Decision-Making in the Primate Brain by Caroline Beth Drucker

Decision-Making in the Primate Brain

Caroline Beth Drucker

Department of Neurobiology Duke University

Date:______Approved:

______Michael L. Platt, Co-Supervisor

______Elizabeth M. Brannon, Co-Supervisor

______Marc Sommer, Chair

______Warren Grill

Dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Neurobiology in the Graduate School of Duke University

2016

ABSTRACT

Decision-Making in the Primate Brain

Caroline Beth Drucker

Department of Neurobiology Duke University

Date:______Approved:

______Michael L. Platt, Co-Supervisor

______Elizabeth M. Brannon, Co-Supervisor

______Marc Sommer, Chair

______Warren Grill

An abstract of a dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Neurobiology in the Graduate School of Duke University

2016

Abstract

Making decisions is fundamental to everything we do, yet it can be impaired in various disorders and conditions. While research into the neural basis of decision- making has flourished in recent years, many questions remain about how decisions are instantiated in the brain. Here we explored how primates make abstract decisions and decisions in social contexts, as well as one way to non-invasively modulate the brain circuits underlying decision-making. We used rhesus macaques as our model organism.

First we probed numerical decision-making, a form of abstract decision-making. We demonstrated that monkeys are able to compare discrete ratios, choosing an array with a greater ratio of positive to negative stimuli, even when this array does not have a greater absolute number of positive stimuli. Monkeys’ performance in this task adhered to

Weber’s law, indicating that monkeys—like humans—treat proportions as analog magnitudes. Next we showed that monkeys’ ordinal decisions are influenced by spatial associations; when trained to select the fourth stimulus from the bottom in a vertical array, they subsequently selected the fourth stimulus from the left—and not from the right—in a horizontal array. In other words, they begin enumerating from one side of space and not the other, mirroring the human tendency to associate numbers with space.

These and other studies confirmed that monkeys’ numerical decision-making follows

similar patterns to that of humans, making them a good model for investigations of the neurobiological basis of numerical decision-making.

We sought to develop a system for exploring the neuronal basis of the cognitive and behavioral effects observed following transcranial magnetic stimulation, a relatively new, non-invasive method of brain stimulation that may be used to treat clinical disorders. We completed a set of pilot studies applying offline low-frequency repetitive transcranial magnetic stimulation to the macaque posterior parietal cortex, which has been implicated in numerical processing, while subjects performed a numerical comparison and control color comparison task, and while electrophysiological activity was recorded from the stimulated region of cortex. We found tentative evidence in one paradigm that stimulation did selectively impair performance in the number task, causally implicating the posterior parietal cortex in numerical decisions. In another paradigm, however, we manipulated the subject’s reaching behavior but not her number or color comparison performance. We also found that stimulation produced variable changes in neuronal firing and local field potentials. Together these findings lay the groundwork for detailed investigations into how different parameters of transcranial magnetic stimulation can interact with cortical architecture to produce various cognitive and behavioral changes.

Finally, we explored how monkeys decide how to behave in competitive social interactions. In a zero-sum computer game in which two monkeys played as a shooter or

a goalie during a hockey-like “penalty shot” scenario, we found that shooters developed complex movement trajectories so as to conceal their intentions from the goalies.

Additionally, we found that neurons in the dorsolateral and dorsomedial prefrontal cortex played a role in generating this “deceptive” behavior. We conclude that these regions of prefrontal cortex form part of a circuit that guides decisions to make an individual less predictable to an opponent.

Contents

Abstract ...... iv

List of Tables ...... xii

List of Figures ...... xiv

List of Abbreviations ...... xxii

Acknowledgements ...... xxiv

1. Introduction ...... 1

1.1 Perceptual Decision-Making ...... 1

1.2 Abstract Decision-Making ...... 21

1.3 Value-Based Decision-Making ...... 27

1.4 Social Decision-Making ...... 34

2. Numerical decision-making ...... 37

2.1 Discrete ratio comparison by rhesus macaques ...... 40

2.1.1 Methods ...... 45

Subjects & Housing ...... 45

Apparatus ...... 46

Training ...... 46

Probe Test Session Structure ...... 51

Numerical Ratio-of-Ratios Probe Trials ...... 51

Control 1: Surface-Area-Only ...... 54

Control 2: Numerosity-Only ...... 56

vii

Control 3: Surface Area versus Numerosity ...... 57

2.1.2 Results ...... 60

Shape Training ...... 60

Ratio Training ...... 60

Numerical Ratio-of-Ratios Testing ...... 60

Control 1: Surface-Area-Only ...... 65

Control 2: Numerosity-Only ...... 67

Control 3: Surface Area versus Numerosity ...... 69

2.1.3 Discussion ...... 70

2.2 Rhesus monkeys map number onto space ...... 75

2.2.1 Methods ...... 81

Subjects ...... 81

Apparatus ...... 82

Training ...... 82

Ordinal Position Testing ...... 84

Number-Space Mapping Testing ...... 86

Hand Preference Assessment ...... 88

2.2.2 Results ...... 90

Ordinal Position Testing ...... 90

Number-Space Mapping Testing ...... 101

2.2.3 Discussion ...... 105

viii

2.3 The intraparietal sulcus may causally contribute to numerical processing: The impact of parietal rTMS on behavior and neuronal activity ...... 110

2.3.1 Experiment 1: rTMS and Behavior, Simultanous Presentation of Stimuli .. 122

Methods ...... 123

Results ...... 127

Discussion ...... 128

2.3.2 Experiment 2: rTMS and Behavior, Sequential Presentation of Stimuli ..... 130

Methods ...... 131

Discussion ...... 141

2.3.3 Experiment 3: rTMS and Electrophysiology ...... 144

Methods ...... 145

Results and Discussion ...... 147

2.3.4 General Discussion ...... 152

3. Competitive social decision-making ...... 157

3.1 Methods ...... 170

3.1.1 Subjects ...... 170

3.1.2 Task (Penalty Shot) ...... 171

3.1.3 Neurophysiological Recordings ...... 174

3.1.4 Pharmacological Inactivation ...... 176

3.1.5 Data Analysis ...... 177

Behavior During the Game ...... 177

Behavior During Goalie Replays ...... 178

Neuronal Firing Rates ...... 179

Muscimol Inactivation ...... 182

3.2 Results ...... 185

3.2.1 Behavior During the Game ...... 185

3.2.2 Behavior During Goalie Replays ...... 188

3.2.3 Neuronal Firing Rates ...... 191

3.2.4 Muscimol Inactivation ...... 205

3.3 Discussion ...... 214

4. General conclusions ...... 219

4.1 Perceptual and Abstract Decisions ...... 219

4.2 Value-Based and Social Decisions ...... 226

4.3 Manipulating Decision Circuits ...... 228

4.4 Conclusion ...... 231

Appendix A: Non-invasive primate head restraint using thermoplastic masks ...... 232

A.1 Introduction ...... 232

A.2 Materials and Methods ...... 235

A.2.1 Subjects ...... 236

A.2.2 Thermoplastic masks ...... 237

Fitting ...... 237

Training ...... 239

A.2.3 Head-post implantation ...... 240

Non-footed head-post ...... 241

Footed head-post ...... 243

Recovery ...... 244

Training ...... 244

A.2.4 Eye-tracking ...... 245

A.2.5 Neuronal recording ...... 246

Recording chamber implantation ...... 246

Electrophysiological data collection & analysis ...... 247

A.2.6 Measurement of head movement ...... 249

A.2.7 Magnetic resonance imaging ...... 249

A.2.8 Calculation of cost ...... 250

A.3 Results ...... 251

A.3.1 Eye-tracking accuracy ...... 251

A.3.2 Neuronal recording stability ...... 253

A.3.3 Head movement ...... 254

A.3.4 MRI clarity ...... 257

A.3.5 TMS coil placement ...... 258

A.3.6 Cost to lab ...... 260

A.4 Discussion ...... 261

A.5 Conclusion ...... 264

References ...... 265

Biography ...... 318

List of Tables

Table 1: Ratio training stimulus array numerosities...... 50

Table 2: Numerical ratio-of-ratio probe stimulus array numerosities...... 53

Table 3: Surface-area-only probe stimulus array surface areas (given in pixels)...... 56

Table 4: Surface Area versus Numerosity probe stimulus array numerosities...... 58

Table 5: Surface Area versus Numerosity probe stimulus array surface areas, in pixels. Here “correct” refers to the array with the greater surface area ratio-of-ratios, and congruence and favorability conditions relate to surface area...... 59

Table 6: Individual subjects’ performance during Standard trials on the first day of testing. “First” through “Fifth” refer to stimulus locations in the array ordered from the bottom of the screen to the top. p-values refer to one-tailed exact binomial tests compared to the 20% expectation...... 92

Table 7: Individual subjects’ performance during Standard trials over all ordinal testing sessions. “First” through “Fifth” refer to stimulus locations in the array ordered from the bottom of the screen to the top. p-values refer to one-tailed exact binomial tests compared to the 20% expectation...... 92

Table 8: Individual subjects’ performance during Red probe trials. “First” through “Fifth” refer to stimulus locations in the array ordered from bottom to top. p-values refer to one-tailed exact binomial tests compared to the 20% chance expectation...... 93

Table 9: Individual subjects’ performance during Blue probe trials. “First” through “Fifth” refer to stimulus locations in the array ordered from bottom to top. p-values refer to one-tailed exact binomial tests compared to the 20% chance expectation...... 93

Table 10: Individual subjects’ performance during Rectangle probe trials. “First” through “Fifth” refer to stimulus locations in the array ordered from bottom to top. p-values refer to one-tailed exact binomial tests compared to the 20% chance expectation...... 93

Table 11: Individual subjects' performance during Trangle probe trials. “First” through “Fifth” refer to stimulus locations in the array ordered from bottom to top. p-values refer to one-tailed exact binomial tests compared to the 20% chance expectation...... 93

xii

Table 12: Individual subjects’ performance during Shifted Up probe trials. “First” through “Fifth” refer to stimulus locations in the array ordered from bottom to top. p- values refer to one-tailed exact binomial tests compared to the 20% chance expectation...... 95

Table 13: Individual subjects’ performance during Shifted Down probe trials. “First” through “Fifth” refer to stimulus locations in the array ordered from bottom to top. p- values refer to one-tailed exact binomial tests compared to the 20% chance expectation...... 95

Table 14: Individual subjects’ performance during Sparse probe trials. “First” through “Fifth” refer to stimulus locations in the array ordered from bottom to top. p-values refer to one-tailed exact binomial tests compared to the 20% chance expectation...... 96

Table 15: Individual subjects’ performance during Dense probe trials. “First” through “Fifth” refer to stimulus locations in the array ordered from bottom to top. p-values refer to one-tailed exact binomial tests compared to the 20% chance expectation...... 96

Table 16: Individual subjects’ performance during Four probe trials. “First” through “Fourth” refer to stimulus locations in the array ordered from bottom to top. p-values refer to one-tailed exact binomial tests compared to the 25% chance expectation...... 99

Table 17: Individual subjects’ performance during Six probe trials. “First” through “Sixth” refer to stimulus locations in the array ordered from bottom to top. p-values refer to one-tailed exact binomial tests compared to the 16.7% chance expectation...... 99

Table 18: Individual subjects’ performance during Standard (Vertical) trials during the Horizontal test condition. “First” through “Fifth” refer to stimulus locations in the array ordered from bottom to top. p-values refer to one-tailed exact binomial tests compared to the 20% chance expectation...... 102

Table 19: Individual subjects’ performance during Horizontal probe trials. “First” through “Fifth” refer to stimulus locations in the array ordered from left to right. p- values refer to one-tailed exact binomial tests compared to the 20% chance expectation...... 102

xiii

List of Figures

Figure 1: General task structure and shape training. Subjects started each trial by touching a red rectangle at the bottom of the screen (left). Then the decision stimuli appeared (middle). If the subject chose S+ (in this case, the black circle), the screen changed color for a short inter-trial interval (top-right), a positive sound played, and a candy was delivered. If she chose S- (in this case, the white diamond), the screen changed color for a longer inter-trial interval (bottom-right), a negative sound played, and no candy was delivered...... 47

Figure 2: Example stimuli presented during ratio training. In these examples, S+ is the black circle, and the correct array is on the left. One example pair of arrays is shown for each of the six congruence X favorability conditions...... 49

Figure 3: Example stimuli presented during Surface-Area-Only probe testing. In these examples, S+ is the black circle, and the correct array is on the left. One example pair of arrays is shown for each of the six congruence X favorability conditions. Note that here congruence and favorability refer to surface area rather than numerosity...... 55

Figure 4: Example stimuli presented during Numerosity-Only probe testing. In these examples, S+ is the black circle, and the correct array is on the left. One example pair of arrays is shown for each of the six congruence X favorability conditions...... 57

Figure 5: Example stimuli presented during Surface Area versus Numerosity probe testing. In these examples, S+ is the black circle, the array with the greater numerical ratio of S+ to S- is on the left, and the array with the greater surface area ratio of S+ to S- is on the right...... 59

Figure 6: Performance during numerical ratio-of-ratios testing. Error bars are standard error of the mean. Black points show the nondifferentially reinforced probe trials; gray points show the differentially reinforced standard trials...... 63

Figure 7: Response time during numerical ratio-of-ratios testing. Error bars are standard error of the mean. Black points show the nondifferentially reinforced probe trials; gray points show the differentially reinforced standard trials...... 65

Figure 8: Subjects’ performance during Surface-Area-Only probe trials. The left plot shows mean accuracy at each surface area ratio-of-ratios, and the right plot shows mean response time at each surface area ratio-of-ratios (error bars are standard error of the mean)...... 67

xiv

Figure 9: Impact of individual item size on subjects’ choices during Numerosity-Only probe trials. Solid lines represent items on the left side, and dashed lines represent items on the right side. Darker shades represent S+ and lighter shades represent S-. As item radius on the left increases for either S+ or S- (solid lines), the proportion of leftward choices increases. Conversely, as item radius on the right increases for either S+ or S- (dashed lines), the proportion of leftward choices decreases, and consequently the proportion of rightward choices increases...... 69

Figure 10: Stimuli presented during ordinal position testing. a) Final phase of training, and standard trials during testing. Subjects were rewarded for selecting the fourth position from the bottom (second from the top). b) Red, c) Blue, d) Rectangle, e) Triangle, f) Up, g) Down, h) Sparse, i) Dense, j) Four, and k) Six probe trials, in which subjects were nondifferentially rewarded...... 84

Figure 11: Stimuli presented during Number-Space Mapping Testing. a) Standard and retraining trials, in which subjects were rewarded for selecting the fourth position from the bottom (second from the top). b) Horizontal probe trials, in which subjects were nondifferentially rewarded. If subjects chose the fourth position from the left (second from the right), this would suggest a left-to-right spatial mapping of number; if they chose the fourth position from the right (second from the left), this would suggest a right-to-left spatial mapping of number...... 88

Figure 12: Performance in the Color condition. Dashed line represents chance expectation. Error bars represent standard error of the mean...... 91

Figure 13: Performance in the Shape condition. Dashed line represents chance expectation. Error bars represent standard error of the mean...... 92

Figure 14: Performance in the Shift condition. Dashed line represents chance expectation. Error bars represent standard error of the mean...... 95

Figure 15: Performance in the Spaced condition. Dashed line represents chance expectation. Error bars represent standard error of the mean...... 96

Figure 16: Performance in the Length condition. Dashed lines represent chance expectation for each trial type. Error bars represent standard error of the mean...... 99

Figure 17: Mean response latency to each position during Standard trials of ordinal position testing...... 101

Figure 18: Performance in Number-Space Mapping Testing. Dashed line represents chance expectation. Error bars represent standard error of the mean...... 102

Figure 19: Custom chamber-centric TMS coil, showing the direction of current around each wing (top) and the induced electric field relative to the threshold for neuronal activation (bottom) in the active (left) and sham (right) configurations. Adapted from Mueller et al. (2014) with permission from the publisher...... 122

Figure 20: Simultaneous-presentation interleaved number and color comparison tasks, performed on a touch-screen. See text for additional description...... 125

Figure 21: Choice accuracy and response time in the simultaneous-presentation number and color comparison tasks following low-frequency offline rTMS...... 128

Figure 22: MRI of Monkey F showing the intraparietal sulcus (within green circle) and the implanted recording chamber (white cylinder above the green circle)...... 131

Figure 23: Interleaved sequential-presentation number and color comparison task. See text for additional description...... 134

Figure 24: Accuracy in the number and color tasks as a function of trial difficulty (4 being the hardest) following treatment. Colors represent the different treatments (see legend), points show actual data and lines show fits from the GLM, and the different plots show performance in each task early and late in the session for trials where the left or right stimulus was correct (see titles)...... 137

Figure 25: Mean response time in the number and color tasks as a function of trial difficulty (4 being the hardest) following treatment. Colors represent the different treatments (see legend), points show actual data and lines show fits from the GLM, and the different plots show performance in each task early and late in the session for trials where the left or right stimulus was correct (see titles)...... 138

Figure 26: Location of subject's touch to left and right targets following treatment early in the session (top) and late in the session (bottom). Each x represents one trial; colored circles represent actual means; and colored squares represent fits from the GLM (see legend)...... 140

Figure 27: Electric field induced by custom-built chamber-centric TMS coil at 100% stimulator output, according to a model (solid line) and measurements in a spherical

xvi

saline flask (dotted line). Adapted from Mueller et al. (2014) with permission from the publisher...... 143

Figure 28: MRI of Monkey S showing the intraparietal sulcus (within green circle) and the implanted recording chamber (white cylinder above the green circle)...... 145

Figure 29: Multi-unit activity following active (red) and sham (blue) low-frequency rTMS, averaged across sessions...... 148

Figure 30: Single-unit activity following active (red) and sham (blue) low-frequency rTMS, averaged across sessions...... 148

Figure 31: Multi-unit activity before and after low-frequency stimulation in individual sessions. Blue plots show active rTMS; red plots show sham rTMS. Plots are organized from the earliest session date (top left) to the latest session date (bottom right). Time in the session is plotted on the horizontal axis, and mean normalized firing rate is plotted on the vertical axis. Numbers at the top of each plot indicate the depth, in millimeters, of the electrode contact below the cortical surface...... 149

Figure 32: Single-unit activity before and after low-frequency stimulation in individual sessions. Blue plots show active rTMS; red plots show sham rTMS. Plots are organized from the earliest session date (top left) to the latest session date (bottom right). Time in the session is plotted on the horizontal axis, and mean normalized firing rate is plotted on the vertical axis. Numbers at the top of each plot indicate the depth, in millimeters, of the electrode contact below the cortical surface...... 150

Figure 33: Difference spectrograms between active and sham low-frequency rTMS, following treatment. The left plot shows the difference in the raw data; the right plot shows the difference in the data normalized to each session’s pre-stimulation average...... 151

Figure 34: Mean normalized power in each frequency band (colors as in legend) before and after active (top) and sham (bottom) low-frequency rTMS...... 152

Figure 35: Schematic of the Penalty Shot game. The two subjects viewed the same image on separate screens. One subject (at left here) played as the "shooter" and controlled the ball, while the other (at bottom here) played as the "goalie" and controlled the bar...... 173

xvii

Figure 36: Coronal MRI showing locations of DL-PFC and DM-PFC recordings in Monkey O. Injections and additional recordings were completed in analogous areas of the right hemisphere...... 175

Figure 37: Coronal MRI showing DL-PFC and DM-PFC injection and recording locations in Monkey Y...... 176

Figure 38: Behavior during standard game trials. Beta values are the coefficients resulting from a sliding-window GLM as described in the text predicting the ball’s final location on the goal-line from the shooter’s saccades (orange curve) or from the ball’s projection on the goal-line (blue curve, or predicting the position of the goalie bar from the shooter’s saccades (purple curve). Shaded areas represent standard error of the beta estimates...... 186

Figure 39: Heat maps showing relationship between the shooter's saccade endpoint and the final position of the ball on the goal-line during six windows before and during the trial. Bins on the x- and y-axis correspond to equally-sized segments of the screen. Colors indicate the number of trials in each bin...... 187

