244 the PRL), instead they generally landed above it. This suggests that the visual system might be very much aware of its own oculomotor 245 ability and could be anticipating the likely direction of the following drift so that the target will fall on the PRL during the middle of the 246 drift segment. In any case, the minutiae of FEM reveals that a PRL that is identified by any of the current methods, including the ISOA 247 approach used here, may be ill-defined. This is a topic of an ongoing investigation.

248
While we measured significant and informative differences in FEM between conditions, we found that differences in FEM between 249 individuals are even greater. The standard deviation of the ISOA between individuals, for example, was roughly twice that between 250 conditions. These differences can be partly explained by experience (Cherici et al., 2012 When considering these small fixational eye movements it is prudent to consider the goals for maintaining a subject's steady 255 fixation, since all fixations are not equal. If the goal of the fixation target is to minimize the overall movement of the eyes (as is the case 256 in many clinical situations), then one must consider which types of FEM are most likely to be an impediment. If the rapid transients from 257 saccades are most likely to have a deleterious effect then it is preferred to rely on an active fixation task so the subject will suppress their 258 microsaccades in order to perform the task. If the goal is to minimize the total area covered by the fixation, then choosing a more passive 259 fixation target is likely to be most effective. Finally, if the goal is to measure FEM and understand their role in regards to ecologically 260 valid situations, it is recommended to use active fixation tasks, since these tasks will reflect real-world discrimination of fine stimuli.

262
This study examined the influence of different fixation targets and tasks on FEM and the location of the PRL in healthy eyes.

263
Using an AOSLO, we developed a new method to locate and follow the target projected on the retina over time relative to the PCD. We