Easily adaptable head-free training system of macaques for tasks requiring precise measurements of eye position

We describe a modified system for training macaque monkeys without invasive head immobilization on visuomotor tasks requiring the control of eye-movements. The system combines a conventional primate chair, a chair-mounted infrared camera for measuring eye-movements and a custom-made concave reward-delivery spout firmly attached to the chair. The animal was seated head-free inside the chair but the concavity of the spout stabilized its head during task performance. Training on visual fixation and discrimination tasks was successfully performed with this system. Eye-measurements, such as fixation-precision, pupil size as well as micro-saccades were comparable to those obtained using conventional invasive head-fixation methods. The system is inexpensive (∼$40 USD material cost), easy to fabricate in a workshop (technical drawings are included), and readily adjustable between animals without the need to immobilize or sedate them for these adjustments. Highlights We developed an approach to train macaque monkeys head-free on visuomotor tasks requiring measurements of eye position The setup is inexpensive, easy to build, and readily adjusted to the animal without the need for sedation The system was tested for training on a visual fixation and a visual discrimination task Eye measurements (fixation precision, pupil size, microsaccades) were comparable to those from head-fixed animals

• The system was tested for training on a visual fixation and a visual discrimination task 23 • Eye measurements (fixation precision, pupil size, microsaccades) were comparable to those 24 from head-fixed animals 25 26 27

Introduction 46
Basic systems neuroscience research of cognitive behavior, such as attention, perception and 47 decision-making, often rely on awake macaque monkeys (Roelfsema and Treue, 2014) 48 performing visuomotor tasks. Animals in such studies are frequently trained extensively prior to 49 data acquisition, and many tasks require tight control of the animals' eye movements (e.g. 50 (Clery et al., 2017)). For precise measurements of eye position the animals' heads are typically 51 fixed to a primate chair using head-posts surgically implanted to the animal's skull (e.g. (Adams 52 et al., 2007;Betelak et al., 2001)). To begin the animal's training using such approaches 53 therefore requires an invasive head-post implantation, and frequently several weeks to months 54 post-surgically for healing and successful osseointegration of the implant into the skull (Betelak 55 et al., 2001) to ensure good stability. 56 57 An approach to allow for head-free training without the need for a surgery is therefore desirable 58 not only as a refinement of research with animals from an animal welfare perspective (e.g. 59 (Prescott et al., 2010)), but also because of its potential to accelerate the training procedure, e.g. sedation. Other approaches using transport-boxes of smaller rhesus and new-world monkeys 65 monitored spontaneous gaze direction to natural images in the absence of operant conditioning 66 (Ryan et al., 2019). We focus here on a non-invasive training systems in combination with a 67 primate chair to ease integration into conventional set-ups using head-fixation. We developed a 68 system that is sufficiently flexible that it only requires coarse measurements of the animal's face, 69 which can be obtained from an animal while seated in a primate chair. The system is integrated 70 in a standard primate chair combined with a commercial eye-tracker. It is inexpensive and 71 simple to build in a standard machine shop (material cost ~$40USD). We show training data 72 from a visual fixation and discrimination task in one animal as well as detailed measurements of 73 eye movements and pupil size, which were comparable to those obtained from the same animal 74 under head-fixation and two additional head-fixed animals. Because of its flexibility and 75 simplicity the system has the potential to be more widely adapted. 76 77 2. Methods 78