Figure 40: Relationship between projection of ball on the goal-line and its final position during the five sessions in which a shooter played against computer replays of prior goalie moves (“Replay sessions,” blue), the five live-opponent sessions prior to these ("Previous sessions," orange), and the five live-opponent sessions following these (“Subsequent sessions,” purple). Asterisks indicate time windows with a significant effect of session type. Shaded areas represent standard error of the beta estimates...... 190

Figure 41: Relationship between saccade end-point and final ball position during the five sessions in which a shooter played against computer replays of prior goalie moves (“Replay sessions,” blue), the five live-opponent sessions prior to these ("Previous sessions," orange), and the five live-opponent sessions following these (“Subsequent sessions,” purple). Asterisks indicate time windows with a significant effect of session type. Shaded areas represent standard error of the beta estimates...... 191

Figure 42: Timing and strength of the signal for Deception Index in all recorded cells of DL-PFC (top row) and DM-PFC (second row), as well as the proportion of cells in each region that were modulated by Deception Index (bottom row), when aligned to trial onset (left column) or offset (right column). Red colors in the heat-maps correspond to lower p-values, or greater modulation...... 194

xviii

Figure 43: Timing and strength of the signal for Goalie Position in all recorded cells of DL-PFC (top row) and DM-PFC (second row), as well as the proportion of cells in each region that were modulated by Goalie Position (bottom row), when aligned to trial onset (left column). Red colors in the heat-maps correspond to lower p-values, or greater modulation...... 196

Figure 44: Timing and strength of the signal for Trial Outcome (win/loss) in all recorded cells of DL-PFC (top row) and DM-PFC (second row), as well as the proportion of cells in each region that were modulated by Trial Outcome (bottom row), when aligned to trial onset (left column). Red colors in the heat-maps correspond to lower p-values, or greater modulation...... 198

Figure 45: Venn diagrams showing overlap in number of cells in DL-PFC (top) and DM- PFC (bottom), aligned to trial onset (left) and offset (right), significantly modulated by deception index, trial outcome (win/loss), and goalie position...... 199

Figure 46: Firing rate modulation by trial-wide Deception Index bin (top) and trial outcome (bottom) for a sample DL-PFC neuron. Shaded areas represent standard error of the mean...... 204

Figure 47: Firing rate modulation by trial-wide Deception Index bin (top) and trial outcome (bottom) for a sample DM-PFC neuron. Shaded areas represent standard error of the mean...... 204

Figure 48: Win rate following saline to the DL-PFC (red bar), muscimol to the DL-PFC (pink bar), saline to the DM-PFC (dark blue bar), and muscimol to the DM-PFC (light blue bar). Error bars represent 95% confidence intervals for the proportions...... 206

Figure 49: Win rate following saline to the DL-PFC (red bar), muscimol to the DL-PFC (pink bar), saline to the DM-PFC (dark blue bar), and muscimol to the DM-PFC (light blue bar), separated by subject (Monkey O on the left; Monkey Y on the right). Error bars represent 95% confidence intervals for the proportions...... 206

Figure 50: Win proportion over the course of the post-injection session. Shaded areas represent standard error of the beta estimates...... 207

Figure 51: Relationship between ball projection on the goal-line and final ball position on the goal following injections in the DL-PFC (left) and DM-PFC (right). Shaded areas represent standard error of the beta estimates...... 209

xix

Figure 52: Relationship between ball projection on goal-line and final position on the goal over the course of the post-injection session, in an early-trial window (250-450 ms, left) and a late-trial window (1050-1250 ms, right), for DL-PFC (top) and DM-PFC (bottom). Shaded areas represent standard error of the beta estimates...... 210

Figure 53: Relationship between saccade end-point and final ball position on the goal following injections in the DL-PFC (left) and DM-PFC (right). Shaded areas represent standard error of the beta estimates...... 212

Figure 54: Relationship between saccade end-point and and final ball position on the goal over the course of the post-injection session, in a pre-trial window (1000-800 ms prior to trial onset, left), a window at the start of the trial (0-200 ms, center), and a mid- trial window (800-1000 ms) for DL-PFC (top) and DM-PFC (bottom). Shaded areas represent standard error of the beta estimates...... 213

Figure 55: Fitting and applying a thermoplastic mask. First the mask is submerged in hot water (a). Then a hole is widened for the animal’s nose and mouth to enable breathing (b). Next the mask is stretched and molded over the sedated animal’s head while she is propped up in the chair used during experiments (c). After cutting holes for the monkey’s eyes, nose, and mouth, the mask is placed on the awake animal’s head for day-to-day use as shown in (d), (e), and (f)...... 239

Figure 56: Eye position traces during a saccade task. Monkeys were required to fixate centrally until a peripheral target appeared at one of eight locations, and then make a saccade to the target. Red and blue curves show the saccade trajectories during the first 120 trials of the session for the mask and post monkey, respectively...... 252

Figure 57: Drift in fixation position. Each point shows the average position of gaze on each trial in the horizontal (left) and vertical (right) direction for the mask monkey (red) and the post monkey (blue). Lines fit by linear regression...... 253

Figure 58: Neuronal recording stability. Sample voltage traces of action potentials recorded from a post monkey (left) and a mask monkey (right) are shown over 10 seconds at the beginning of a session (top), and 50 minutes into the session (bottom). Data were collected and spikes were sorted in Plexon...... 254

around the “roll” axis, the middle row shows rotation around the “pitch” axis, and the bottom row shows rotation around the “yaw” axis...... 255

Figure 60: Power spectra of head movement. Following a fast Fourier transformation, the power spectra of head rotational velocity in each of the three axes – roll (top), pitch (middle), and yaw (bottom) – is shown for two mask monkeys (left column), one calm (brighter color) and one restless (darker color), and two post monkeys (right column), one calm and one restless. These plots represent the same data as in the previous figure...... 256

Figure 61: Magnetic resonance images. Sample MRI images from head-post monkey (left), and a mask monkey (right). Coronal sections (top) and sagittal sections (bottom) are shown through the center of each monkey’s recording cylinder over parietal cortex. Red arrow in bottom-left image indicates shadow cast by post...... 258

Figure 62: TMS coil placement. The chair set-up for a head-post (left) partially blocks the movement of the TMS coil around the head, whereas the chair set-up for a mask (right) enables much more latitude in coil placement. Red arrows indicate TMS coil...... 260

xxi

List of Abbreviations

ACC Anterior cingulate cortex

ANOVA Analysis of variance

ANS Approximate number system

BOLD Blood-oxygen-level dependent

DI Deception index

DL-PFC Dorsolateral prefrontal cortex

DM-PFC Dorsomedial prefrontal cortex

EEG Electroencephalography

FEF Frontal eye field

FF Favorable-favorable

fMRI Functional magnetic resonance imaging

GLM General linear model

IPS Intraparietal sulcus

ITC Inferior temporal cortex

LFP Local field potential

LIP Lateral intraparietal area

MRI Magnetic resonance image

MT Middle temporal area

MUA Multi-unit activity

xxii

OFC Orbitofrontal cortex

PCC Posterior cingulate ocrtex

PFC Prefrontal cortex

PPC Posterior parietal cortex

RDM Random dot motion

RF Receptive field rTMS Repetitive transcranial magnetic stimulation

S+ Positively-reinforced stimulus

S- Negatively-reinforced stimulus

SE(M) Standard error (of the mean)

SNARC Spatial-numerical association of response codes tDCS Transcranial direct current stimulation

TMS Transcranial magnetic stimulation

ToM Theory of mind

TPJ Temporoparietal junction

UF Unfavorable-favorable

UU Unfavorable-unfavorable

VIP Ventral intraparietal area

VM-PFC Ventromedial prefrontal cortex

VTF Vibrotactile frequency

xxiii

Acknowledgements

My advisors and committee members – Drs. Michael Platt, Liz Brannon, Marc

Sommer, and Warren Grill – have been wonderfully supportive of me and have fostered my development as a scientist. They and others have collaborated on the projects that form the basis of the chapters in this dissertation. Dr. Brannon, an undergraduate named

Marley A. Rossa, and I designed the ratio experiment together; Marley, other research assistants, and I collected the data; I analyzed the data; and I wrote the manuscript with their valuable input. Dr. Brannon also advised me on the spatial-numerical association study, and helped me put together that manuscript. All four of my committee members have guided the experiments using transcranial magnetic stimulation. The study on strategic social decision-making was designed by a post-doctoral associate, Jean-

Francois Gariépy, in Dr. Platt’s lab; he collected the data, which I analyzed and synthesized into a manuscript. Part of the analysis relied on timing of moves derived from a behavioral model developed by Dr. Jeff Beck. The thermoplastic mask method for non-invasive primate head restraint described in the appendix was based on a design by

Dr. Chris Machado, and honing the technique for our purposes was a joint effort in Dr.

Platt’s lab among myself, Monica L. Carlson, Dr. Nicholas K. DeWind, and Dr. Koji

Toda.

The research presented here would not have been possible without the funding provided by the Holland-Trice Graduate Fellowship in Brain Science and Disease and

xxiv

the National Science Foundation Graduate Research Fellowship Program under Grant

No. 1106401 to me; NIH Grant 5R01-EY-019303-03 and the James McDonnell Scholar

Award to Dr. Brannon which funded the monkey number-space mapping study and the monkey ratio comparison study; a Duke Institute for Brain Sciences Incubator Award and NIH Grants 1R1-NS-078687-01 and 1R01-NS088674-01 to Drs. Platt, Sommer, Grill, and others which funded the transcranial magnetic stimulation studies and, in part, the development of the thermoplastic mask method; the JSPS Postdoctoral Fellowship for

Research Abroad to Koji Toda, NIH grant 1R01-MH-095894-01 and the Simons

Foundation Autism Research Award number 304935 to Dr. Platt and others which also funded the development of the thermoplastic mask method; NIH grant R01MH095894 and Department of Defense grant W81XWH-11-1-0584 to Dr. Platt and Dr. Steve W.C.

Chang (who also helped design the “penalty shot” experiments), NIH grant

K99MH099093 to Dr. Chang, and the Fonds de la Recherche en Santé du Québec grant

25559 and the Canadian Institutes for Health Research grant 84765 to Dr. Gariépy, which funded the “penalty shot” study.

Working with non-human primates is a team effort, and I am grateful to all the members of the Platt and Brannon labs, as well as the veterinary and technical staff at

Duke, who helped make that possible. I am also grateful for their friendship and support, as well as that of the broader neuroscience community at Duke, both in and out of the lab.

xxv

Finally, I would not have made it to this point without the love, encouragement, and support of my family.

1. Introduction

Every action we take results from a decision, conscious or not. Some decisions reflect our sensory perceptions: is the traffic light red or green? Others involve our personal valuation systems: which flavor of ice cream do we want? Should we donate to a charity? Decisions can occur in various contexts, and can be much more complex than these; nonetheless, a decision in general can be defined as a deliberative process that results in the commitment to a categorical proposition (Gold & Shadlen, 2007). Impaired decision-making processes can be problematic for individuals and for society. For these reasons, and because decision-making plays a role in everything we do, there is great interest in uncovering the neuronal mechanisms guiding decision-making. In this dissertation, I will present experiments that clarify the cognitive and neurobiological processes involved in various types of decisions. As background, in this chapter I review what is known about the neural basis of two fundamental types of decisions— perceptual and value-based—as well as what has been uncovered about decisions in more abstract or social contexts, relevant to the experiments presented here, and introduce the present study questions.

1.1 Perceptual Decision-Making

The systematic study of how the brain produces perceptual decisions began with work by Bill Newsome and colleagues in the late 1980s. They showed monkeys random

dot motion (RDM) stimuli, in which dots in a cloud moved in various directions. The monkeys had to report, using an eye movement to one of two lights, in which of two opposite directions the majority of the dots were moving. When very few of the dots moved coherently in one direction, the monkeys’ behavioral performance was at chance level; that is, they randomly guessed which light to look at. But as the percentage of coherently-moving dots increased, the monkeys’ performance also improved

(Newsome, Britten, & Movshon, 1989).

Simultaneously, Newsome and colleagues recorded the action potentials of individual neurons in the subjects’ middle temporal visual area (MT/V5), a region of extrastriate visual cortex sensitive to the direction and speed of motion (Maunsell & Van

Essen, 1983; Zeki, 1974). They found that MT neurons fired more strongly when a majority of dots moved in a given neuron’s preferred direction, and the difference between a single neuron’s firing for motion in the preferred and anti-preferred directions was greater the higher the coherence among dots. The researchers postulated that a decision about the direction of motion could be represented by the difference between a single neuron’s firing rate and that of its “antineuron” preferring the opposite direction of motion (Newsome et al., 1989).

Because the monkeys’ behavioral choice corresponded only modestly to the activity of any individual neuron but strongly to the activity of the population of recorded neurons, this hypothesis was later revised to the pooling of information across

many direction-sensitive neurons to form a decision (Britten, Newsome, Shadlen,

Celebrini, & Movshon, 1996; Britten, Shadlen, & Newsome, 1992; Shadlen, Britten,

Newsome, & Movshon, 1996). Importantly, neuronal activity predicted the monkey’s choice on error trials when the monkey chose the incorrect direction, and even when there was no coherence among the dots (i.e. all dots moved in different directions, and there was no objectively correct choice), demonstrating that the neuronal population’s activity did not simply reflect the stimulus, but related to the subject’s decision.

Moreover, chemical lesions of MT impaired subjects’ ability to perform the RDM task

(Newsome & Pare, 1988), while electrical microstimulation of MT neurons biased the subject’s decision toward the preferred direction of the stimulated neurons (Ditterich,

Mazurek, & Shadlen, 2003; Salzman, Murasugi, & Britten, 1992; Salzman & Newsome,

1994), confirming the functional link between activity of these neurons and the animal’s choice.

The middle superior temporal visual area (MST/V5a) receive inputs from MT and also shows motion sensitivity but with larger receptive fields (Desimone &

Ungerleider, 1986; Felleman & Van Essen, 1991). Neurons in this area showed similar response properties to those in MT when subjects performed the RDM task (Celebrini &

Newsome, 1994), and microstimulation of MST biased subjects’ behavior in much the same way as stimulation of MT (Celebrini & Newsome, 1995).

While the above evidence indicates that MT and MST play a role in representing the subject’s perception of motion, it does not speak to how the decision is instantiated.

Since the monkeys report their decisions with an eye movement, the search for such a signal began in brain areas related to generating eye movements. For example, during the RDM task, the superior colliculus, which participates in the generation of eye movements (Mays & Sparks, 1980; Wurtz & Goldberg, 1971), was found to contain neurons that—rather than strictly representing motor signals related to the parameters of the forthcoming saccade—were selective for motion direction, increasing their activity gradually when motion coherence was low and more quickly when motion coherence was high (Horwitz & Newsome, 1999; 2001). Microstimulation of the frontal eye field

(FEF), which also participates in the preparation of saccades (Bruce & Goldberg, 1985), suggested that the oculomotor decision was represented in this area as well (Gold &

Shadlen, 2000). Similarly, the activity of neurons in the lateral intraparietal area (LIP)— another area involved in planning eye movements, which receives inputs from areas MT and MST and in turn projects to the SC and FEF (Andersen, Asanuma, Essick, & Siegel,

1990; Andersen, Brotchie, & Mazzoni, 1992)—related not only to the direction of the forthcoming saccade, but also to the strength of the sensory signal: the more coherent the stimulus, the more quickly the neuronal firing rate increased, and the higher the firing rate remained through the delay period before a saccade was required (Shadlen &

Newsome, 1996; 2001).

To clarify how neuronal activity corresponds to the decision process, the RDM task paradigm was modified so that, rather than waiting through a delay period after the RDM stimuli disappeared, subjects were free to indicate their choice at any point during the stimulus presentation. Since correct choices resulted in a reward, subjects were motivated to respond as quickly as possible. This enabled researchers to identify the point in time when subjects committed to a decision. In this “reaction time” (RT) version of the task, not only did choice accuracy increase, but choice speed also increased (i.e. RT decreased) as motion coherence increased. The firing rates of LIP neurons recorded during this task increased when saccades to indicate the subject’s choice were directed into the neuron’s receptive field, as expected from previous studies, and firing rates increased more rapidly during strong motion coherence. More interesting was that, regardless of motion coherence, firing rates always reached the same level roughly fifty milliseconds before the saccade (Roitman & Shadlen, 2002).

These results in LIP supported a sequential-sampling model of decision-making

(Gold & Shadlen, 2007; P. L. Smith & Ratcliff, 2004). In the drift diffusion or race model, neuronal activity in sensory areas represents momentary evidence for or against a given choice. Another brain area accumulates this evidence, with firing rates reflecting the integral of the evidence for and against a decision over time. A threshold level of activity in this accumulation area would represent a commitment to a decision. Activity in this latter area would then represent a “decision variable.” In the case of the RDM task,

sensory evidence was represented in areas MT and MST, while the accumulation of evidence occurred in area LIP. Indeed, by introducing temporary noise into the RDM stimuli, it was shown that neurons in LIP do integrate the sensory evidence over time

(Huk & Shadlen, 2005).

Additional support for this model came from a microstimulation study during the RT version of the RDM task (Hanks, Ditterich, & Shadlen, 2006). Stimulation of both

MT and LIP biased choices toward the stimulated neurons’ receptive fields (RF) and decreased RTs for choices toward the RF while increasing RTs for choices away from the

RF. However, the overall shape of these changes differed between the two areas, and a computational model indicated that the changes in MT were best explained by a change in the momentary evidence and not the accumulated evidence, whereas the changes in

LIP were best explained by a change in the accumulated evidence.

Further experiments using the RDM task have elucidated various details of the decision process. For example, the process of accumulating evidence incurs a cost that increases with time in a trial for both monkeys and humans (Drugowitsch, Moreno-Bote,

Churchland, Shadlen, & Pouget, 2012), and changing the task regime to favor either high-speed or high-accuracy decisions revealed that changes in speed-accuracy tradeoff are accompanied by changes in firing rate of LIP neurons from the beginning of decision formation, rather than changes in the decision threshold (Hanks, Kiani, & Shadlen,

2014). Moreover, by including a lower-reward “opt out” choice which subjects could

sometimes select after their motion-direction decision to be guaranteed a reward when they did not feel certain about their decision, researchers were able to determine that the same LIP neurons representing the decision variable also encode the animal’s confidence in its decision (Fetsch, Kiani, Newsome, & Shadlen, 2014; Kiani & Shadlen, 2009).

Other brain regions besides LIP have been associated with a decision variable.

While LIP neurons seem to track the accumulation of sensory evidence whether the animal’s choice must be communicated via an eye movement or an arm movement, neurons in the nearby medial intraparietal area (MIP)—which signal the location of potential reach targets, as well as the intention to move the arm to a particular location

(Batista, Buneo, Snyder, & Andersen, 1999)—represented a decision variable more strongly when choices were communicated via a reach rather than a saccade (de

Lafuente, Jazayeri, & Shadlen, 2015). Some FEF neurons also encoded the decision variable when choices must be communicated via a saccade, while other FEF neurons were modulated by motion strength before and after the subject had committed to a decision (L. Ding & Gold, 2012), suggesting that this region may help to establish a memory trace that enables updating of choice values following the reward outcome. The dorsolateral prefrontal cortex (DL-PFC) also represents aspects of the decision in this task (Kim & Shadlen, 1999), and population dynamics in the PFC have revealed mechanisms of changes-of-mind (Kiani, Cueva, Reppas, & Newsome, 2014) and context- dependent decisions (Mante, Sussillo, Shenoy, & Newsome, 2013).

The drift diffusion model has been extended to account for changes of mind

(Albantakis & Deco, 2011; Resulaj, Kiani, Wolpert, & Shadlen, 2009) and decisions among more than two alternatives (Churchland & Ditterich, 2012; Churchland, Kiani, &

Shadlen, 2008). Additionally, variations on the simple linear drift diffusion model have been proposed, such as biophysically realistic nonlinear attractor models (Brunel &

Wang, 2001; Wong & Wang, 2006; Wong, Huk, & Shadlen, 2007) and the incorporation of firing rate variance as an important aspect of neuronal computation during decision- making (Churchland et al., 2011). However, note that some authors have argued that on individual trials, single neurons in LIP show discrete step changes in their firing rates during the RDM task, and the ramping pattern of activity only emerges when responses are averaged across trials (Latimer, Yates, Meister, Huk, & Pillow, 2015).

In parallel to the development of the RDM task to probe neural mechanisms of a visual decision, somatosensory decisions were being investigated using a vibrotactile frequency (VTF) paradigm originally developed by Vernon Mountcastle and colleagues in the 1960s. They applied brief mechanical vibrations to the hands of humans and monkeys, and first determined—based on human detection thresholds and recordings from peripheral nerve fibers in monkeys—that two distinct systems subserve the perception of mechanical vibrations: one system involving the tactile or Meissner corpuscles detects low-frequency (5-50Hz) vibrations, while another system involving the lamellar or Pacinian corpuscles detects high-frequency (50-400Hz) vibrations

(Mountcastle, Talbot, Darian-Smith, & Kornhuber, 1967). Next they showed that distinct groups of cells in somatosensory cortex (postcentral gyrus) respond selectively to either low- or high-frequency vibrations, with the low-frequency cells using a rate code—that is, the vibration frequency was indicated by the firing rate of cells—while the high- frequency cells used a place or “labeled line” code, where different neurons were tuned to different frequencies (Hyvarinen, Sakata, Talbot, & Mountcastle, 1968).

Beyond detecting various frequencies, subjects could also compare two frequencies, f1 and f2, presented in sequence by pressing a button with the non- stimulated hand to indicate whether the first or second frequency was higher. The discrimination curves—how often the subject reported that each presented frequency was greater than a reference frequency—were similar in monkeys and humans (LaMotte

& Mountcastle, 1975). In both species, discrimination followed Weber’s law: the amount of change needed to detect a difference between two stimuli was proportional to the magnitude of the frequencies. Additionally, unilateral lesions of monkey primary and secondary somatosensory cortex destroyed this discrimination ability for vibrations presented to the contralateral hand (LaMotte & Mountcastle, 1979), indicating that neurons within these regions are necessary for some aspect of the decision process in this task.

Investigations into how the brain achieves the VTF comparison thus began in primary somatosensory cortex (S1). In an initial study, periodicity in S1 neuronal firing

was found to correspond to the period of the vibratory stimulus (Mountcastle,

Steinmetz, & Romo, 1990). This modulation occurred whether the subject was engaged in the comparison task or was passively presented with VTF stimuli, suggesting that S1 is involved in perceiving somatic frequencies, but not necessarily in comparing them.

Similarly, neurons in owl monkey S1 (Brodmann area (BA) 3b), but not in BA 3a, exhibited lower variance in their periodicity during VTF presentation with improved behavioral performance, i.e. better discrimination was associated with decreased neuronal cycle variance (Recanzone, Merzenich, & Schreiner, 1992). Moreover, in both of these studies, firing rates of S1 neurons were not found to vary substantially with stimulus frequency.

However, the lab of Ranulfo Romo questioned whether periodicity in interspike intervals, and not firing rate, encoded VTF stimuli. They first observed that when monkeys were trained with a constant base stimulus (i.e. f1 never changed), as had been the case during both training and testing in previous studies, they failed at tests where the base stimulus varied; in other words, they simply categorized f2 as “low” or “high” rather than comparing it with f1. However, when they were trained with a varying f1, they successfully performed the comparison between f2 and f1 (Hernandez, Salinas,

Garcı́a, & Romo, 1997). Next the researchers inserted microelectrodes into S1 (BA 3b) and replaced the second VTF stimulus with a sequence of injected current pulses at a particular frequency. Monkeys showed similar behavioral performance in

microstimulation trials as in trials where both f1 and f2 were presented mechanically.

Moreover, this held true whether pulses were delivered periodically or aperiodically, suggesting that the overall rate—rather than periodicity—of S1 neurons indicates frequency (Romo, Hernández, Zainos, & Salinas, 1998). Indeed, when recording from neurons in S1 and secondary somatosensory cortex (S2), they found that firing rate modulations in both areas were enhanced when frequency perception was relevant for task performance, and—unlike periodicity—correlated with task performance (Salinas,

Hernandez, Zainos, & Romo, 2000). Indeed, neurometric thresholds based on firing rates in S1 were very similar to the animal’s psychometric threshold; that is, firing rates of S1 neurons could be used to predict subjects’ performance in the VTF task (Hernandez,

Zainos, & Romo, 2000). Thus, just as MT and MST encoded the sensory evidence in the visual RDM task, S1 and perhaps S2 encoded the sensory evidence in the somatosensory

VTF task.

A later experiment revealed that, while some S2 neurons encoded the frequency of each VTF stimulus, other S2 neurons had firing rates that were a function of both f1 and f2 during f2 presentation, and corresponded to the monkey’s decision following f2 presentation (Romo, Hernández, Zainos, Lemus, & Brody, 2002). This latter population of S2 neurons could thus be thought of as representing the decision. The PFC was found to encode the frequency of the first stimulus during the delay period, holding f1 in working memory for later comparison (Brody, Hernandez, Zainos, & Romo, 2003;

Romo, Brody, Hernández, & Lemus, 1999). Furthermore, the ventral premotor cortex

(VPC) encoded not only the current and remembered stimuli, but also their comparison, as well as motor commands to report the result of this comparison (Hernandez, Zainos,

& Romo, 2002; Romo, Hernández, & Zainos, 2004). Together these results indicate that

S1 encodes the current somatosensory stimulus, driving subsequent cortical areas, including S2, PFC, and VPC, to combine past and current sensory information and form a decision.

However, since the decision was always reported immediately, it was difficult to disentangle neuronal signals related to the commitment to a decision versus to a motor response. In contrast with the RDM task paradigm, where early studies required subjects to wait through a delay period between stimulus presentation and response and the RT version was only introduced later, the VTF task paradigm had essentially begun with an RT version. When researchers introduced a delay between stimulus presentation and the subject’s response, they were able not only to separate the motor signal from a comparison- or decision-related signal, but also to probe the nature of what is held in working memory between the stimuli and the response. When recording from the medial premotor cortex (MPC), including the presupplementary and the supplementary motor areas, they found that during the delay period between f2 and the motor response, some neurons coded a categorical choice—firing more strongly when f2>f1 or vice versa—whereas others carried information only about f1 or f2. Some neurons even

switched what they encoded over the course of the delay period. Moreover, activity of these neurons, even those that only encoded each individual stimulus and not their comparison, correlated with the subject’s eventual motor choice on both correct and error trials, indicating that they contribute to the decision (Lemus et al., 2007). The authors suggest that such a scheme, where the original pieces of sensory evidence and not just a decision about their comparison are maintained in working memory, could facilitate updating the decision following additional consideration or if more evidence should arise.

The above pattern of neuronal firing was not limited to MPC. In a subsequent study using the same task, Hernández and colleagues recorded not only from MPC, but also from S1, S2, PFC, VPC, and dorsal premotor cortex (DPC) contralateral to the stimulated finger, and PFC, VPC, DPC, and M1 contralateral to the responding hand.

They found that all areas had neurons that responded differentially during each stimulus presentation, and that all areas except S1 contained neurons that maintained f1 in memory during the delay between f1 and f2, and neurons that encoded f1, f2, or the decision about which was greater during and after f2 presentation. When VTF stimuli were delivered to the finger in the absence of the task, only neurons in S1 and S2 responded, suggesting that the signals observed broadly across cortex during the task everywhere but S1 were specific to the decision-making process (Hernandez et al., 2002).

In addition to single-unit firing, oscillatory activity has been associated with performance of the VTF discrimination task. Local field potentials (LFPs) recorded in S1, and to a lesser extent S2, but not those recorded in DPC, MPC, or M1, were entrained to the frequency of each stimulus during its presentation. During the delay before the decision, beta-band (15-30 Hz) power in all these areas corresponded to the subject’s choice on each trial, building up gradually in M1 and left MPC in advance of the response (Haegens et al., 2011a). Additionally, alpha-band (8-14 Hz) power decreased during the task in all areas, with the largest decrease occurring during stimulus presentation in S1 and S2, and during the decision delay in DPC, MPC, and M1. Firing rates correlated positively with alpha power in S1, but negatively in S2, MPC, and M1, and in all regions, alpha phase predicted neuronal spiking, with higher firing rates occurring at the trough of the alpha cycle and lower firing rates occurring at its peak.

Furthermore, alpha power on each trial correlated inversely with discrimination accuracy, indicating that it was relevant for task performance (Haegens, Nacher, Luna,

Romo, & Jensen, 2011b). Additionally, delta-band (1-4 Hz) coherence between parietal and frontal cortices was found to be high during the decision period, and its level was modulated by which decision the subject made (Nacher, Ledberg, Deco, & Romo, 2013).

Finally, modulations in beta- and theta-band (4-7 Hz) power were observed in the ventral posterolateral nucleus of the somatosensory thalamus during a VTF detection task, and this modulation was weakened during passive presentation of VTF stimuli

(Haegens et al., 2014). Together these results indicate that oscillatory activity across various regions of the cortex, and even a subcortical area, contribute to somatosensory decision-making.

Just as for the RDM task, which presented only one stimulus and did not require subjects to compare stimuli stored in working memory, various models have been proposed to explain not just where and when, but how, mechanistically, the sequential

VTF discrimination is performed. Mutual inhibition between an excitatory and an inhibitory population could enable the fast initial encoding of f1 into working memory, followed by a slowly-decaying maintenance in working memory, and then quick decision-making about whether f2 is greater than f2 (Machens, Romo, & Brody, 2005).

Alternatively, integral feedback control, which performs a subtraction over time by sending inhibition from an integrator, which stores the short-term memory of the stimulus, back to comparison neurons that receive sensory input, could also explain the observed neuronal results, and may have a plausible anatomical substrate in PFC neurons whose excitatory projections target inhibitory neurons in upstream cortical areas (P. Miller & Wang, 2006). As reviewed by Xiao-Jing Wang (2012), these and other recurrent network models can explain some but not all experimentally-observed aspects of perceptual decisions, and operate at a level of abstraction—population dynamics in state space—complementary to that of the drift-diffusion model, which involves ramping-to-threshold in the temporal domain, and which similarly captures some but

not all of observed behavior (Wang, 2012). Others have proposed that probabilistic population codes can implement optimal strategies in binary perceptual decisions, thus outperforming either the drift-diffusion or the attractor network model (Drugowitsch &

Pouget, 2012). However, such a model has less of a biological basis, and according to

Drugowitsch & Pouget (2012), experiments are in progress to test whether this theory accurately predicts neural activity in realistic situations where stimuli have varying reliability and subjects’ confidence may correspondingly vary.

Perceptual decisions can, of course, occur in any modality, and recent studies have extended beyond vision and somatosensation. For example, the anterolateral but not middle-lateral belt region of monkey auditory cortex has been found to contribute to the sensory evidence in an auditory decision-making task regarding whether a noisy stimulus contained more low- or high-frequency tone bursts (Tsunada, Liu, Gold, &

Cohen, 2015). This type of auditory evidence is likely converted into a decision in the

PFC (Russ, Orr, & Cohen, 2008).

Rodents are often preferred to monkeys as model organisms due to the lower costs—both financially and ethically—involved in maintaining them in the lab, and more importantly, the neuroscientific and genetic tools available for them. In recent years, perceptual decision-making tasks have been developed that rodents are capable of performing. For instance, rats presented with pulses of evidence over time were able to report whether more pulses had come from the left or the right (Brunton, Botvinick, &

Brody, 2013). (Interestingly, a computational model revealed that, in this paradigm, both rats and humans accumulated evidence noiselessly.) Recordings in rat frontal orienting fields (FOF) and posterior parietal cortex (PPC) subjected to tuning curve assays revealed that PPC neurons encoded the gradual accumulation of evidence, whereas FOF neurons encoded the presently-favored decision. Optogenetic silencing of FOF only influenced behavior when the silencing occurred at the end of the perceptual stimulus, supporting this area’s role in committing to a categorical choice (Hanks et al., 2015).

Moreover, pharmacological inactivation of FOF impaired performance in a manner most consistent with a disruption in the output of an evidence accumulator, whereas inactivation of the PPC did not impair auditory decisions, but did impair internally- guided decisions (Erlich et al., 2015). Together these results indicate that—as in monkeys—decisions are represented in a frontal region in rats, whereas evidence accumulation in some contexts occurs in posterior parietal regions.

This paradigm has been extended to examine multisensory decision-making in rats and humans. Both species were able to optimally combine visual and auditory event streams over time to make judgments about whether the event rate was high or low

(Raposo, Sheppard, Schrater, & Churchland, 2012; Sheppard, Raposo, & Churchland,

2013), suggesting that rats can serve as a useful model organism in studies of decision- making. Neuronal recordings in rats indicated that a dynamic network of neurons in

PPC entered different states during the decision and movement epochs (Raposo,

Kaufman, & Churchland, 2014). Mice have even been shown to perform as well as rats in an auditory decision-making task (Jaramillo & Zador, 2014), and they can also integrate auditory and visual cues to improve their decisions (Siemann et al., 2014), paving the way for detailed studies of the genes, neurons, and circuits involved in perceptual decisions (see also (Brody & Hanks, 2016).

On the other end of the spectrum, the emergence of non-invasive neuroimaging techniques over the past two decades has enabled the investigation of decision-making processes in humans, guided by the findings from electrophysiological studies in other animals. For example, Heekeren and colleagues took advantage of the fact that faces and houses activate distinct regions of temporal cortex (Epstein & Kanwisher, 1998; Ishai,

Ungerleider, Martin, Schouten, & Haxby, 1999; Kanwisher, McDermott, & Chun, 1997;

G. McCarthy, Puce, Gore, & Allison, 1997). They presented subjects in an fMRI scanner with low- or high-noise images of faces and houses, and subjects pressed a button after a delay to indicate whether the image was a face or a house. They found that activity in the left DL-PFC covaried with the difference in signal between the face- and house- selective regions of ventral temporal cortex, was greater when images had low noise— making decisions easier—rather than high noise, and predicted behavioral performance

(Heekeren, Marrett, Bandettini, & Ungerleider, 2004). The involvement of the DL-PFC was not limited to the face/house task: when subjects performed the RDM task, this region, as well as the left posterior cingulate cortex, left inferior parietal lobule, and left

fusiform/parahippocampal gyrus responded more to high- than low-coherence stimuli, regardless of whether subjects used a button press or a saccadic eye movement to indicate their choice (Heekeren, Marrett, Ruff, Bandettini, & Ungerleider, 2006).

Moreover, when subjects were asked to report the direction of RDM displays, activity in the human MT complex (hMT+) predicted the reported direction, regardless of whether motion direction was ambiguous (0% coherence) or unambiguous (50 or 100% coherence) (Serences & Boynton, 2007). These studies in the visual domain provided initial support for the idea that in humans, as in monkeys, sensory evidence may be represented in perceptual brain areas, whereas decisions are represented in the PFC.

Studies in the auditory and somatosensory domains also adhered to this pattern.

For example, when subjects identified speech sounds, the BOLD signal in an area adjacent to primary auditory cortex predicted accuracy, consistent with a role in representing the sensory evidence, while the signal in an inferior frontal region predicted response time, consistent with a role in forming the decision (Binder,

Liebenthal, Possing, Medler, & Ward, 2004). When subjects performed the VTF discrimination task, the relative frequency difference f2-f1 modulated activity changes in the left DL-PFC, the dorsal anterior cingulate cortex (dACC), and the anterior insula, but only in left DL-PFC was this modulation greater on correct than incorrect trials, suggesting that the DL-PFC is involved in veridical accumulation of sensory evidence

(Pleger et al., 2006).

As task and analysis paradigms have diversified, so have the regions implicated in various stages of perceptual decision-making. For instance, when different visual stimuli—faces or places, where images varied in their level of noise—were arbitrarily associated with saccadic eye or hand-pointing movements, regions of the medial parietal and posterior intra-parietal cortex integrated sensory evidence that led to a decision outcome (Tosoni, Galati, Romani, & Corbetta, 2008). An accumulator model predicted activity in the intraparietal sulcus (IPS) during the RDM task (Kayser, Buchsbaum,

Erickson, & D'Esposito, 2010), and when response modality was cued to subjects following a delay after RDM stimulus presentation, BOLD activity in the FEF, IPS, inferior frontal sulcus, and anterior insula increased with decreasing motion coherence

(T. Liu & Pleskac, 2011). At the same time, others have reported that separating motor preparation from the perceptual decision results only in signals related to the accumulation of evidence in the inferior frontal cortex (Filimon, Philiastides, Nelson,

Kloosterman, & Heekeren, 2013). In one study, activity only in a subregion of the right insula was found to have a timecourse consistent with the accumulation of sensory evidence when subjects responded in a two-color version of the RDM task either with a saccadic eye movement or with a manual button press (Ho, Brown, & Serences, 2009).

Nonetheless, most regions reported to reflect a decision variable were found in the prefrontal or posterior parietal cortices. Indeed, a recent review of fMRI studies taking a model-based approach to analyzing the contribution of different brain regions

to perceptual decision-making found that overall, the accumulation of evidence is associated with a fronto-parietal network, while adaptation and individual differences in decision threshold are associated with a fronto-basal ganglia network, and choice bias is associated with both networks (Mulder, Van Maanen, & Forstmann, 2014). Whereas fMRI provides high spatial resolution, electroencephalography (EEG) provides better temporal resolution, and studies using this technique have also confirmed that brain activity evolves in time in a manner consistent with the drift diffusion model (S. P. Kelly

& O'Connell, 2015). A pair of studies using continuous EEG monitoring and two different task paradigms—a gradual-change detection task (O'Connell, Dockree, &

Kelly, 2012) and the RDM task (S. P. Kelly & O'Connell, 2013)—even localized a neural instantiation of the decision variable to an EEG component known as the centro-parietal positivity (CPP). Altogether, results in humans to date have largely confirmed the predictions of sequential sampling models of perceptual decision-making.

1.2 Abstract Decision-Making

Most of the studies discussed so far have investigated decisions based strictly on perceptual features. However, many perceptual decisions are more abstract, requiring the transformation of sensory evidence into some other type of information reflecting the meaning or behavioral relevance of the stimuli. For example, when deciding whether a spherical orange-colored object would be good to eat, we need to somehow abstract

away from the purely visual features and mentally categorize the object as a fruit or a toy ball. In general, categorization is distinguished by the presence of sharp boundaries along a continuum, where stimuli on one side of the boundary are placed into one category, while stimuli on the other side of the boundary are placed into another (Ashby

& Maddox, 2005; Liberman, Cooper, Shankweiler, & Studdert-Kennedy, 1967; Liberman,

Harris, Kinney, & Lane, 1961; Wyttenbach, May, & Hoy, 1996). Two stimuli that lie very close together on the continuum will be considered different if they fall on opposite sides of a boundary, whereas stimuli that lie far apart will be considered the same if the fall in between boundaries. Classifications such as food versus non-food or friend versus foe are crucial to animals’ survival, so it is not surprising that other animals, including non-human primates and pigeons, are able to conceptualize visual stimuli into different categories (Fabre-Thorpe, Richard, & Thorpe, 1998; W. A. Roberts & Mazmanian, 1988;

Zentall, Wasserman, Lazareva, Thompson, & Rattermann, 2008).

The neural underpinnings of categorical decisions were first studied by

Freedman and colleagues (2001, 2002), who trained monkeys to classify images as cats or dogs. To disentangle abstract categorization from clustering based on physical similarity, they applied a three-dimensional morphing system to different images of cats and dogs in order to create novel stimuli that were linear combinations of the various original images. In this way the shape of animal varied continuously, and a category boundary between cats and dogs was defined so that images on one side of the

boundary could vary widely in their physical features, whereas images on opposite sides of the boundary could appear quite similar. Monkeys performed a delayed match- to-sample task in which a sample image appeared, and following a delay a test image appeared. If the test came from the same category as the sample, subjects had to release a lever to earn a reward; if the two images came from different categories, subjects had to continue holding the lever through another delay period, and release it during a subsequent test stimulus that matched the category of the sample. Subjects performed this task with high accuracy even when the test stimulus was a 60:40 mix of dog and cat, or vice versa.