Design of the reward spout for head-free training 79
To train the animal without head-fixation, we used a custom-made concavely shaped reward 80 spout mounted firmly to the chair (schematics and technical drawings in Fig. 1a and c-f, 81 respectively). It consisted of an engineering thermoplastic (copolymere polyoxymethylene, 82 POM-c, "Delrin", DuPont) milled to a concave cone whose base pointed towards the animal and 83 in whose center a reward tube (OD 4mm, ID 2mm) was inserted and secured with a screw (Fig.  84 1 c, red arrow). The thermoplastic material was chosen for its sturdiness without being brittle to 85 withstand the animal's attempts to bite into the rim of the concavity. A wide slit in the downward 86 facing side of the cone ensured that no liquid accumulated inside the concavity of the spout. 87 The blunt top of the cone transitioned to a solid cylindrical part (Fig. 1c). Both conical and 88 cylindrical component were milled in one piece. The depth and diameter of the cone was 89 chosen such that a) it reduced the range of possible head positions from which rewards could 90 be sampled and b) no shadows were cast on the eye to allow for good quality monocular eye-91 signals. Note that while we designed this spout to approximately match the ventral-dorsal extent 92 (3-4 cm) and width of the upper jaw (4-5 cm) of the animal, our measurements were coarse and 93 did not seem critical. Indeed, we also initially tried a substantially longer spout (87.1 mm instead 94 of 66.9 mm, cf. Fig. 1c), which allowed for overall satisfactory measurements of eye signals 95 although it was more prone to cast shadows on the animal's eye and hence not used further. 96 Moreover, small adjustments could be made by changing the distance by which the reward tube 97 protruded from the bottom of the concavity inside the reward spout. The conical design of the 98 spout in our system was done for convenience of the milling in the manufacturing process. 99 Given the ease to obtain good eye signals with this design there was no need to further refine 100 the shape of the spout. But if, e.g. shadows from the spout led to deteriorated signal quality a 101 narrower shape better tailored to the bridge the noise (resulting in a more triangular cross-102 section of the spout) could be considered. 103

104
The cylindrical part of the spout was screwed to two aluminum rods (OD 10mm), one on each 105 side (Fig. 1e, "front rods"). Note that the spout had to be tightly screwed to the rods to ensure 106 that the monkey could not rotate the spout around the axis of the aluminum rods. These rods 107 (oriented parallel to the front of the primate chair) were then mounted via aluminum cross-108 connectors ( Fig. 1e) to two aluminum rods (OD 10mm, Fig. 1e, "side rods") that were mounted 109 to the vertical walls of the chair (see Fig. 1b, oriented parallel to the sides of the chair) via 110 aluminum clamps (Fig. 1c). These latter rods remained mounted to the chair between training 111 sessions, while the former and the reward spout were removed between sessions. Only the 112 screws in the cross-connectors (red arrows in Fig. 1b)  This study was approved by the local authorities (Regierungspraesidium Tübingen). We 143 collected data from three male rhesus monkeys, A, M and K (Macaca mulatta; K: 6.5kg, M: 8kg 144 and A: 12kg) performing a standard visual fixation task (monkey M and K only under head-145 fixation and monkey A both head-free and under head-fixation), an orientation discrimination 146 task (monkey A) and a disparity discrimination task (all three monkeys under head-fixation). The 147 monkeys were implanted under general anesthesia with a titanium head-post base on their skull, 148 under their skin, and we developed this head-free training system to take advantage of the post-149 surgical period for osseointegration in animal A. He was naïve to any behavioral training in a 150 laboratory setting except for climbing into a standard primate chair. After the animal was trained 151 using the head-free system, a metal holder was screwed into the titanium base of the head-152 holder to allow for conventional head-fixation. 153 154

Behavioral training 155
While monkey M and K received conventional fixation training under head-fixation, monkey A 156 was initially trained using the head-free system. First, he was habituated to the spout of our 157 system and the reward delivery (four sessions). He was then trained on a standard visual 158 fixation task by gradually increasing the fixation duration. We here analyzed the results of all 46 159 sessions of the fixation training in this animal. Following the fixation training we initiated training 160 on an orientation discrimination task. We initially used the contrast of the distractor target as an 161 additional cue and report here the initial 16 sessions during which both targets were at full 162 contrast such that the only cue the animal could use was the stimulus orientation. 163

Visual fixation task 164
The monkeys were required to fixate within a window around the fixation dot 0.1° dva in the 165 center of the monitor to receive juice or water rewards. During the initial training sessions we 166 progressively increased the fixation duration until it reached 2 sec. Once the animals could 167 reliably fixate for 2 sec, we started to present a peripheral visual stimulus (typically a drifting 168 luminance grating) on the screen. 169 170 2.5.2. Orientation discrimination task 171 After animal A learned to maintain stable fixation using the head-free system, we began to train 172 him on a two-alternate forced-choice (2AFC) orientation discrimination task, similar to (Nienborg 173 and Cumming, 2014). Once the animal acquired fixation the stimulus appeared (typically for 2 174 sec), as well as two choice targets, a horizontally and vertically oriented Gabor, respectively, 175 presented above and below the fixation marker. The vertical position of the choice target was 176 randomized. Once the central fixation marker disappeared the animal was allowed to make his 177 saccade indicating the choice. A saccade to the target whose orientation matched that of the 178 stimulus was rewarded. To discourage the animal from guessing, the available reward size was 179 increased based on his task performance. After three consecutive trials with correct choices, the 180 available reward size was doubled compared to the original reward size. After four consecutive 181 trials with correct choices, the available reward size was again doubled (quadruple compared to 182 the original size) and remained at this size until the next error. After every error trial, the 183 available reward size was reset to the original. For the analyses in In the orientation discrimination task, the stimuli were 2D Gabor whose orientation and phase 199 was randomly changed on each video-frame (60Hz). Orientation signal strength on each trial where erf denotes the error function, and µ and σ are the mean and standard deviation of the 218 fitted cumulative Gaussian distribution, respectively. The standard deviation σ was defined as 219 the psychophysical threshold and correcponds to the 84% correct level. 220 221

Preprocessing of eye traces 222
We transformed the eye position x to velocity v, which represented a moving average of 223 velocities over 5 data samples (Engbert and Kliegl, 2003): where Δt corresponds to 1/sampling rate. Eye position values were reconstructed using these 225 velocity values to suppress noise (Engbert and Mergenthaler, 2006): 226

Training on a standard visual fixation task 271
We devised this head-free training system to train animals prior to any surgery and to take 272 advantage of the post-surgical period (3-6 months, (Betelak et al., 2001)) aimed at ensuring 273 successful osseointegration of the base-part of a two-part head-fixation implant (Fig. 2a). 274 275 We tested this system in one male animal (A) naive to any behavioral training other than to 276 enter the primate chair. The animal was seated in a standard primate chair and trained via 277 operant conditioning to stabilize his head in a fixed position to receive fluid rewards (Fig. 2b, left). 278 Although the animal could move his head freely, the concavity of the reward-spout reduced the 279 variability of the head position whenever he was seeking rewards (Fig. 2b, right) to allow for 280 reliable measurements of monocular eye-position. After four sessions of habituation to the 281 reward spout we were able to start monitoring the animal's eye position using the chair-mounted 282 video-based eye-tracker during a standard visual fixation task. Example measured eye traces 283 and pupil size are shown in Fig. 2c. Using this head-free system, the monkey was able to fixate 284 successfully for 2 sec within 6 training session, which was somewhat shorter than the number of 285 sessions required in two other animals using conventional head-fixation (13 and 28 sessions in 286 animal M and K, respectively, Fig. 3b). In session seven, following the animal A's reliable ability 287 to fixate for 2 sec we began to simultaneously present a visual stimulus peripherally during the 288 fixation period. Starting in session eight, we could reduce the size of the fixation window width 289 and height to below 2.5 o x 2.5 o (Fig. 3c). These increased fixation requirements resulted in a 290 relatively stable proportion of fixation breaks across sessions (Fig. 3d), while the animal worked 291 for increasingly longer sessions over the course of training (Fig. 3e). 292 To examine the quality of the eye-position measurements using this system, we quantified the 293 fixation precision as the variance of eye position (Fig. 4a) and fixation span (Cherici et al., 2012) 294 both within (Fig. 4b) and across (Fig. 4c) trials. We observed that these values improved over 295 the course of training (Spearman's rank correlation with session number; 4b: r = -0.59, p < 10 -4 296 4c r = -0.76, p < 10 -8 ). After about 30 sessions the fixation precision reached values that 297 approached those obtained for fully trained animals under head-fixation (Fig. 4b, c, right). This 298 shows that comparable fixation precision can be reached in our head-free system to that with 299 conventional head-fixation using implanted head-posts. 300 301