The activity of many lateral PFC neurons recorded during the task was elevated during the sample or delay period. Importantly, these neurons exhibited different firing rates for stimuli from different categories, but similar firing rates for stimuli within categories. Moreover, when one monkey was retrained with two new category boundaries orthogonal to the original boundary, it readily learned to perform the three- category task, and PFC neurons showed categorical responses reflecting these new boundaries, and not the previously-learned boundary. Thus a sharp category boundary between sometimes perceptually similar cats and dogs, and also between arbitrarily- defined novel categories, was observed both in the animals’ behavior and in the responses of lateral PFC neurons. Additionally, the activity of many PFC neurons differentiated between “match” and “non-match” choices during the test stimulus,

indicating that both category information and the subject’s decision were represented in the lateral PFC (Freedman & Riesenhuber, 2002; Freedman, Riesenhuber, Poggio, &

Miller, 2001).

On the other hand, the inferior temporal cortex (ITC), which plays a role in visual shape processing and object identification (Desimone, Albright, Gross, & Bruce, 1984;

Hung, Kreiman, Poggio, & DiCarlo, 2005) and might be expected to play a role in image categorization, did not show category selectivity. These neurons also did not fire differentially during match and non-match decisions. Instead, neurons in ITC were tuned to particular stimuli, with decreasing responses as the image became less visibly similar to the neuron’s preferred stimulus (Freedman, Riesenhuber, Poggio, & Miller,

2003). Experience with particular exemplars of the cat-dog morphs sharpened ITC neurons’ tuning (Freedman & Assad, 2006), further confirming this area’s role in representing specific visual stimuli, but not categories.

To determine whether the posterior parietal cortex, and in particular area LIP— which, as discussed above, is involved in visually-guided action as well as perceptual decision-making, and also plays a role in spatial and shape-based attention (Sereno &

Amador, 2006)—participates in categorization, Freedman and colleagues used RDM stimuli where the dots could move primarily in any of twelve directions. They arbitrarily defined a boundary to divide these twelve directions into two categories, and trained monkeys to perform a delayed-match-to-category task on these motion direction

categories. Neurons in LIP showed strong category selectivity, with little variation in their firing to different directions within the same category. Moreover, when monkeys were retrained with a new category boundary, their behavior and the responses of LIP neurons shifted accordingly. Recordings in MT did not show this pattern; instead, MT neurons were strongly direction-selective (Freedman & Assad, 2006).

Overall these results point to similar modes of category processing in the

“dorsal” and “ventral” visual streams (Milner & Goodale, 2008; Mishkin, Ungerleider, &

Macko, 1983). In the ventral stream, as observed during dog/cat shape categorization,

ITC was sensitive to the low-level visual features of the images, whereas PFC—which receives input from ITC—was sensitive to the abstract category to which each image belonged. In the dorsal stream, as observed during motion direction categorization, MT was sensitive to the purely perceptual direction of motion, whereas LIP—which receives input from MT—encoded the category membership of each direction. This pattern suggests a division between visual feature encoding and the more abstract and learning- dependent encoding necessary for categorization (Freedman & Miller, 2008).

Further experiments confirmed that LIP plays a preeminent role in categorization. Although LIP neurons showed strong modulation by whether or not stimuli were presented in their receptive fields (RFs), during the dot-motion categorization task they also encoded category membership even when stimuli were presented outside their RFs (Freedman & Assad, 2009). Moreover, even though

responses were reported with an arm movement (lever release), LIP showed stronger and earlier category selectivity than MIP (Swaminathan, Masse, & Freedman, 2013), an area involved in arm movements, as noted above. This finding supports the hypothesis that LIP activity reflects the categorization process itself, rather than decision-related motor actions. Finally, decision-related signals were stronger, more reliable, and shorter- latency in LIP than PFC, indicating that LIP is independently involved in visual categorization and this signal does not reflect feedback from PFC (J. K. Fitzgerald,

Swaminathan, & Freedman, 2012; Swaminathan & Freedman, 2012).

An interesting combination of perceptual processing and abstract decision- making occurs during decisions about number: that is, how many items are there? Like many purely perceptual features, numerical discriminations follow Weber’s Law; that is, if distance between two numbers is held constant, the difficulty in discriminating them increases as their magnitude increases (Dehaene, Dehaene-Lambertz, & Cohen, 1998), or in other words, the discriminability between two stimuli depends on the ratio between them (Meck & Church, 1983). At the same time, numerosity is a concept that must be abstracted from perceptual stimuli. Four mugs of coffee and four elephants are quite distinct in the visual, olfactory, auditory, somatosensory, and presumably gustatory modalities, yet we can appreciate that they share the property of four-ness. Furthermore, number appears to be a core cognitive system, present across development and across the animal kingdom (Feigenson, Dehaene, & Spelke, 2004). For these reasons, numerical

discrimination provides a good test case for theories about the mechanisms of decision- making. How conserved are these mechanisms across species? Which brain areas causally contribute to which parts of an abstract decision? Can neuromodulation techniques improve such decisions? These questions will be addressed in section 2.

1.3 Value-Based Decision-Making

Whether choosing one of two options in an experimental task or choosing how to behave in everyday life, all decisions are guided by expectations about their outcome.

We behave in ways we think will yield some benefit. Deciding how to act is therefore thought to rely on representations of expected value for each available action. The form of this value may vary—primary reinforcers such as food, secondary reinforcers such as money, experiences such as “fun,” and even abstract concepts such as fairness can all carry value. Regardless, a rational actor should make decisions based upon which choice will deliver the greatest value.

Rangel and colleagues (2008) proposed a framework for value-based decision- making that is useful for guiding and interpreting neurobiological investigations. They posited that every choice involves five basic computations: a representation of the decision problem, in which internal and external states as well as the set of possible actions are identified; valuation of each different possible action; action selection, where a choice is made based on the relative values; outcome evaluation, where the desirability of

the outcome is measured; and finally, learning or updating the representation, valuation, and action-selection processes based on the feedback measures in order to improve future decisions. Additionally, they divided valuation systems into three distinct types:

Pavlovian systems that assign values to certain evolutionarily appropriate behaviors, such as approach, avoidance, and consumption, in response to environmental cues; habit systems in which values are assigned to any action through repeated training; and goal- directed systems in which values are assigned to actions based on the values of their outcomes in different contexts. Finally, they noted that various factors—including risk and uncertainty, delay between action and outcome, and social context—can modulate the values that each system assigns to actions (Rangel, Camerer, & Montague, 2008). The sequence of recognizing the current state of the environment, evaluating possible actions, choosing an action, and then updating the action values based on the outcome is also known as reinforcement learning (Ito & Doya, 2011).

The first stage in this sequence of computations—how the brain determines what the relevant internal and external states and possible actions are—is not perfectly understood, but certainly involves aspects of perception and perceptual decision- making, discussed above. The remaining stages have received a fair amount of study.

Valuation, according to one recent meta-analysis of 81 fMRI studies, predominantly involves the ventromedial prefrontal cortex (VM-PFC), ventral striatum, and posterior cingulate cortex (PCC), with a posterior-to-anterior gradient for the representation of

concrete-to-abstract rewards (Clithero & Rangel, 2014). Another meta-analysis of over

200 fMRI studies determined that, while one set of regions—including the anterior insula, dorsomedial prefrontal cortex (DM-PFC), dorsal and posterior striatum, and thalamus—could show either positive or negative modulation by subjective value, the

VM-PFC and ventral striatum consistently showed increased BOLD signal for positive subjective value. This correlation occurred both when options were presented to subjects, and when outcomes were delivered (Bartra, McGuire, & Kable, 2013). These analyses suggest that a network involving the VM-PFC (including the orbitofrontal cortex (OFC)) and ventral striatum is involved in both valuation and outcome evaluation. Action selection and learning, on the other hand, seem to rely on the lateral

PFC (reviewed in (Dixon & Christoff, 2014)).

Electrophysiology experiments in other animals have further clarified the neuronal mechanisms involved in value-based decision-making, although the areas implicated in each stage have been widespread throughout the brain. The earliest of these studies, work by Wolfram Schultz and colleagues, spoke to the outcome evaluation and learning stages. They argued that observations of learning in other animals could be explained by changes in their expectations about future rewards and punishments. Following extensive evidence that the neuromodulator dopamine plays a role in reward-dependent learning (Koob, 1992; Le Moal & Simon, 1991; Phillips, Brooke,

& Fibiger, 1975; Wise & Rompré, 1989), they recorded from midbrain dopaminergic

neurons in monkeys while delivering rewards. When the subject received an unexpected reward, these neurons increased their firing. After several pairings of a cue followed by a reward, the neurons began to fire in response to the cue, and returned to baseline during reward delivery. If the conditioned cue was delivered and no reward followed, these neurons decreased their firing at the time when a reward would have been expected. Dopaminergic neurons thus encoded what is known as a reward prediction error—the difference between the animal’s expectation of reward, and what it actually received (Schultz, Dayan, & Montague, 1997). Representing this quantity is central to reinforcement learning.

Such a representation relies on the animal’s ability to compute expected value.

The expected value of an action should be determined by the gain that may result from the action, combined with the probability that the action will result in that gain. These two variables do appear to play a role in guiding the behavior of many different animals

(Caraco, Martindale, & Whittam, 1980; Krebs, Kacelnik, & Taylor, 1978; Montague,

Dayan, Person, & Sejnowski, 1995; Schoener, 1971; Werner & MittelBach, 1981). Platt and

Glimcher (1999) hypothesized that neurons in area LIP, which prepares certain motor actions and also participates in perceptual decisions as described above, might compute expected value in order to guide choices. They required monkeys to fixate their gaze centrally on a screen, and then, following an instruction, make a saccadic eye movement to either a left or a right target. If subjects made a saccade to the correct target, they

could earn a juice reward, but if they made a saccade to the incorrect target, they did not receive anything. In different blocks of trials, the experimenters varied either the amount of juice delivered—that is, the expected gain—for a leftward or rightward saccade, or the probability that a leftward or rightward saccade would be instructed, i.e. the outcome probability. They found that a large proportion of LIP neurons were significantly modulated by expected gain or by outcome probability early in the trial, before the movement was instructed. Additionally, when there was no instructional cue and subjects were free to make a saccade to whichever target they wanted, they tended to match the proportion of choices to each target with the relative expected gain available from each target, and the firing rates of LIP neurons correlated with the expected value of saccades into their RFs (Platt & Glimcher, 1999). Thus the parietal cortex participates in the representation of expected value.

Since these seminal studies, reward-related signals have been found in multiple regions of the primate brain (reviewed in (Schultz, 2015)), including for example the amygdala (Bermudez & Schultz, 2010) and the PFC, where the DL-PFC may incorporate information about an animal’s past choices and their payoffs (Barraclough, Conroy, &

Lee, 2004) and the OFC may encode item rather than action value (Padoa-Schioppa &

Assad, 2006). Other studies have further implicated LIP in representing value (Sugrue,

Corrado, & Newsome, 2004), with one demonstrating that, in the context of the drift diffusion model, expected value information influences the starting point, and not the

rate, of the accumulation of sensory evidence in LIP neurons (Rorie, Gao, McClelland, &

Newsome, 2010).

Subcortical regions have also been found to encode action values. For instance, when subjects chose between leftward and rightward handle turns to receive a probabilistic reward, neurons in the striatum fired in accordance either with the expected value of a leftward turn or that of a rightward turn (Samejima, Ueda, Doya, &

Kimura, 2005). Citing these and other studies in nonhuman primates, Samejima and

Doya (2007) proposed a hierarchical reinforcement learning model in which four subloops of the cortico-basal ganglia network—the lateral prefrontal-caudate head loop, orbitofrontal-ventral striatal loop, posterior parietal-caudate body network, and premotor/motor-putamen network—are involved in selecting different levels of actions, i.e. selecting a strategy, a target object, a spatial target, or a movement, based on different levels of states, i.e. the present task goal, the environmental state, the subject’s history of actions and experienced outcomes, and its sensory perceptions (Samejima &

Doya, 2007). Following additional work in rodents, this model was refined to include the algorithms employed by each network during reinforcement learning (Ito & Doya,

2011).

A related characterization of reinforcement learning involves how the value function is updated. Classically, the reward or penalty following an action, i.e. the reward prediction error, is used to update the value function; this is known as simple or

model-free reinforcement learning. However, an actor may also be able to update value functions based on its own motivational state or the state of its environment, even without directly receiving reward or punishment. Known as model-based reinforcement learning, this type of updating is thought to underlie goal-directed behavior (Daw, Niv,

& Dayan, 2005; D. Lee, Seo, & Jung, 2012). There is evidence to suggest that model-free and model-based reinforcement learning call on some of the same brain networks, such as the monkey pallidus-habenula-dopamine pathway (Bromberg-Martin, Matsumoto,

Hong, & Hikosaka, 2010) and the human ventral striatum (Daw, Gershman, Seymour,

Dayan, & Dolan, 2011), but also on distinct networks, with, for example, prefrontal dopamine modulating model-based and striatal dopamine modulating model-free learning (Doll, Bath, Daw, & Frank, 2016). Determining their joint and distinct neural implementations is an active area of study (Boureau, Sokol-Hessner, & Daw, 2016; D.

Lee et al., 2012; Shohamy & Daw, 2015).

Thus far I have discussed how the brain represents the basic computations involved in value-based decision-making. As Rangel et al. (2008) pointed out, various contextual factors, such as uncertainty about outcomes (Platt & Huettel, 2008), can mediate how decision-making proceeds. Recently, researchers have begun to explore the neurobiology of how social context influences our decisions. We turn to this nascent area of study next.

1.4 Social Decision-Making

Other people can influence the decisions we make. This fact was first demonstrated by Solomon Asch in the 1950s, in his experiments showing that subjects’ decisions about a simple perceptual feature—which line in a set was longest—could be swayed by the reports of other people (Asch, 1956; R. Bond & Smith, 1996). The mere presence of other people can also change our behavior; for example, among adolescents, the presence of peers has been shown to increase risky decisions (Gardner & Steinberg,

2005) as well as exploratory behavior and rate of learning from experience (Silva,

Shulman, Chein, & Steinberg, 2015). How we make decisions not just when others are present, but in our interactions with them, has only recently begun to be explored by neuroscientists.

Decisions in social contexts add two elements onto nonsocial, value-based decisions (D. Lee & Seo, 2016). The first is the potential for other-regarding preferences; that is, whether the actor wishes to increase or decrease the well-being of others (Du &

Chang, 2015; Kurzban, Burton-Chellew, & West, 2015). Monkeys, like humans, have other-regarding preferences (Chang, Winecoff, & Platt, 2011), enabling investigation of the neuronal basis of these preferences. In the context of a reward allocation task, the anterior cingulate gyrus encoded monkeys’ allocation of reward to themselves or to others, while the OFC encoded the subject’s own receipt of reward, and the anterior

cingulate sulcus encoded allocation of reward to the other monkey or to no one (Chang,

Gariépy, & Platt, 2013).

The second factor in social decisions is that the outcome of a decision can depend on the behavior of others. For instance, when two people are walking towards each other on a sidewalk, each may choose to move to a side so they do not collide. If they move to opposite sides, collision is avoided—a positive outcome—but if they move to the same side, they will collide—a negative outcome. Thus the decision of which way to move in this situation should depend on a person’s prediction about how the other person will move. Making these predictions depends on having a “theory of mind”

(ToM)—that is, representing the intentions and goals of others (Premack & Woodruff,

1978; Yoshida, Dolan, & Friston, 2008), also known as mentalizing. Neuroimaging studies have revealed that the mentalizing system in humans involves the cingulate cortex, medial prefrontal cortex (mPFC), and temporoparietal junction (TPJ) (Behrens,

Hunt, & Rushworth, 2009; C. D. Frith & Frith, 2006). These regions—particularly the mPFC and TPJ—seem to be involved specifically in social, and not nonsocial, decisions

(D. Lee & Seo, 2016).

Whether and to what extent other animals exhibit a ToM remains controversial.

Nonhuman primates may represent what others see, hear, know, and perhaps intend, but as yet there is no evidence that they represent others’ beliefs (Call & Tomasello, 2008;

Drayton & Santos, 2014). The paucity of evidence for nonhuman primate ToM has held

back neurophysiological investigations of social decision-making. One study of strategic reasoning in a simple game against a computer opponent indicated that monkeys could change their behavior to counter the computer’s strategy, and neurons in the DM-PFC were involved in guiding decisions based on inferences about the opponent’s strategy

(Seo, Cai, Donahue, & Lee, 2014). These findings raise the possibility that, in competitive scenarios, monkeys may in fact infer the intentions of their opponents. Thus, as reported in section 3, we devised an interactive zero-sum game for monkeys that encouraged each player to develop complex strategies in order to outcompete the opponent. Our goal was to further elucidate the neuronal mechanisms of social decision-making, and in particular, to explore the neuronal basis of deception.

2. Numerical decision-making

Numbers are central to our everyday lives. When purchasing goods, planning events, telling the time, preparing our taxes, determining the winner of an election, constructing objects or buildings, and a host of other essential tasks, we rely on an understanding of quantity. Humans’ ability to use precise numbers and perform increasingly complex computations seems to be founded upon the approximate number system (ANS), in which a fuzzy sense of numerical quantity is represented in the brain without numerical symbols or words (Dehaene, 1999; 2011; Feigenson et al., 2004).

Evidence for this claim (Piazza, 2010) comes from findings that higher math skills tend to correlate with ANS acuity (DeWind & Brannon, 2012; Halberda, Mazzocco, &

Feigenson, 2008; Libertus, Feigenson, & Halberda, 2011; Lyons & Beilock, 2011), pre- and early-school ANS acuity predicts future math achievement (Gilmore, McCarthy, &

Spelke, 2010; M. M. M. Mazzocco, Feigenson, & Halberda, 2011b; Starr, Libertus, &

Brannon, 2013), ANS acuity is lower in children with a specific impairment in math known as developmental dyscalculia (Mazzocco, Feigenson, & Halberda, 2011a), and training the ANS can improve math proficiency (Park & Brannon, 2013). Because the

ANS is crucial for our numerical skills, understanding its biological basis can inform how we treat disorders such as acalculia and dyscalculia (Ardila & Rosselli, 2002) as well as how we teach math, and could provide novel ways to improve math skills in healthy individuals.

Abundant evidence indicates that the ANS is an evolutionarily ancient system, with many other species displaying an ability to perceive and even compare numbers of items (see section 2.2 and (Uller, 2008)). Other animals may benefit from an appreciation of quantity in a variety of ways (Pahl, Si, & Zhang, 2013), such as assessing the size of one’s ingroup versus a competitive outgroup (McComb, Packer, & Pusey, 1994; M. L.

Wilson, Britton, & Franks, 2002), counting eggs and regulating clutch size to avoid brood parasitism (Lyon, 2003), and foraging optimally (Gallistel & Gelman, 2000; Harper,

1982). An important, defining feature of the ANS across species is that it follows Weber’s law: the discriminability of two sets depends on the numerical ratio between them

(Dehaene et al., 1998; Piazza, 2010). This similarity justifies the use of other animals as model organisms for investigating the neurobiological basis of numerical reasoning.

Additional similarities between the human and non-human ANS have been discovered in rhesus macaques. During nonsymbolic number comparison, monkeys’ accuracy and response times depend on the ratio between the two numbers in a manner strikingly similar to humans (Cantlon & Brannon, 2006a). Monkeys are also able to perform approximate addition, selecting an array of items containing roughly the sum of two previously-presented arrays, and their performance again resembles that of humans in the same task (Cantlon & Brannon, 2007a). Monkeys can also perform subtraction

(Cantlon, Merritt, & Brannon, 2016).

Monkeys’ performance in numerical tasks not only follows Weber’s law, but shows several other signatures of human numerical reasoning. When instructed to choose either the larger or smaller of two nonsymbolic numerosities, monkeys exhibit a semantic congruity effect akin to that observed in humans, responding more quickly when instructed to choose the smaller (or larger) number of dots when the presented numerosities were small (or large, respectively) (Cantlon & Brannon, 2005), and, as for humans, whether numerosities were considered small or large depended on the context in which they were presented—that is, the range of values presented in that session

(Jones, Cantlon, Merritt, & Brannon, 2010). When performing addition, monkeys exhibit a problem size effect, where accuracy decreases as sum magnitude increases, and a tie effect, where accuracy is higher when the two addends are identical compared to when they are different, both of which characterize human symbolic addition (Cantlon et al.,

2016).

We aimed to further probe the numerical abilities and tendencies of rhesus monkeys, both to shed light on the evolutionary origins of our math skills and to more firmly establish this species as a viable model organism in neurobiological studies of numerical decision-making. First we investigated whether monkeys exhibit another important mathematical skill: the ability to compare fractions (section 2.1). Next we explored whether monkeys share humans’ propensity to associate number with space

(section 2.2). Given our positive findings in these studies, which added to the parallels

between human and rhesus monkey number processing, along with findings that similar regions of the brain—most notably the intraparietal sulcus—are involved in numerical reasoning in rhesus macaques and humans (Cantlon, Platt, & Brannon, 2009;

Nieder & Dehaene, 2009), we conducted a number of pilot experiments to attempt to establish a paradigm for investigating the neuronal mechanisms underlying the effects of transcranial magnetic stimulation (TMS) on cognition, using macaque numerical decision-making as a model (section 2.3).

2.1 Discrete ratio comparison by rhesus macaques

We interpret proportions frequently in our everyday lives, such as when we are doubling a recipe, determining how much an item will cost after a discount, or computing a tip. We are also capable of comparing ratios. For instance, what is the relative generosity of a poverty-stricken mother of three who gives $10 faithfully to a charity each Sunday, versus a much wealthier member of the community who gives a little more money? Other typical situations calling on proportional reasoning include comparing the concentration of two solutions in a chemistry experiment or two batting averages of baseball players. Unfortunately, many children have trouble learning about proportions and their symbolic counterpart, fractions (Hartnett & Gelman, 1998; Siegler,

Fazio, Bailey, & Zhou, 2013). Thus an important question is whether and how the mind represents proportions prior to formal education.

One way to address this issue is to investigate the proportional reasoning abilities of individuals in other cultures that lack formal education. Fontanari and colleagues (Fontanari, Gonzalez, Vallortigara, & Girotto, 2014) tested various aspects of probabilistic cognition in preliterate and prenumerate adults from two indigenous

Mayan groups. In one experiment, they asked participants to bet on which of two sets, each composed of a different ratio of winning to non-winning chips, was more likely to produce a winning chip. The Mayan adults tended to choose the set with a greater proportion of winning chips, even when the quantities were varied such that the correct set actually contained the smaller absolute number of winning chips. Those findings suggest that proportional reasoning does not depend on formal math education.

Another line of research has demonstrated that preverbal infants attend to proportion. Denison and Xu (Denison & Xu, 2010) showed 12- to 14-month-old infants two cups: one with 40 pink and 10 black lollipops, and the other with 10 pink and 40 black lollipops. The experimenter then drew a lollipop from each cup such that infants could not see its color, and placed it in another container. The majority of infants crawled to the container corresponding to the cup with the larger proportion of their preferred color. In a sequence of follow-up experiments, they ruled out a number of different heuristics that the infants might have used, such as preferring the cup with the greater absolute quantity of the preferred color, avoiding the cup with the greater absolute quantity of the non-preferred color, or attending only to whether each

individual cup had more of the preferred or non-preferred color (Denison & Xu, 2014).

These studies demonstrate that infants behave in accordance with probabilistic expectations based on proportions.

Even younger babies appear to extract information about proportions from a visual display. After viewing habituation displays that contained yellow and blue shapes at a particular ratio, 6-month-old infants looked longer to displays that contained yellow and blue shapes with a novel ratio when the two ratios differed by a factor of 2, but not when they differed by a factor of 1.5 (McCrink & Wynn, 2007). These results indicate not only that preverbal infants are sensitive to ratios but also that their ability to discriminate between ratios depends on the difference between the ratios. Thus Weber’s law (see eg. (Pica, Lemer, Izard, & Dehaene, 2004)) well characterizes infants’ ability to compare sets based on their ratio.

A variety of experimental paradigms have been used with nonhuman animals to test for proportional reasoning. When foraging competitively, birds and fish distribute themselves among patches according to the ratio of patch profitabilities (Harper, 1982;

Godin & Keenleyside, 1984; Gallistel, 1990). The rate of pecking in pigeons is proportional to the ratio of positive to negative items in a stimulus array (Emmerton,

2001; Honig & Stewart, 1989; 1993). Bonobos, chimpanzees, gorillas, and orangutans make statistical inferences based on proportions (Rakoczy et al. 2014). In that study, apes were shown two containers with mixtures of a preferred and non-preferred food item.

An experimenter drew one item from each container and placed the items in opaque cups. When the apes were given a choice between the two opaque cups, individuals of all four ape species tested tended to choose the cup corresponding to the container with a more favorable ratio of preferred to nonpreferred food items, even when the absolute quantities favored the other container (Rakoczy et al., 2014); see also (Hartnett &

Gelman, 1998; Siegler et al., 2013; Woodruff & Premack, 1981).

Finally, Vallentin and Nieder (2008) demonstrated proportional reasoning in rhesus monkeys by training them to match pairs of lines based on the ratio of the two lines within each pair. Monkeys were shown a sample stimulus composed of two lines.

The ratio between the length of the lower line and the upper line was 1:4, 2:4, 3:4, or 4:4 during training, or 3:8 and 5:8 during testing. After a delay, the monkeys were shown a second pair of lines. They were to release a lever if the ratio between the length of the lower line and upper line was the same as during the sample, and to continue holding the lever if the ratio was different. Monkeys transferred to novel ratios and performed this task with accuracies similar to human subjects (Fontanari et al., 2014; Vallentin &

Nieder, 2008).

Here we ask whether the proportional reasoning skills of rhesus monkeys extend beyond those demonstrated by Vallentin & Nieder (2008) in two ways. First, can monkeys decide which of two ratios is more favorable? Most real-world proportional decisions require a judgment to be made about which is greatest or least, not about the

equivalence or non-equivalence of two ratios. Second, can monkeys make this decision when stimuli are composed of discrete elements rather than continuous quantities?

Our research questions also touch upon another debate that has percolated in the field of animal cognition. How and when can animals reason about relations between relations? This capacity may be the foundation for analogical reasoning (Denison & Xu,

2010). Some authors have argued that understanding second-order relations requires language (Denison & Xu, 2014; Premack, 1983) and may be an ability unique to great apes (McCrink & Wynn, 2007; R. K. R. Thompson & Oden, 2000) or humans (Emmerton,

2001; Honig & Stewart, 1989; 1993; Penn, Holyoak, & Povinelli, 2008). However, there are reports that other species, including monkeys and birds, succeed at relational matching-to-sample tasks: they can categorize individual arrays as containing all-same or all-different elements, and then report whether the types of arrays are the same (both all-same or both all-different) or different (one all-same and the other all-different)

(Cook & Wasserman, 2007; Fagot & Maugard, 2013; Flemming, Beran, Thompson,

Kleider, & Washburn, 2008; Flemming, Thompson, Beran, & Washburn, 2011; Rakoczy et al., 2014; Smirnova, Zorina, Obozova, & Wasserman, 2015; Truppa, Piano Mortari,

Garofoli, Privitera, & Visalberghi, 2011; Vonk, 2003). Similarly, when comparing two ratios, subjects must apprehend the first-order relation between elements within each stimulus array, as well as the second-order relation between the ratios in the two arrays.

Determining which of two arrays has the more favorable ratio is therefore an example of

understanding relations between relations.

We presented monkeys with two arrays that were each composed of S+ (positive rewarded stimulus) and S- (negative rewarded stimulus) items. We rewarded the monkeys for selecting the array with the higher ratio of S+ to S- items. Following training, we tested monkeys on probe trials with novel numerosities and novel ratios between the ratios in each array. Finally we tested the monkeys on a variety of control conditions to tease out the relative contributions of numerosity and surface area on their decisions.

2.1.1 Methods

Subjects & Housing

Subjects were two adult female rhesus macaques (Macaca mulatta, Monkey B: 12 years old, Monkey C: 12 years old). Both subjects had previously been trained to use a touch-screen, and had participated in a study on magnitude bisection. They had also received minimal training on numerical matching and ordering. They were pair-housed together in a vivarium at Duke University, and were separated each day to participate in the experiment. Fresh fruit and Purina monkey chow were provided daily, and additional treats such as dried fruit or nuts were also provided daily. Water consumption was restricted during some periods of this study for unrelated experiments.

Apparatus

A 15-inch touch-sensitive computer monitor (Elo TouchSystems, Menlo Park,

CA) and a food pellet reward delivery system (Med Associates, St. Albans, VT) were attached to the front of the subjects’ home cage. Stimulus presentation, reward delivery, data collection, and data analysis were performed via custom-written programs in

MATLAB (Mathworks, Natick, MA) with the Psychophysics Toolbox add-on (D. H.

Brainard, 1997) (https://psychtoolbox.org).

Training

Training progressed through two main phases: shape training and ratio training.

In shape training monkeys learned to associate one of two shapes (black circle and white diamond) with reward, making it S+ while the other shape was S-. For Monkey B, the black circle was S+ while the white diamond was S-, and for Monkey C these contingencies were reversed. During shape training, subjects initiated trials by touching a red rectangle at the bottom-center of the screen (Figure 1). The two shapes then appeared on a gray background, one on the left and one on the right side of the monitor

(side randomized). If the subject touched S+, the screen turned the color of that shape for a 1-second inter-trial interval (ITI), a pleasant “ding” sound played, and a candy (either a mini Reese’s Pieces or a mini M&M) was delivered. If she touched S-, the screen turned the color of that shape for a 2-second ITI, a “buzz” sound played, and no candy was delivered. Each session consisted of at least 100 trials. The performance criterion to

advance from shape training to ratio training was 80% accuracy (selecting S+ on at least

80% of trials) in a single session.

Ratio training was designed to train the monkeys to select the array that had a more favorable ratio of S+ to S-. Monkeys were presented with two intermixed arrays of

S+ and S- items with a red circle surrounding each array (Figure 2). All items had the same width and height. The physical arrangement of S+ and S- items within each red circle was trial-unique and randomly generated such that exact stimulus configurations could not be memorized. The side on which the correct array was displayed was also determined randomly on each trial, with equal probability given to the left and the right.

When the subject touched one of the arrays, the circle around that array turned blue for 47

0.25s. If the subject chose the array with the higher ratio of S+ to S-, a “ding” sound played, the screen turned the color of the S+ for 1s, and a candy was delivered. If she chose the array with the lower ratio, a “buzz” sound played, the screen turned the color of the S- for 2-5s and no candy was delivered.

We defined the ratio-of-ratios on a given trial as the ratio of S+ to S- on the correct side, divided by the ratio of S+ to S- on the incorrect side. During training, all trials had a ratio-of-ratios of 2 or 4 with 16 different absolute values, shown in Table 1.

We constructed these sets such that on half the trials, the array with the greater absolute numbers of S+ and S- (and thus the greater absolute number of items) was also the array with the greater ratio of S+ to S- (congruent); while on the other half of the trials, the greater absolute number of items occurred in the array with the lesser ratio of S+ to S-

(incongruent). In this way subjects could not perform above chance expectations if they relied on a strategy of choosing the array with the greater number of S+. Instead the structure required that monkeys consider the relative numbers of S+ to S-. A second consideration in stimulus control was to ensure that monkeys needed to compare ratios between the two arrays in order to perform successfully, rather than only considering a single array. Thus, in addition to unfavorable-favorable (UF) trials in which one array had more S- than S+ while the other had more S+ than S-, we also included favorable-favorable

(FF) trials in which both arrays had a greater number of S+ than S-, and unfavorable- unfavorable (UU) trials in which both arrays had a greater number of S- than S+. In this

way subjects could not be successful by examining just one array, deciding if there were more S+ than S- in that array, then choosing it if the decision were “yes” and choosing the other if the decision were “no”. This yielded a 2x3 design for training trial conditions: each trial was either congruent or incongruent, and either UF, FF, or UU.

The relative frequency of each trial type within each session was varied according to subjects’ performance (for example, including a larger proportion of any

trial type for which the monkey was not performing well) until the last 21 sessions, at which point each of the 6 trial types was presented with equal frequency in an order determined randomly each day. Each 120-trial training session was conducted on a separate day and lasted until the subject completed all trials, or did not interact with the screen for 5 minutes. Training continued for approximately 180 days and accuracy surpassed 75%.

Table 1: Ratio training stimulus array numerosities.

Probe Test Session Structure

Test sessions included standard trials, which were differentially reinforced and consisted of the training values described above (n = 120), and probe trials that were nondifferentially reinforced (n = 32 or 36). On probe trials, regardless of which array the subject selected, the positive sound, shorter ITI, and candy reinforcement were delivered. Thus subjects could not learn which probe arrays were correct over the course of testing. Probe trials were randomly intermixed with standard (training) trials, except that the first 10 trials of every session were standard trials. Four different categories of probe trials were included in separate sessions, detailed below: numerical ratio-of-ratios, surface-area-only, numerosity-only, and surface area versus numerosity. We considered a test session complete if the subject completed at least 90% of the trials, except for the final session of surface area versus numerosity testing, when a technical malfunction resulted in the presentation of only 104 total trials to each subject, which they both completed.

Numerical Ratio-of-Ratios Probe Trials

To test whether the monkeys would generalize what they had learned in training to stimulus arrays with novel numerosities and novel ratios-of-ratios among the numerosities, we introduced probe trials with each of six novel ratios-of-ratios: 1.5, 2.5,

3, 3.5, 6, and 10. We divided these into two stimulus sets (Table 2), and ran five testing sessions with each stimulus set for each subject. Each session included two exemplars of 51

each of the array sets in Table 2 (the correct array appeared on the left once, and on the right once).

Table 2: Numerical ratio-of-ratio probe stimulus array numerosities.

Control 1: Surface-Area-Only

During training and numerical ratio-of-ratios probe trials, all S+ and S- items were the same size, such that total surface area varied with numerosity. To test whether subjects were exclusively relying on the relative surface area of the S+ and S- items, we introduced probe trials in which monkeys were presented with a single S+ and a single

S- on each side of the monitor as shown in Figure 3. In this way numerical ratio-of-ratios could not serve as a cue. Crucially, we varied the size of the shapes so that the ratio-of- ratios of total surface areas between the array with the more favorable ratio of surface area of the S+ to surface area of S- (the “correct” array) and the array with the less favorable ratio of surface areas (the “incorrect” array) was 1.5, 3, or 6 as shown in Table

3. Each set of surface areas was presented twice in a session (the correct array appeared on the left once, and on the right once), yielding 36 probe trials per session. Subjects were given three sessions of this control condition.

Table 3: Surface-area-only probe stimulus array surface areas (given in pixels).

Control 2: Numerosity-Only

To test whether subjects would select the array with the greater numerical ratio- of-ratios in the absence of surface area cues, we introduced probe trials in which total surface area was equated for S+ and S- within each array and across the two arrays.

Specifically, we altered individual item size so that the total surface area of S+ on the correct side, total surface area of S- on the correct side, total surface area of S+ on the incorrect side, and total surface area of S- on the incorrect side were all equal (Figure 4).

Numerosities in these probe trials were the same as in training (Table 1), with each numerosity set appearing twice in a session as a probe trial, yielding 32 probe trials per session. Subjects were given three sessions of this control condition.

Control 3: Surface Area versus Numerosity

To further investigate the potential contributions of numerosity and surface area in guiding subjects’ decisions, we introduced probe trials in which numerical ratio-of-

ratios and surface area ratio-of-ratios would lead to opposite choices. On these trials, the numerical ratio-of-ratios between the left and the right array was 4 (or 0.25), while the surface area ratio-of-ratios between the left and the right array was 0.25 (or 4, respectively) (Figure 5). Thus, the array with the greater numerical ratio was the array with the lesser surface area ratio, and vice versa. We included six different numerical sets (one for each of the congruence X favorability categories; Table 4), and presented each one six times according to six different surface area sets (again, one for each of the congruence X favorability categories; Table 5), yielding 36 total probe trials, which were each presented once in a session. Subjects were given three sessions of this control condition.

Table 4: Surface Area versus Numerosity probe stimulus array numerosities.

2.1.2 Results

Shape Training

Both subjects quickly learned to select S+. By the second training session, they each exceeded the 80% accuracy criterion for advancing to ratio training.

Ratio Training

Subjects successfully learned to perform the ratio comparison task over many training sessions (180 for Monkey C, 179 for Monkey B). Over the final 10 training sessions, during which each monkey completed all 120 trials per session, Monkey B had a mean accuracy of 82.4% correct per session (SEM = 0.8%), and Monkey C had a mean accuracy of 77.6% correct per session (SEM = 1.4%). Given a binomial distribution with a success probability of 50%, a two-tailed P-value of 0.05, and a 120-trial session, performance worse than 40.8% or greater than 59.2% exceeds chance expectations. Both subjects performed above that level on each of the final 21 days of training, when the 6 trial types were presented with equal frequency.

Numerical Ratio-of-Ratios Testing

To determine whether subjects’ decisions were based on ratio-of-ratios or instead whether they had memorized the training sets, we presented nondifferentially reinforced trials with novel ratio-of-ratios values (“probe trials”) mixed in with standard training trials that were differentially reinforced (“standard trials”). Subjects continued

to perform above chance expectations on the standard trials during these test sessions

(Figure 6; P < 0.001 for both binomial tests). Importantly, subjects also performed above chance expectations on probe trials with novel ratio-of-ratio values (mean = 83.5%, SEM

= 1.4%, binomial P < 0.0001). This was true for each of the novel ratio-of-ratio values

(Figure 6; P < 0.001 for all binomial tests).

As shown in Figure 6, performance increased as the ratio-of-ratios increased. A generalized linear model (GzLM) following a binomial distribution predicting whether a given probe trial was correct from the four factors subject, ratio-of-ratios, congruence condition, and favorable condition confirmed that accuracy increased with the ratio of ratios and thus followed Weber’s law (beta weight for ratio-of-ratios: 0.2437, P < 0.0001).

The overall model was significantly different from the constant model (χ2709 = 116, P <

0.0001). There was a marginal effect of subject (beta = -0.417, P = 0.0614), due to the fact that Monkey C (mean = 81.1%, SEM = 2.1%) was less accurate than Monkey B (mean =

85.9%, SEM = 1.9%). There was an effect of congruency (beta = 1.65, P < 0.0001), indicating that subjects performed better on congruent (mean = 92.7%, SEM = 1.4%) than incongruent (mean = 74.3%, SEM = 2.3%) trials. Finally, there was an effect of favorability condition, with subjects performing best on UU trials (mean = 91.1%, SEM =

1.9%), slightly less well on UF trials (mean = 87.5%, SEM = 2.1%), and least well on FF trials (mean = 71.9%, SEM = 2.9%). The difference between UU and UF was significant

(beta = -1.59, P < 0.0001), as was the difference between UU and FF (beta = 1.17, P <

0.0001), while the difference between UF and FF was not (beta = -0.845, P = 0.18). Within each of these favorability categories, however, performance was still significantly above chance expectations (P < 0.0001 for all binomial tests).

In order to determine whether subjects selectively attended to certain aspects of the stimulus arrays – for instance, only looking at one side of the screen, or only taking

S+ or S- into account – we fit a GzLM that predicted a left-side array choice based on the following factors: number of S+ on the left, number of S- on the left, number of S+ on the right, number of S- on the right, and subject. We found that each of the four quantities of interest contributed significantly to the subjects’ decisions (beta for number of S+ on the left: 0.342, P < 0.0001; beta for number of S- on the left: -0.156, P < 0.0001; beta for number of S+ on the right: -0.251, P < 0.0001; beta for number of S- on the right: 0.142, P

< 0.0001). The overall model was significantly different from the constant model (χ2709 =

493, P < 0.0001). There was no significant effect of subject (beta = 0.315, p = 0.16). These results indicate that both subjects considered the relations between all the quantities on the screen when making their decisions.

Reaction time (RT) also decreased as ratio-of-ratios increased on probe trials

(Figure 7). A GzLM predicting RT from the factors subject, ratio-of-ratios, congruence condition, and favorable condition was fit to the data after all RTs that exceeded the

97.5th percentile (1.23s), or two standard deviations above the mean, were excluded. The overall model was significantly different from the constant model (F690 = 24.6, P < 0.0001).