Training on an orientation discrimination task 302
Once the animal achieved good fixation performance we began to train him on an orientation 303 discrimination task (see Methods). During this training we gradually increased the contrast of 304 the incorrect target until the only cue available to the animal to solve the task was the orientation 305 of the stimulus. Here, we analyzed the initial 16 sessions for which both targets were at full 306 contrast such that the animal had to rely on the orientation of the stimulus to solve the task (Fig.  307  2018). We therefore wondered whether our measurements of the eye signals in the head-free 318 system were of sufficient quality to observe such modulation in this task as well. In the task we 319 used, reward size was changed in a systematic way based on performance (see Methods). We 320 therefore computed the average pupil size in the last 250ms during the stimulus presentation, 321 (as done for head-fixed animals in (Kawaguchi et al., 2018)), and compared this metric for small 322 and large available reward trials across the 16 sessions analyzed here. We found that pupil size 323 was significantly larger in large available reward trials (p = 0.011; Wilcoxon signed rank test; Fig.  324 6b), very similar to our results in head-fixed animals ( (Kawaguchi et al., 2018), their Fig. 3e). traces from two example trials in Fig. 6a). The rate of microsaccades was comparable to that 334 obtained under head-fixation (head-free animal A: 1.14 ± 0.20 s -1 ; head-fixed animal A: 1.64 ± 335 0.19 s -1 ; head-fixed animal M: 2.19 ± 0.13 s -1 ; head-fixed animal K: 1.23 ± 0.24 s -1 ; mean ± SD) 336 and to values observed in human observers, e.g. (Cherici et al., 2012). Moreover, the 337 microsaccades showed the characteristic linear relationship between peak velocity and 338 amplitude (Zuber et al., 1965), which was similar to that obtained under head-fixation in the 339 same and two additional animals (Pearson correlation: r = 0.61 for 16 sessions, 0.55 for 7 340 sessions in head-free and head-fixed animal A, respectively; r=0.70 in 8 sessions and r=0.47 in 341 14 sessions in head-fixed animal M and K, respectively; p<<10 -10 in all cases, Fig. 7b). 342 343

Transitioning to head-fixation following the head-free training 344
After a five-month period of training using the head-free system we transitioned animal A to 345 conventional head-fixation using an implanted head-post. Since the animal had become 346 accustomed to the voluntary engagement in the task and sometimes rotated his body while 347 being seated in the chair we were initially concerned that head-fixation might substantially 348 disrupt his trained visual fixation behavior. But the animal adapted quickly to the head-fixation 349 and required only 10-15 additional sessions of fixation training (Fig. 4b and c, middle, green 350 abscissa) to approach the fixation precision of the fully trained animals that only received 351 fixation training under head-fixation. 352 353 Together, these results show that good quality eye-signal measurements can be obtained with 354 this head-free system, allowing for the training on sensorimotor tasks, and that an animal 355 trained head-free can then readily adapt to head fixation. 356

Discussion 357
We described a modified non-invasive system to train macaque monkeys without head-fixation. 358 It has the advantages that, unlike previous non-invasive designs ( An important observation was that the transition between the head-free system to conventional 375 head-fixation required only minimal re-training, suggesting very little cost in time to employ initial 376 head-free training even when the animal will ultimately engage in experiments requiring head-377 fixation. This observation, together with the simplicity, flexibility and low material cost should 378 result in a low threshold to adopt this system to more efficiently train the animals as well as 379 improve animal welfare (Prescott et al., 2010), by refining existing set-ups using video-based 380 eye tracking. Finally, the quality of the eye measurements, which was comparable to those 381 obtained under head-fixation, makes the system amenable to combine with neuronal recordings. 382 These may be a tethered configuration with, e.g. chronically or semi-chronically implanted 383 electrodes (e.g. (Ruff et al., 2016)), but would require training the animal to tolerate the touch to 384 the head associated with connecting the cables of the recording system, or wireless recordings 385 left two columns show data for animal A for head-free and head-fixed task performance, 470 respectively. The third and forth column show data for head-fixed animals M and K during task 471 performance, respectively. a) Example eye traces with labeled microsaccades (black) for two 472 example trials. b) The characteristic relationship between microsaccade amplitude and peak 473 velocity was also found in animal A in the head-free condition, similar to the head-fixed animals.

475
Author contributions 476 K.K, L.S, P.P, and S.C performed data collection. K.K analyzed data. H.N conceived of the 477 head-free system and supervised the project.