Each of the four factors had a significant effect on RT: subject (beta = -0.291, P < 0.0001), indicating that Monkey B (mean = 0.573s, SEM = 0.008) responded more quickly overall than Monkey C (mean = 0.682s, SEM = 0.010); ratio-of-ratios (beta = 0.011, P = 0.043), indicating that RT decreased as ratio-of-ratios increased (note that beta weight sign is opposite to the direction of the effect due to the use of a gamma distribution with a 63

reciprocal link function); congruence condition (beta = 0.091, P = 0.0031), indicating that subjects responded more quickly on congruent (mean = 0.611s, SEM = 0.009) than incongruent (mean = 0.648s, SEM = 0.010) trials; and favorability condition (between UF and FF: beta = 0.140, p = 0.00015; between UF and UU: beta = 0.153, P < 0.0001; between

UU and FF: beta = -0.013, P = 0.74), indicating that subjects responded more quickly on

UU trials (mean = 0.608, SEM = 0.011) and FF trials (mean = 0.612s, SEM = 0.011) with no significant difference between them, and significantly more slowly on UF trials (mean =

0.670, SEM = 0.013). The finding that subjects responded more quickly to higher ratios- of-ratios even when controlling for individual differences (subject) and the effects of congruence and favorability conditions supports the hypothesis that subjects attended to and made decisions based on ratio-of-ratios values, with their behavior conforming to

Weber’s law.

Control 1: Surface-Area-Only

During training and numerical ratio-of-ratios testing, both S+ and S- items were held constant in size, such that the ratio-of-ratios of total surface areas varied systematically with the ratio-of-ratios of numerosities. To determine the degree to which subjects relied on the ratio between total surface area of S+ to S- to make their decisions, we tested the monkeys in three control conditions described in the Methods. In the first of these, as shown in Figure 3, we presented only one of each shape on each side so that the numerical ratio of ratios was 1 (i.e. 1:1/1:1). Crucially, we varied the size of each shape such that the surface area ratio-of-ratios were equal to 1.5, 3, or 6, which were 65

values the monkeys succeeded at in the numerical ratio-of-ratio probes. If subjects used ratios of total surface area rather than numerosity, they should have chosen the side with the more favorable surface area ratio-of-ratios in these probe trials. However, subjects performed near the 50% chance level on these trials. Monkeys chose the side with the more favorable surface area ratio-of-ratios on 49.5% of trials (SEM = 3.5%, binomial P = 0.53). This held true at each of the three surface area ratio-of-ratio values

(Figure 8), with performance close to exceeding chance level only for the easiest ratio-of- ratios value of 6 (mean accuracy = 58.3%, SEM = 5.9%, binomial P = 0.063). A GzLM predicting whether a given probe trial was correct from the factors subject, surface area ratio-of-ratios, surface area congruence condition, and surface area favorability condition revealed that accuracy was modulated by surface area ratio-of-ratios (beta =

0.178, P = 0.034) and congruence condition (beta = 1.67, P < 0.0001), and by favorability condition (between UF and FF: beta = 0.848, P = 0.028; between UF and UU: beta = 0.713,

P = 0.064; between UU and FF, beta = 0.135, p = 0.71), but not by subject (beta = 0.329, p =

0.28). The overall model was significantly different from the constant model (χ2202 = 40.6,

P < 0.0001). RT was not significantly modulated by surface area ratio-of-ratios on these probe trials; a GzLM with factors subject, surface area ratio-of-ratios, surface area congruence condition, and surface area favorable condition revealed that none of these factors had a significant impact on RT (the model was not significantly different from the constant model (F198 = 0.51, p = 0.77, and all beta p’s > 0.1)). These results indicate that

subjects did not exclusively use total surface area ratio-of-ratios to guide their decisions during training or in the numerical ratio-of-ratios probe trials; however, the analysis also revealed some sensitivity to the surface area ratio-of-ratios.

Control 2: Numerosity-Only

To further explore the monkeys’ use of numerical ratio-of-ratios versus other visual cues, we removed total surface area as a cue by equating the total surface area of each of the four sets of shapes such that surface area ratio-of-ratios was equal to 1

(Figure 4). The numerosities and numerical ratio of ratios were identical to those used in training and shown in Table 1. Surprisingly, subjects were unsuccessful at these probe trials, choosing the correct numerical ratio-of-ratios on only 51.0% of trials (SEM = 3.6%, binomial P = 0.36). To assess whether individual item size controlled performance, we fit a GzLM predicting whether subjects chose the left-side array based on the following

factors: radius of S+ on the left, radius of S- on the left, radius of S+ on the right, radius of

S- on the right, and subject. We found that radius of S+ on the left (beta = 0.089, p = 0.045) and on the right (beta = -0.108, P = 0.017) were significant predictors of choosing the left side, and radius of S- on the left (beta = 0.074, P = 0.066) was marginally significant, while radius of S- on the right (beta = -0.056, P = 0.21) was not significant. There was no effect of subject (beta = -0.066, P = 0.23). The overall model was significantly different from the constant model (χ2186 = 51.7, P < 0.0001). In a GzLM on response latency with the same factors, the overall model was again significantly different from the constant model (F179 = 12.4, P < 0.0001). There was a significant effect of subject (beta = -0.298, P <

0.0001) and of S- radius on the left side (beta = -0.010, P = 0.049; all other P’s > 0.1). These results suggest that subjects’ decisions were strongly modulated by individual item size.

Monkeys tended to choose the side with the larger individual shapes, particularly for the S+, though the trend was similar for S- (Figure 9). It is important to note that since all items had the same radius during training and numerical ratio-of-ratios probe testing, subjects could not have used this strategy in those conditions.

Control 3: Surface Area versus Numerosity

In the final control condition, we directly pitted surface area and numerosity against each other. On probe trials, the numerical ratio-of-ratios between the left and right arrays was either 4 or 0.25, while the surface area ratio-of-ratios was 0.25 or 4, respectively. Thus the side with the more favorable numerical ratio-of-ratios was also the side with the less favorable surface area ratio-of-ratios, and vice versa (Figure 5). In

these probe trials, subjects chose the side with the more favorable surface area ratio-of- ratios on 59.5% of trials (SEM = 3.5%), which was significantly above chance level

(binomial P = 0.0032). However, this performance still fell well below subjects’ performance during numerical ratio-of-ratios probes where subjects could have used both numerosity and surface area (83.5%; χ2 = 52.1, P < 0.0001), suggesting they were not solely using surface area ratio-of-ratios during those trials. Moreover, performance was once again strongly modulated by item size, according to a GzLM predicting a left-side array choice from the factors subject (beta = -0.246, P = 0.61), radius of S+ on the left (beta

= 0.32, P < 0.0001), radius of S- on the left (beta = 0.145, P = 0.0002), radius of S+ on the right (beta = -0.307, P < 0.0001), and radius of S- on the right (beta = -0.184, P < 0.0001).

The model was significantly different from the constant model (χ2189 = 141, P < 0.0001).

Together with the other control conditions, these results suggest that subjects’ choices were influenced by changing item radius and total surface area, but also suggest that the monkeys did not solely rely on these factors during training or numerical ratio-of-ratios probe testing.

2.1.3 Discussion

The findings presented here demonstrate proportional reasoning abilities in rhesus macaques. First we showed that monkeys could learn to choose between two intermixed arrays of positive and negative stimuli based on which array had a higher ratio of positive to negative stimuli. In novel, nondifferentially reinforced probe trials, 70

subjects continued to choose the more favorable array. Importantly, monkeys exhibited above-chance performance even when the absolute quantity of positive stimuli was lower in the array with the higher ratio, showing that they were attending to relative quantity. Their performance improved as the ratio between ratios increased, indicating that ratio discrimination follows Weber’s law.

Continued probe testing with controls for surface area, however, suggested that the monkeys relied on a combination of cues. The stimuli were discrete arrays of positive and negative items that were homogeneous in size, so numerical ratio and surface area ratio were confounded throughout training and novel ratio-of-ratios probe testing. Thus monkeys could have attended to numerical ratio-of-ratios, surface area ratio-of-ratios, or both. Our three control conditions reveal that subjects used both numerical and surface area cues and did not exclusively use one or the other.

Specifically, monkeys failed to perform above chance expectations in the first control condition, which completely removed numerosity ratio-of-ratios as a potential choice dimension. This demonstrates that monkeys could not solve the problem when required to attend only to the surface area ratio-of-ratios and thus that they were unlikely to have been solely attending to the surface area ratio-of-ratios in training and probe testing. In the second control condition, when the stimuli were designed such that monkeys could not use the surface area ratio-of-ratios, they again performed poorly, suggesting that they were unable to attend to the pure numerosity ratio-of-ratios information. Moreover,

there was also clear evidence that the size of the items influenced performance. Finally, in the third control condition when surface area and numerosity ratio-of-ratios were pitted against each other, monkeys showed a small but significant bias to use surface area over numerosity.

The data are consistent with the idea that surface area and number served as redundant cues and that the monkeys did not exclusively rely on either cue. This finding parallels data from infants in which the ordinal relationship between quantities was detected when surface area and numerosity were redundant cues but not when either dimension was presented in isolation (Suanda, Tompson, & Brannon, 2008). Results in fish also suggest that providing multiple cues to magnitude – i.e. both number and continuous quantities like surface area – can facilitate performance in a numerosity comparison task (Agrillo et al. 2011). Indeed, continuous visual variables influence human adults’ perception of nonsymbolic number (DeWind et al. 2015; Gebuis &

Reynvoet 2012a, 2012b), perhaps because such variables are often confounded with numerosity in daily life. Thus it may be unsurprising that monkeys integrated information from multiple sources to aid performance in our task.

The small bias exhibited by the monkeys for surface area over numerosity in the third control condition was unexpected in light of prior evidence that monkeys attend more to number than surface area during a match-to-sample task (Cantlon & Brannon

2007). It could be that it is more natural for animals to consider ratios of continuous

quantities and magnitudes of discrete quantities. However, it is also possible that features of our experimental design – such as the order of the control conditions – produced the greater reliance on surface area we observed. Subjects may have learned to rely on or ignore certain cues during the numerosity-only and surface-area-only probe trials which later influenced their performance during surface area versus numerosity testing.

An important point, however, is that the monkeys in our study were trained with stimuli that confounded number and surface area. It is an open question whether monkeys could make pure numerical ratio judgments if their initial training were with numerosity stimuli that controlled for surface area. However, training in our task already required many months, likely due to the complex nature of the stimuli in which many different features could have been relevant. Based on our findings, we would predict that monkeys could ultimately succeed at ratio comparison when surface area and other continuous visual variables are controlled during training, but such a task – made more difficult by removing helpful cues – may take even longer for them to learn.

Our study nevertheless demonstrates that monkeys are able to make decisions about ratios of visually discrete items, as opposed to continuous quantities like line length. This is the first time such a skill has been demonstrated in monkeys and thus extends prior work showing that monkeys can make ratio judgments about continuous quantities (Hayden, Heilbronner, & Platt, 2010; Vallentin & Nieder, 2008). The

distinction between continuous and discrete quantities is important because children struggle with nonsymbolic proportions involving discrete quantities more than they do with continuous quantities (Boyer, Levine, & Huttenlocher, 2008; Jeong, Levine, &

Huttenlocher, 2007; Spinillo & Bryant, 1999). One explanation for children’s increased difficulty with discrete ratios is that learned explicit mathematical processes with whole numbers, such as counting, interfere with correct intuitive processes, such as relative visual comparison (Boyer et al., 2008; C. A. Thompson & Opfer, 2008). Monkeys’ success at comparing ratios of discrete items suggests that similar tasks can be solved by intuitive processes in humans. Moreover, accuracy in a non-symbolic fraction magnitude comparison task correlated with overall math achievement in fifth grade children (Fazio, Bailey, Thompson, & Siegler, 2014). This evidence supports the notion that intuitive strategies should be encouraged while explicit strategies like counting should be discouraged when children are learning about discrete proportions and fractions.

Our results also demonstrate that rhesus macaques can reason about relations between relations. They made decisions based not only on the first-order relation between S+ and S- within each array, but also on the second-order relation between the ratios of S+ to S- in the two arrays. This finding adds to the recent literature that animals other than apes can reason analogically (Cook & Wasserman, 2007; Fagot & Maugard,

2013; Flemming et al., 2008; 2011; Smirnova et al., 2015; Truppa et al., 2011; Vonk, 2003),

and argues against the hypotheses that such skills require language or symbol training

(Premack, 1983; R. K. R. Thompson & Oden, 2000), or may be unique to humans (Penn et al., 2008). However, our results do not speak to whether monkeys can match relations, nor to whether they spontaneously reason analogically, as has been found in apes

(Vonk, 2003) and crows (Smirnova et al., 2015). Whether rhesus macaques identify relations between relations without explicit training remains an open question.

In conclusion, we demonstrate that rhesus macaques are able to compare arrays of discrete items based on the ratio of the items within each array, with performance following Weber’s law. These findings extend prior work demonstrating that rhesus monkeys can match ratios of continuous line lengths (Vallentin & Nieder 2008) and that human infants are sensitive to the equivalence or non-equivalence between arrays based on the ratios of discrete items (McCrink & Wynn 2007). Proportional reasoning clearly does not require language or formal education. An important question is how we can harness these primitive nonverbal capacities to facilitate proportional reasoning in children.

2.2 Rhesus monkeys map number onto space

Number and space are integrally linked in the human mind (Bueti & Walsh,

2009; Fias & Fischer, 2005; Hubbard, Piazza, Pinel, & Dehaene, 2005). A prominent behavioral manifestation of this interaction is the SNARC (Spatial-Numerical

Association of Response Codes) effect, whereby smaller numbers are associated with one side of space and larger numbers with the other. In the original demonstration of the

SNARC effect, subjects making parity judgments (i.e. odd or even) were faster to respond to smaller numbers with a left-side response key and larger numbers with a right-side response key (Dehaene, Bossini, & Giraux, 1993). This pattern has been replicated in a variety of experimental paradigms and populations (reviewed in Fias &

Fischer, 2005), including in the auditory modality among both blind and sighted subjects

(Castronovo & Seron, 2007), indicating that the number-space mapping is amodal.

Interestingly, in cultures that read from right to left the SNARC effect can be attenuated or even reversed, suggesting that experience may drive the mapping of number onto space (Shaki & Fischer, 2008; Shaki, Fischer, & Petrusic, 2009; Zebian, 2005).

Although the specific orientation of the SNARC effect varies with cultural conventions for reading direction, its presence across cultures suggests that the spatial mapping of numbers may be a universal cognitive strategy (Gobel, Shaki, & Fischer,

2011). In fact, preschool children, who have not begun learning to read, are quicker and more accurate at finding an object hidden in a numbered compartment when compartment numbers increase from left to right rather than right to left, indicating they expect numbers to increase from left to right (Opfer, Thompson, & Furlong, 2010; Opfer

& Furlong, 2011). These results suggest that spatial-numerical associations are not a product of extensive formal education or reading experience. Moreover, even preverbal

infants, who have not yet learned to count or use measurement tools like rulers, are sensitive to the relationship between number and space (de Hevia & Spelke, 2010).

When habituated to dot arrays (i.e. non-symbolic numbers) presented in either increasing or decreasing order, eight-month-old infants looked longer (indicating a novelty response) to line lengths presented in the opposite, but not the same, order.

Furthermore, when infants were shown several exemplars of dot arrays paired with lines such that the line length either increased (positive pairing) or decreased (inverse pairing) with number of dots, infants extracted the rule from the positive but not the inverse pairing. Infants also showed a preference for positive pairings between numbers and line lengths over inverse pairings. These findings indicate that human infants have a predisposition to relate number to space and that our privileged number-space mapping does not depend upon language. An important question then is whether number is mapped onto space in any nonhuman species.

Rugani and colleagues found evidence for a left-to-right mental number line in baby chickens. First, they showed that 5-day-old domestic chicks (Gallus gallus) could learn to identify the third, fourth, or sixth hole in a series of ten identical holes sagittal to the chicks’ starting point, even when distance between holes was varied (Rugani,

Regolin, & Vallortigara, 2007). Critically, when the line of holes was rotated 90° so that it was horizontal to the chicks’ starting point, chicks that had learned to select the fourth hole preferentially approached the fourth hole from the left, and rarely the fourth hole

from the right. This finding was investigated further in newborn chicks and adult nutcrackers (Nucifraga columbiana). When trained to select the fourth (or sixth) item in a sagitally-oriented line of sixteen identical items, both species preferentially chose the fourth (or sixth, respectively) item from the left – but not the right – in a horizontally- oriented line (Rugani et al., 2010). However, when inter-element distances were varied during training or test, chicks were equally likely to choose the correct position from the left or the right (Rugani et al., 2011). The authors interpret their results as evidence for asymmetric processing of number and space. They posit that the purely ordinal aspect of the array is bilaterally represented in the left and right cerebral hemispheres, whereas the purely spatial aspect is unilaterally represented in the right hemisphere, biasing attention to the left side of space only when spatial cues (such as constant inter-element distance) could be used to complete the task. This explanation seems viable in birds, who have lateralized visual fields and a complete crossing of nerves at the optic chiasm, coupled with minimal interhemispheric connections (lack of corpus callosum), which results in visual information from one side of space being represented almost entirely in the opposite hemisphere (Larsson, 2013; Rogers et al., 2013). This strong lateralization likely underlies the numerous behavioral asymmetries observed in birds (Rogers et al.,

2013). Primates, on the other hand, receive visual information from both sides of space in each cerebral hemisphere, due to frontally placed eyes, only partial crossing of nerves at the optic chiasm, and communication between the hemispheres (Larsson, 2013). Thus

although Rugani and colleagues found a SNARC-like effect in a non-human species, the mechanism driving this effect could be different from that driving the effect in humans.

Exploring whether monkeys orient numerical representations spatially might yield more information about the evolutionary precursors of the spatial mapping of number in humans.

The SNARC effect depends upon the ordinality inherent in number representation. The ability to identify a particular ordinal position in a sequence is crucial for many everyday activities, such as following directions (i.e. take the third right) or locating a car in a parking lot (i.e. in the eighth spot of the fifth row). Using numbers to denote rank or ordinality is theoretically distinguished from using numbers to denote quantity or cardinality. Indeed, behavioral (Franklin et al., 2009; Rubinsten &

Sury, 2011; Turconi, et al., 2006), neurological (Butterworth, 1997; Turconi & Seron,

2002), and neuroimaging (Tang et al., 2008; Turconi et al., 2004; Zorzi et al., 2011) evidence indicates that distinct cognitive processes are engaged when accessing the ordinal rather than the cardinal meaning of numbers. A broad range of animal species can compare sets to choose the greater or lesser numerical value: birds (e.g. Aïn et al.,

2008; Bogale et al., 2011; Roberts, 2010), fish (e.g. Agrillo et al., 2010; Gómez-Laplaza &

Gerlai, 2012; Piffer et al., 2011; Stancher et al., 2013), amphibians (e.g. Krusche et al.,

2010; Uller et al., 2003), non-human primates (e.g. Beran, 2008; Brannon & Terrace, 1998;

Hanus & Call, 2007), and other mammals (e.g. Jaakkola et al., 2005; Uller & Lewis, 2009;

Vonk & Beran, 2012; Ward & Smuts, 2007).

Other research shows that animals can learn to respond to arbitrary stimuli in prescribed ordinal sequences and represent the abstract ordinal position of each item

(e.g. monkeys: Chen et al., 1997; Terrace, 2005; Terrace et al., 2003; domestic chicks:

Rugani et al., 2007). Pigeons show more limited ability to encode ordinal positions of arbitrary items in a sequence ( Scarf & Colombo, 2010; Terrace, 1993; Terrace et al., 1996).

Rats (Davis & Bradford, 1986; Suzuki & Kobayashi, 2000), bees (Dacke, 2008), and chicks

(Rugani et al., 2007) moving through space can learn to select a target landmark based on how many identical landmarks they have already passed. Only one previous study tested the ability of a non-human primate to respond selectively to a particular ordinal position in a sequence. Using the Wisconsin General Test Apparatus, Ruby trained a single rhesus monkey to select the third block from the left in a formboard from three, four, five, six, or seven identical blocks in a row (Ruby, 1984). However, in that study the food reward was always placed in the hole under the correct block, leaving open the possibility that the monkey used odor cues. Spacing between the blocks was also held constant throughout the experiment, so the monkey may have relied on spatial position rather than ordinal position.

We developed a set of touch-screen tasks for rhesus macaques that parallel the experiments Rugani and colleagues conducted with birds (Rugani et al., 2007; 2010;

2011). By using a touch-screen, we avoided experimenter cuing and odor cuing, and gained precise control over the appearance and spatial placement of stimuli. First we trained monkeys to select the fourth item from the bottom in a vertical array of five identical items presented on a touch-screen. We then tested their ability to identify the fourth position over changes in the appearance, location, and spacing of the stimuli.

Next we rotated the array ninety degrees into a horizontal line to assess whether monkeys map number to space in a preferred direction. If monkeys do not have a spatially oriented mental number line, they should respond to the fourth position from the left and fourth position from the right with equal likelihood. If macaques do have a spatially oriented mental number line, they should preferentially select the fourth position from the left, suggesting a left-to-right orientation, or the fourth position from the right, suggesting a right-to-left orientation, but not both.

2.2.1 Methods

Subjects

Four adult male rhesus macaques served as subjects (Macaca mulatta, mean age =

9.8 years, range = 7-15 years). Monkeys were housed singly (n=3) or in pairs (n=1) in a vivarium, and those housed in pairs were separated for testing. All animals had participated in previous touch-screen tasks unrelated to the present study. Fresh fruit

and Purina monkey chow were provided daily. The monkeys’ water consumption was restricted for unrelated experiments.

Apparatus

A 15-inch touch-sensitive computer monitor (Elo TouchSystems, Menlo Park,

CA) and a food pellet reward delivery system (Med Associates, St. Albans, VT) were attached to the front of each macaque’s home cage. Stimulus presentation, reward delivery, data collection, and data analysis were performed via custom-written programs in MATLAB (MathWorks, Natick, MA).

Training

Training sessions consisted of 50 trials and were designed to train the monkeys to respond to the fourth oval from the bottom of a five-oval array presented along the vertical midline of the screen (Figure 10a). Each oval measured 2.4 cm wide and 1.4 cm high. The bottom-most point of the bottom-most oval was 7 cm above the bottom of the screen, and there were 2.2 cm between oval centers. Each trial required initiation by touching a red rectangle presented in the bottom-center of the screen. The placement of this “start” stimulus was chosen to anchor monkeys to the bottom of the screen so that they would likely perceive the target oval as the fourth from bottom rather than second from top.

In the first phase the target oval (fourth from bottom) was bright green with three horizontal white stripes and in subsequent phases it became increasingly darker until it was solid black and matched the other ovals. In the first phase of training incorrect responses to other targets were recorded but did not end the trial and each trial continued until the monkey touched the target oval. In subsequent phases incorrect responses ended the trial. A touch was registered for a particular shape if it fell within the (invisible) rectangle bounding the top, bottom, left, and right border of that shape.

Correct responses elicited a 1-2 second green screen, a pleasant auditory signal, and the delivery of a miniature M&M or Reese’s Pieces candy. Incorrect responses elicited a blue screen presented for 2-4 seconds, an unpleasant auditory stimulus, and no food reward.

A performance criterion of 80% was required to advance from the first to the second phase of training. The performance criterion for advancement was then decreased because as training phases progressed, the target oval became less distinguishable from the surrounding ovals and trials became more difficult. The criterion for advancing from the second and third phases was 70%, and from the fourth and fifth phases 60%, accuracy in a single complete session. Once monkeys reached the final (sixth) training phase all five shapes were solid black ovals (Figure 10a). The criterion for advancing to testing was attainment of 50% accuracy during each of two consecutive sessions. Note that throughout all phases chance performance was 20%.

Ordinal Position Testing

Five test conditions were designed to assess the generality of the ordinal rule that the monkeys had learned. In each condition, two types of probe trials were intermixed with standard trials. The standard trials were identical to phase 6 training trials and were differentially reinforced (Figure 10a). The probe trials were nondifferentially positively reinforced: that is, whichever ordinal position the subject selected, the screen turned green, a pleasant sound was played, and a candy was delivered. Each test session began with a string of at least three standard trials; otherwise the standard and probe trials were randomly intermixed. Sessions consisted of 80 trials total: 48 standard trials,

16 of one probe type, and 16 of another probe type.

Two of the five conditions were designed to assess whether the rule the monkeys learned would transfer to physically distinct elements. In the ‘Color’ condition all of the stimuli were either red (‘red’ probe trials, Figure 10b) or blue (‘blue’ probe trials, Figure

10c), but remained oval-shaped and in the same location and size as standard trials. In the ‘Shape’ condition all of the stimuli were either rectangles (‘rectangle’ probe trials,

Figure 10d) or triangles (‘triangle’ probe trials, Figure 10e), but remained black and in the same location and size as standard trials.

Two of the five conditions were designed to distinguish whether the monkeys had learned the ordinal or physical position of the target oval. In the ‘Shift’ condition the array of five black ovals was shifted up (‘up’ probe trials, Figure 10f) or down (‘down’ probe trials, Figure 10g) on the screen by 2.2 cm with respect to standard trials such that the third or fifth oval, respectively, was in the same physical position as the target

(fourth) oval on standard trials. The size, shape, color, and spacing of the elements within the array were identical to standard trials. In the ‘Spaced’ condition the first oval at the bottom of the array was in the same position as in standard trials, but spacing among the five ovals was increased to 3.3 cm (‘sparse’ probe trials, Figure 10h) or decreased to

1.65 cm (‘dense’ probe trials, Figure 10i) between centers of adjacent stimuli such that the third or fifth oval was in the same physical position as the target fourth oval on standard trials, where distance between centers of adjacent stimuli was 2.2 cm. The size, shape, and color of stimuli within the array were identical to standard trials.

A final condition tested whether the monkeys were anchoring to the bottom or top of the array. In the ‘Length’ condition the stimulus array consisted of four (‘four’ probe trials, Figure 10j) or six (‘six’ probe trials, Figure 10k) ovals rather than five, but the size, shape, color, and spacing of stimuli within the array, as well as the distance from the bottom of the screen to the bottom of the array, were identical to standard trials.

Two of the monkeys were tested with four sessions on each of the first four conditions in a blocked fashion (they were not tested on the Length condition because that condition was added after they had completed the study). Given that we were most interested in performance during the first session of each condition, and performance had remained stable in each condition over the four sessions, the remaining two monkeys were tested with a single session on each of the five conditions. The order of conditions was randomized and different for each monkey.

Number-Space Mapping Testing

Each trial was initiated when the subject touched a red rectangle at the bottom- center of the screen. Standard trials were identical to the standard trials in ordinal position testing except that elements were circles rather than ovals so that they would appear the same whether the array was horizontal or vertical (Figure 11). Subjects received retraining sessions between the two test phases to familiarize them with the

circle stimuli. Each re-training session consisted of 50 trials in which five black circles

(1.75cm diameter, 3.0cm between adjacent centers) were presented along the screen’s vertical midline (the bottom of the bottom-most circle was 4cm above the lower screen edge; Figure 11a). Subjects received at least two sessions of retraining and proceeded from retraining to testing when they completed either one session with at least 80% correct (n=3), or five consecutive sessions with over 60% correct (n=1).

As in ordinal testing, during Number-Space Mapping testing positively reinforced probe trials were randomly intermixed with differentially reinforced standard trials.

Standard trials were identical to retraining trials: five circles appeared along the vertical midline, and subjects were rewarded for touching the fourth from the bottom (Figure

11a). In probe trials, the array was rotated 90° around its center, to display a horizontal line of five circles with size and spacing identical to standard trials (Figure 11b). Each session consisted of 48 vertical standard trials and 24 horizontal probe trials. Because we were interested in monkeys’ spontaneous behavior in the horizontal probes before they began building up a reinforcement history (i.e. before they discovered that touching any position in these trials would yield the same reward), only the first half of trials in each subject’s first Number-Space Mapping testing session were included for analysis.

Hand Preference Assessment

It is essential to rule out an alternative explanation for the monkeys’ preference to choose the circle second from the right. Monkeys may have had a side-bias in responding to the touch screen, perhaps due to preferential use of one hand or the other.

Such a scenario is unlikely, since rhesus macaques as a population do not exhibit

“handedness,” though they and other Old World monkeys may have a slight left-hand preference (McGrew & Marchant, 1997; Papademetriou et al., 2005). Nonetheless the possibility of a hand or side preference was tested in these subjects in the following three ways:

1) ‘Hand Used’: The hand each subject used for touching the screen during the experiments described above was recorded on one training session and subsequently observed throughout the study to monitor for changes.

2) ‘Basic Touch Screen Response Speed’: Monkeys B, C, and O participated in an additional touch screen task, where each trial consisted of a single black circle varying in size (1.5-3.5cm radius) and appearing at a random location on a white background.

Monkey Y was no longer available for testing at the time of this assay. The subject had to touch the circle to earn a candy reward and move on to the next trial. Subjects completed

50 trials each, and their response latencies (time from black circle appearing to the subject touching it) were recorded. Trials were coded as to whether the stimulus fell on the left or right half of the screen. For each of the two categories we coded whether the left or right hand was used to respond and mean response time.

3) ‘Fruit-in-Cup Test’: Monkeys B, C, and O participated in a cup- and fruit- grabbing task, in which chunks of dried fruit were placed at the bottom of a small paper cup. Monkey Y was no longer available for testing at the time of this assay. The inside of the cup was shown to each monkey while he was in one part of his enclosure, and then the cup was placed in another part of the monkey’s enclosure. The monkey was given access to this second area of his enclosure, and the hand he used to grab the cup as well as the hand he used to reach inside the cup, remove the fruit and eat the fruit were recorded. Each subject completed five trials (i.e. five cups).

2.2.2 Results

Ordinal Position Testing

On average subjects required 16 training sessions (range = 13 to 19) before moving to the test sessions. During training and every testing condition, subjects always completed the full number of trials constituting each session. One-tailed exact binomial tests were used to determine whether each position was selected above the 20% chance expectation. In the text and figures we report data on all subjects together; individual subjects’ accuracy and binomial test p-values are shown in Tables 6-19. Subjects selected the correct position on standard trials more than expected by chance from the first day of testing (70.31%, binomial p<0.0001) (Table 6). This performance was maintained throughout testing (73.50%, binomial p<0.0001) (Table 7). No other position was selected with frequencies above chance expectations (first: 0.81%, second: 2.66%, third: 12.85%, fifth: 9.49%; all binomial p’s>0.9). Chi-square tests with one degree of freedom confirmed that the fourth position was chosen with a different frequency than any other position

(first: χ2=977.5, p<0.0001; second: χ2=919.3, p<0.0001; third: χ2=647.7, p<0.0001; fifth:

χ2=729.0, p<0.0001).

Responding to the fourth ordinal position from the bottom was robust to changes in the appearance of the elements in the array. During nondifferentially reinforced probe trials, subjects continued to choose the fourth position from the bottom in both the Color

(Figure 12, Table 8 & Table 9) and Shape (Figure 13, Table 10 & Table 11) conditions, when the arrays were composed of red ovals (54.69%, binomial p<0.0001), blue ovals

(67.19%, binomial p<0.0001), black rectangles (71.88%, binomial p<0.0001), or black triangles (64.06%, binomial p<0.0001). None of the other positions exceeded chance level in any of the Color or Shape probe conditions (all binomial p’s>0.2). Chi-square tests with one degree of freedom confirmed that the fourth position was chosen at a significantly different level than any other position in each probe condition (all χ2’s>23.2, all p’s<0.0001). All individual monkeys followed this pattern, except for Monkey B who selected the third non-significantly more than the fourth position in triangle probes of the Shape condition (χ2=0.533, p=0.47).

100 Standard Red 80 Blue

40 Percent Selected

0 1 2 3 4 5 Position from Bottom

Figure 12: Performance in the Color condition. Dashed line represents chance expectation. Error bars represent standard error of the mean.

100 Standard Triangles 80 Rectangles

40 Percent Selected

0 1 2 3 4 5 Position from Bottom

Figure 13: Performance in the Shape condition. Dashed line represents chance expectation. Error bars represent standard error of the mean.

Table 10: Individual subjects’ performance during Rectangle probe trials. “First” through “Fifth” refer to stimulus locations in the array ordered from bottom to top. p- values refer to one-tailed exact binomial tests compared to the 20% chance expectation.

Table 11: Individual subjects' performance during Trangle probe trials. “First” through “Fifth” refer to stimulus locations in the array ordered from bottom to top. p- values refer to one-tailed exact binomial tests compared to the 20% chance expectation.

Additional transfer conditions revealed that subjects based their decisions upon the ordinal, rather than physical, position of the elements in the array. In the Shift

(Figure 14, Table 12 & Table 13) and Spaced (Figure 15, Table 14 & Table 15) conditions, the fourth oval moved to a different location on the screen and a different ordinal position occupied the physical location where the fourth oval appeared in standard trials. Subjects continued to select the fourth position, whether stimuli were shifted up

(68.75%, binomial p<0.0001) or down (54.69%, binomial p<0.0001), or spaced more sparsely (57.81%, binomial p<0.0001) or densely (64.06%, binomial p<0.0001) as compared with standard trials. The only other positions to approach statistical significance in any of the Shift or Spaced probe conditions were the fifth position in

Shifted Down trials (26.56%, binomial p=0.075) and the third position in Sparse trials

(28.12%, binomial p=0.042). Even so, in these and all other Shift and Spaced probe conditions, chi-square tests with one degree of freedom confirmed that the fourth position was chosen at a significantly different level than any other position (all

χ2’s>10.4, all p’s<0.01).

100 Standard Shifted Up 80 Shifted Down

40 Percent Selected

0 1 2 3 4 5 Position from Bottom

Figure 14: Performance in the Shift condition. Dashed line represents chance expectation. Error bars represent standard error of the mean.

100 Standard Sparse 80 Dense

40 Percent Selected

0 1 2 3 4 5 Position from Bottom

Figure 15: Performance in the Spaced condition. Dashed line represents chance expectation. Error bars represent standard error of the mean.

All individual monkeys followed this pattern, except that Monkey B chose the fifth (75.00%, binomial p<0.0001) and not fourth (18.75%, binomial p=0.402) position above chance level in Shifted Down trials. A few individual monkeys selected the third position above chance expectation, but not more than the fourth position, in certain conditions: during Dense trials for Monkey Y (50.00%, binomial p=0.0015; compared to the fourth position, χ2= 0.126, p=0.723); Sparse (50.00%, binomial p=0.0015; compared to the fourth position, χ2= 0.126, p=0.723), Shifted Up (37.50%, binomial p=0.0267; compared to the fourth position, χ2= 0.130, p=0.719), and Shifted Down (37.50%, binomial p = 0.0267; compared to the fourth position, χ2= 0.508, p=0.476) trials for

Monkey C; and Sparse trials for Monkey O (37.500%, binomial p = 0.0267; compared to the fourth position, χ2=1.129, p=0.288). The tendency to select the third position above chance level in the Dense and Shifted Down trials, as well as in the Triangle trials described for Monkey B above, may have resulted from the monkey accidentally touching the screen too early as his hand traveled up from the “Start” response. In other conditions, subjects’ behavior could have resulted from spatial coding. Monkey B chose the fifth position in Shifted Down trials, Monkeys C and O chose the third (but not as much as the fourth) position in Sparse trials, and Monkey C chose the third (but not as much as the fourth) position in Shifted Up trials. These aberrations suggest that subjects may have encoded the precise physical location of the target oval during Standard trials,

but that this geometric representation weakly conflicted with the ordinal numeric representation only in certain conditions for certain monkeys.

To examine whether monkeys anchored their responses from the bottom or the top of the array, we conducted the Length condition probe trials, where the total number of ovals in the array was reduced to four or increased to six. Results indicated that subjects chose both the fourth position from the bottom and the second position from the top more than would be expected by chance (Figure 16, Table 16 & Table 17). When there were four ovals (chance = 25%), subjects chose the second position from the top on

53.12% of trials (binomial p<0.0001) and the fourth position from the bottom on 37.50% of trials (binomial p=0.038). A chi-square test indicated that preference for the fourth from the bottom and the second from the top did not differ significantly (χ2=1.58, p=0.21). When there were six total ovals (chance = 16.67%), subjects chose the second position from the top on 43.75% of trials (binomial p=0.0001) and the fourth position from the bottom on 37.50% of trials (binomial p=0.0011). A chi-square test again indicated that preference for the fourth from the bottom and the second from the top did not differ significantly (χ2=0.26, p=0.61). Thus in both Length probe conditions subjects were equally likely to choose the fourth from the bottom and the second from the top.

100 Standard (Five Stimuli) 90 Less (Four Stimuli) More (Six Stimuli) 80

Percent Selected 30

0 1 2 3 4 5 6 Position from Bottom

Figure 16: Performance in the Length condition. Dashed lines represent chance expectation for each trial type. Error bars represent standard error of the mean.

An examination of response latencies on standard trials, however, suggests that subjects typically anchored their standard trial decisions from the bottom of the array.

Latency was measured as the elapsed time between the stimuli appearing (following the subject touching the “start” rectangle) and the subject touching one of the stimuli in the array. Mean latencies increased with position selected as follows (standard error in parentheses): selected position one, 0.37 sec (0.04); two, 0.70 sec (0.07); three, 0.90sec

(0.04); four, 0.88 sec (0.01); five, 0.96sec (0.03) (Figure 17). A one-way ANOVA revealed a significant effect of stimulus position on latency (F(4) = 6.82, p<0.0001). A linear regression on latency by stimulus position revealed a positive linear trend (β = 0.071, p<0.001). The location of the “start” rectangle at the bottom-center of the screen was selected specifically to anchor subjects to this position, and the pattern of response times confirms that this anchoring was successful. Although this anchoring was physical and we cannot rule out the possibility that monkeys may have encoded the sequence from top to bottom while simultaneously moving through space from bottom to top, it seems likely that subjects visually scanned the array from bottom to top and thus ordered items from bottom to top. Importantly, our main finding does not hinge on which interpretation is correct.

100

0.8

0.6

0.4 Mean Latency (sec)

0.2

0 1 2 3 4 5 Position from Bottom

Figure 17: Mean response latency to each position during Standard trials of ordinal position testing.

Number-Space Mapping Testing

Subjects required 2.75 sessions on average to meet retraining criterion (range = 2-

5 sessions). During testing, all subjects maintained their high performance on standard vertical trials, selecting the fourth position from the bottom on 81.25% of trials (Table

18), which was well above the 20% chance expectation (binomial p<0.0001). No other position was selected with above chance levels (all binomial p’s>0.99).

On average subjects selected the fourth position from the left (i.e. second from the right) on 60.42% of horizontal probe trials, substantially above the 20% chance expectation (binomial p<0.0001) (Figure 18, Table 19). No other position was chosen above chance expectation (all binomial p’s>0.3). Notably, the fourth position from the 101

right (i.e. second from the left) was only selected on 12.50% of trials. The monkeys’ preference for the fourth position from the left suggests that macaques map number onto space in a left-to-right orientation.

100 Standard (Vertical) Horizontal 80

40 Percent Selected

0 1 2 3 4 5 Position from Bottom or Left

Figure 18: Performance in Number-Space Mapping Testing. Dashed line represents chance expectation. Error bars represent standard error of the mean.

Table 19: Individual subjects’ performance during Horizontal probe trials. “First” through “Fifth” refer to stimulus locations in the array ordered from left to right. p-

102

values refer to one-tailed exact binomial tests compared to the 20% chance expectation.

Individually, three of the four subjects followed this pattern of choosing the fourth position from the left – and no other position – above chance level. The remaining subject, Monkey C, chose the third from the left as well as the fourth from the left above chance (third: 50.00%, binomial p<0.01; fourth: 50.00%, binomial p<0.01). Although this subject deviated somewhat from the common pattern, he did not behave in a manner consistent with a right-to-left oriented mental number line, as he never chose the fourth position from the right. Instead it seems that rotating the array more generally disoriented him and resulted in less precise responding.

One alternative explanation for our finding that monkeys chose the fourth position from the left is that the monkeys may have an inherent side bias. That is, perhaps some aspect of the experimental set-up or the monkeys’ predispositions influenced them to touch stimuli closer to the right-hand side of the screen. We believe that this is unlikely because this would have resulted in a preference for the spot furthest to the right (i.e., the fifth from the left); however, subjects selected that position on only

15.62% of trials on average, which did not differ from chance expectations (binomial

103

p>0.8). Nonetheless we assessed monkeys’ hand and side preferences in three different ways, described in Methods section 2.6.

Half of the subjects (n=2, Monkeys B & C) exclusively used their right hands and half (n=2, Monkeys O & Y) exclusively used their left hands during both the training and testing phases of the experiment. When tested with the basic touch-screen response assessment we found that all three monkeys that we were able to test were equally fast at responding to the left and right halves of the screen (two-tailed t-test, t145 = 0.791, p =

0.430 for all monkeys together; Monkey B: t47 = 0.356, p = 0.723; Monkey C: t47 = 0.949, p =

0.348; Monkey O: t47 = 0.652, p = 0.518). Finally we used a fruit cup test modeled after prior studies (e.g. the TUBE task: Hopkins et al., 2003; Hopkins, 2013). Such a task can tease out whether one hand is specialized for gross support (picking up the cup) while the other is specialized for fine manipulations like extracting the treat and moving it toward the mouth. The three subjects tested here each showed a different pattern of behavior in this task. Across the five trials, one subject (Monkey B) typically held the cup with his left hand and reached in with his right; another (Monkey O) typically held the cup with his right hand and reached in with his left; while the other (Monkey C) held the cup with his right hand on half the trials but his left on the other half, switching between the two during one trial, and used the opposite hand to reach into the cup.

Thus as a population, our subjects did not show a preference to use a particular hand for gross support or fine manipulation.

104

Together these three assessments indicate that monkeys in this study did not have an overall hand preference or a predisposition to respond to stimuli on the right side of the screen. These results concord with previous findings that while individual animals may prefer to use the left or right hand for particular tasks, at the population level rhesus macaques do not exhibit a strong hand preference (McGrew & Marchant,

1997; Papademetriou et al., 2005). Hence our findings in Number-Space Mapping testing do not reflect a simple side bias.

2.2.3 Discussion

We investigated whether rhesus monkeys, like humans, map number to space.

To test this question, we first trained monkeys to select a specific ordinal position in a vertical array of homogeneous items. All four monkeys rapidly learned to identify the fourth oval from the bottom of a five-oval vertical array. They maintained high performance when the color or shape of the stimuli was altered, indicating that their responses did not depend upon the particular appearance of the array elements.

Moreover, they continued to select the fourth item even when its physical location was changed. Their performance was robust to variations in inter-element spacing and to position of the array on the screen, showing that they were not relying upon distance from the end of the array or from the edge of the screen as a cue.

105

Our main finding was that on probe trials when the stimulus array was rotated ninety degrees to a horizontal line, monkeys preferentially selected the fourth position from the left. Such a result could emerge from three possible interpretations: 1) monkeys learned to respond to the fourth position from the bottom, and number is mapped onto space in such a way that biases them to orient from the left rather than the right; 2) monkeys learned to respond to the second position from the top, and number is mapped onto space in such a way that biases them to orient from the right rather than the left; 3) some monkeys follow the first pattern and others follow the second pattern. All three of these interpretations suggest a SNARC-like effect is present in monkeys. The experiment was designed to anchor monkeys’ responses to the bottom of the screen, so the first interpretation seems most likely, but the second interpretation remains possible. The third interpretation seems least parsimonious, as it would require that certain subjects anchored to the bottom of the array during training, and then precisely those subjects exhibited a left-to-right spatial mapping of number, while other subjects anchored to the top of the array during training, and then precisely those subjects exhibited a right-to- left spatial mapping of number. However, even in this unlikely scenario, rhesus macaques could be said to exhibit a SNARC-like effect with individual differences regarding the orientation of the number line. Thus the finding that monkeys consistently selected the same ordinal position in the rotated five-element array – whether this

106

position is considered the fourth from the left or second from the right – indicates that monkeys map number to space.

We ruled out an alternative explanation of a more general side bias in our participants. Monkeys in general do not exhibit a strong hand preference at the population level (McGrew & Marchant, 1997; Papademetriou et al., 2005). Moreover, we found no evidence for a consistent handedness preference or side bias in our monkeys.

There are unfortunately few prior studies of visual scene scanning preferences in monkeys, but those that exist have not revealed any consistent left-to-right or right-to- left viewing tendency (Fitch & Braccini, 2013). Thus it is unlikely that our results stem from simple side biases. Additionally, the prior task experience of the monkeys in this study could not have set up the spatial-numerical association observed here. All four subjects had participated in several previous experiments, but only two of these involved numerical judgments. In one study, Monkeys B and Y were presented with two stimulus arrays at the left and right of the screen, and were rewarded for touching the array with a greater number of dots (Jones et al., 2014). In another unpublished study,

Monkey B was presented with two stimulus arrays at the left and right or top and bottom of the screen, and was rewarded for making an eye movement toward the array with a greater number of dots. In both of these experiments, the location of the larger array was perfectly counterbalanced, so this task should not have caused subjects to associate either side with lesser or greater numerosities or with a higher reward

107

probability. The rewarded location was similarly counterbalanced in all other non- numerical eye movement, joystick, and touch-screen experiments in which these four monkeys were previously tested. Therefore our results cannot be explained by subjects’ task history.

Our results complement those of Rugani et al. (2007, 2010, 2011) in birds.

Nutcrackers and domestic chicks approached the fourth (or sixth, respectively) item from the left in a horizontally-oriented row of sixteen items following training to select the fourth (or sixth, respectively) item from the starting point in a sagitally-oriented line of sixteen items (Rugani et al., 2010). It is notable that we found a similar pattern in monkeys despite differences in the experimental paradigm: our subjects selected shapes on a touch screen while sitting in their home cages, whereas the birds walked through space; and for our subjects, the fourth item from the left actually fell on the right half of the screen, whereas the fourth (or sixth) item from the left was in the left half of space for the birds. The presence of the space-number mapping in all three species even with these experimental differences raises the question of whether similar neural mechanisms may underlie the effect in birds and monkeys – and, in turn, humans. Although birds’ visual systems and brains are generally far more lateralized than those of primates

(Larsson, 2013; Rogers et al., 2013), it may be that some of the hemispheric specialization for representing numeric and spatial information posited to produce the effect in chicks

108

(Rugani et al., 2011) persists in primates. Future studies in monkeys could investigate this possibility.

Several explanations have been proposed for the SNARC effect in humans, some of which emphasize specific cultural experiences. The verbal-spatial account proposes that particular words are associated with each other - “small” with “left” and “large” with “right” – and so the concept of “small” activates the concept of “left” and facilitates leftward responses for small values (Gevers et al., 2010). Such a mechanism requires use of language, which our monkeys do not possess. Similarly, reading or writing direction

(Shaki et al., 2009; Shaki & Fischer, 2008; Zebian, 2005) cannot contribute to a space- number mapping in monkeys. Another theory is that finger counting gives rise to spatial-numerical associations: individuals who begin counting on their left hand come to associate smaller numbers with the left side of space (Fischer & Brugger, 2011).

However, our monkeys have never been taught to count on their hands and more generally monkeys have not shown evidence of spontaneous finger enumeration. While verbal associations and finger counting or reading habits appear to modulate the

SNARC effect in humans, our finding in monkeys suggests that none of these is the primary driver. Instead explanations that do not require specific lifetime experiences are more likely, including the original idea of a spatial coding of number that exists in long- term memory (Dehaene et al., 1993), and the more recent working memory account which holds that a given ordinal position in working memory is associated with a

109

particular region of space (van Dijck & Fias, 2011). Additional investigations are needed to determine which of these latter mechanisms underlies the SNARC-like effect in monkeys.

In conclusion, our novel demonstration of a space-number mapping in a nonhuman primate suggests that the interplay between these two types of magnitudes in human cognition has a biological basis. Evidence that cultural factors can mediate the

SNARC effect (Fischer & Brugger, 2011; Gevers et al., 2010; Shaki et al., 2009; Shaki &

Fischer, 2008; Zebian, 2005) has raised the possibility that the interaction between number and space could be entirely learned. However, we have shown that monkeys – who do not read, write, or count on their hands and have no experience with items like rulers – also seem to associate number with space. Our results, along with those of de

Hevia and Spelke (2010) in human infants and Rugani et al. (2007, 2010, 2011) in birds, support the view that the predisposition to map number onto space is not a cultural invention. Instead this tendency has deep developmental and evolutionary roots.

2.3 The intraparietal sulcus may causally contribute to numerical processing: The impact of parietal rTMS on behavior and neuronal activity

Transcranial magnetic stimulation (TMS) consists of sending a brief current through a coil of wire held over the scalp to produce a magnetic field which passes through the skull undisturbed and induces a temporary electric field in the brain. Since

110

its advent nearly three decades ago (Barker, Jalinous, & Freeston, 1985), TMS has become increasingly popular as a non-invasive means for modulating brain activity in several different contexts. Cognitive neuroscientists use TMS to transiently alter activity in a specific brain region to assess the role of that region in a particular task or process

(Nevler & Ash, 2015; Sandrini, Umiltà, & Rusconi, 2011). Neurosurgeons implement

TMS for presurgical motor and language mapping in order to spare these crucial functions during removal of a tumor or epileptogenic focus (Eldaief, Press, & Pascual-

Leone, 2013; Picht, 2014). Psychiatrists treat patients suffering from drug-resistant major depressive disorder with TMS, resulting in reduction of depressive symptoms and in some cases complete remission (Connolly, Helmer, Cristancho, Cristancho, &

O'Reardon, 2012). In addition to depression, other ailments such as Alzheimer’s disease

(J. Lee, Choi, Oh, Sohn, & Lee, 2016), post-traumatic stress disorder (Berlim & Van Den

Eynde, 2014), schizophrenia (Slotema, Blom, van Lutterveld, Hoek, & Sommer, 2014),

Parkinson’s disease and other movement disorders (Kamble, Netravathi, & Pal, 2014), anxiety disorders, tinnitus, migraine, and chronic pain may be alleviated by TMS

(Wassermann & Zimmermann, 2012). It may augment psychotherapy (Bajbouj &

Padberg, 2014), aid in post-stroke rehabilitation (Allendorfer, Storrs, & Szaflarski, 2012;

Lefaucheur et al., 2014), attenuate cravings in addiction (Hanlon et al., 2015), and improve functioning in Autism Spectrum Disorders (Oberman, Rotenberg, & Pascual-

Leone, 2013). Finally, TMS holds promise as a tool to improve mental function in healthy

111

individuals (Bagherzadeh, Khorrami, Zarrindast, Shariat, & Pantazis, 2016; Horschig,

Zumer, & Bahramisharif, 2014; K. E. Hoy et al., 2015; Luber, 2014; Luber & Lisanby,

2013) as well as those with neuropsychiatric disorders (Demirtas-Tatlidede,

Vahabzadeh-Hagh, & Pascual-Leone, 2013). Despite these widespread applications, our understanding of the neural mechanisms by which TMS exerts its effects remains incomplete, thus limiting our ability to optimize its efficacy (Muller-Dahlhaus &

Vlachos, 2013). One of the largest gaps in our knowledge of TMS mechanisms is how it impacts neuronal activity to produce behavioral changes.

TMS is applied in discrete pulses that can vary in intensity, number, and pattern.

While single and paired pulses are useful for experimental manipulations, their effects seem to be transient. Longer trains of many TMS pulses delivered at specific time intervals, a class of protocols known as repetitive TMS (rTMS), are of greater therapeutic interest because they may have long-lasting effects (Aleman, 2013; Wassermann &

Zimmermann, 2012). In current practice, rTMS is delivered at a multitude of patterns, frequencies, intensities, and durations, with little consensus as to which parameters yield the strongest effects (Pell, Roth, & Zangen, 2011). Moreover, producing an improvement in behavior sometimes requires increasing and other times requires decreasing activity in a particular brain region. While motor cortex excitability, as measured by motor evoked potentials or resting motor threshold, has been shown to decrease following “low-frequency” (1Hz or lower) rTMS or a more complex pattern

112

known as continuous theta burst stimulation (cTBS) (Di Lazzaro et al., 2011), and increase following “high-frequency” (5Hz or greater) rTMS (P. B. Fitzgerald, Fountain, &

Daskalakis, 2006) or intermittent theta burst stimulation (Di Lazzaro et al., 2011), such findings have never been extended to other brain regions nor to higher-order cognitive functions.

The question of how rTMS impacts the brain has been approached from several angles (see (Chervyakov, Chernyavsky, Sinitsyn, & Piradov, 2015; Tang, Thickbroom, &

Rodger, 2015) for recent reviews). For instance, after a train of rTMS, a single pulse over motor cortex may produce a larger or smaller motor evoked potential (MEP) in the corresponding muscle, indicating an increase or decrease in cortical excitability (P. B.

Fitzgerald et al., 2006). rTMS may be combined with neuroimaging methods to evaluate changes in cerebral blood flow (functional magnetic resonance imaging (fMRI) and positron emission tomography (PET)), glucose metabolism (PET), or electric field potential on the scalp (electroencephalography (EEG)), revealing diverse activation patterns both at the site of stimulation and at more remote network nodes depending on stimulation protocol (Reithler, Peters, & Sack, 2011). Despite an abundance of such studies, no clear pattern has yet emerged linking the frequency of rTMS with its aftereffects (Veniero, Vossen, Gross, & Thut, 2015).

Depending on what tracer is used, PET can also map changes in neuromodulator levels. For example, high-frequency rTMS to left dorsolateral prefrontal cortex leads to

113

increased dopamine release in medial prefrontal regions (Cho & Strafella, 2009).

Additionally, pharmacological interventions combined with behavioral or MEP assessments can reveal whether certain chemicals or receptor types are necessary for rTMS to yield an effect (Kapogiannis & Wassermann, 2008), implicating for example N- methyl-D-aspartate (NMDA) receptors (Y.-Z. Huang, Chen, Rothwell, & Wen, 2007; Teo,

Swayne, & Rothwell, 2007). These methods in humans reveal large-scale brain dynamics associated with rTMS but generally have coarse spatial resolution and cannot directly address how rTMS affects neuronal action potentials.

Recent studies applying rTMS to slice cultures or intact animals have begun to give insight into the finer-grained impact of rTMS on individual neurons. For example,

Allen et al. (2007) applied short rTMS trains to visual cortex of anesthetized cats, and recorded single unit activity, local field potentials, tissue oxygenation, and hemodynamic responses, finding that all measures increased synchronously following stimulation in a dose-dependent manner (Allen, Pasley, Duong, & Freeman, 2007).

Tischler et al. (2011, 2012) implemented a novel mini-coil design that fits inside a nonhuman primate neural recording chamber and found that 60 or 100 low-frequency pulses over Macaca fascicularis motor cortex resulted in a brief increase followed by a longer-lasting decrease in cortical neuronal firing (Tischler et al., 2012; 2011). However, the much smaller size and round shape of their stimulating coil create a different electric

114

field distribution from standard human figure-eight-shaped TMS devices (Tischler et al.,

2011), so this paradigm is not directly comparable to human rTMS.

Vlachos et al. (2012) applied 10Hz repetitive magnetic stimulation to mouse entorhinal-hippocampal slice cultures and found increases in excitatory post-synaptic potential amplitude and frequency, as well as increased dendritic spine size and accumulation of GluA1-containing AMPA receptors, dependent upon NMDA receptor activation (Vlachos et al., 2012). This experiment points toward a long-term potentiation

(LTP)-like mechanism of synaptic plasticity following rTMS. Follow-up studies clarified that high-frequency rTMS could induce synaptic potentiation of excitatory synapses on proximal dendrites in hippocampal area CA1, dependent upon voltage-gated sodium channels, L-type voltage-gated calcium channels, and NMDA receptors (Lenz et al.,

2014); and that it could reduce the strength of GABA-ergic inhibitory synapses onto CA1 dendrites via structural changes in postsynaptic gephyrin scaffolds (Lenz et al., 2016).

However, it is unclear whether similar results would emerge in non-hippocampal brain regions, which are more accessible by rTMS in the intact brain. Changes in calcium release and in the expression of certain genes that modulate apoptosis and neurite outgrowth have been observed following low-intensity repetitive magnetic stimulation of cortical neuron cultures (Grehl et al., 2015). Still, the conclusions of such studies cannot necessarily be extended to human subjects. The effects of rTMS may be quite different in an interacting network of multiple neuronal nodes versus in a slice culture,

115

given the changes observed in brain regions distal to the stimulation site using neuroimaging as discussed above.

While observations in the cat (Allen et al., 2007; Aydin-Abidin, Moliadze, Eysel,

& Funke, 2006; Kozyrev, Eysel, & Jancke, 2014; Moliadze, 2003) offer an improvement in this regard, they are still limited by the fact that animals were anesthetized during stimulation and recording, which could have disrupted the balance between excitatory and inhibitory neurotransmission (Bestmann, 2008; Krnjevic, 1992) and do not allow assessment of the animals’ behavior. Indeed, studies in rats suggest that rTMS can have opposite effects on levels of the immediate early gene BDNF and the GluR1 subunit of the AMPA receptor depending on whether the animals were awake or anesthetized during stimulation (Gersner, Kravetz, Feil, Pell, & Zangen, 2011). One study in awake mice found an increase in the levels of two proteins involved in synaptic plasticity following five days of high-frequency rTMS to frontal cortex (Etiévant et al., 2015), but the impact of these changes on neuronal activity and behavior were not examined. Thus an important gap in our knowledge of rTMS is how it influences neuronal activity in awake subjects, and furthermore how these neuronal changes relate to behavioral outcomes.

Most studies of rTMS mechanisms to date have targeted the frontal or occipital lobe (Hoogendam, Ramakers, & Di Lazzaro, 2010; Reithler et al., 2011). This tendency may be due to the fact that the frontal lobe includes the dorsolateral prefrontal cortex, an

116

effective site for depression treatment (Hovington, McGirr, Lepage, & Berlim, 2013), as well as the motor cortex, where responses can be readily measured via MEPs (Di

Lazzaro et al., 2011; P. B. Fitzgerald et al., 2006). The occipital lobe includes the visual cortex, where responses can be measured by elicitation of phosphenes (perceived phantom spots of light) (Amassian et al., 1998). Interestingly, individual motor and phosphene thresholds are not correlated (Boroojerdi et al., 2002; L. M. Stewart, Walsh, &

Rothwell, 2001), underscoring that TMS effects may differ among brain regions. In contrast with frontal and occipital regions, less attention has been directed toward the effects of rTMS on the parietal lobe. The parietal cortex carries out many cognitive processes, including attentional reorienting, episodic memory encoding and retrieval, space and number processing, and even theory of mind (Cabeza, Ciaramelli, &

Moscovitch, 2012; Constantinidis, Bucci, & Rugg, 2013), making it a prime target for rTMS to produce cognitive enhancement. In particular, electrically stimulating the parietal cortex during learning may produce long-term changes in numerical skills

(Kadosh, Soskic, Iuculano, & Kanai, 2010), raising the possibility that rTMS might be a viable treatment option for mathematical disabilities such as acalculia and dyscalculia

(Ardila & Rosselli, 2002). Moreover, initial studies have suggested that stimulating regions of parietal cortex could relieve symptoms of depression (Schutter & van Honk,

2005; van Honk, Schutter, Putman, de Haan, & d'Alfonso, 2003) and Parkinson’s Disease

(van Dijk, Møst, Van Someren, Berendse, & van der Werf, 2009). Because of its potential

117

as a therapeutic target, the parietal cortex is an important region in which to evaluate the impact of rTMS.

While it is generally not possible to record intracranial electrical activity in humans, such recordings can be accomplished in our close evolutionary relatives, rhesus monkeys. In order to translate the results of an rTMS study in monkeys to humans, a behavior must be selected that monkeys and humans perform similarly, using similar cognitive architecture. One cognitive domain that fits this bill is numerical processing.

When comparing non-symbolic stimulus arrays based on numerosity, monkeys and humans exhibit comparable ratio-dependent performance. The studies presented in the previous two sections of this chapter on the ability of monkeys to compare ratios

(Drucker, Rossa, & Brannon, 2015b) and on the interaction between spatial and numerical processing in monkeys (Drucker & Brannon, 2014a) further demonstrate the similarities between the two species’ number systems. Moreover, the parietal cortex supports number processing in both species. In particular, the intraparietal sulcus has been repeatedly implicated in neuroimaging investigations of human number perception and manipulation (Nieder & Dehaene, 2009). Lesion (Dehaene & Cohen,

1997; Takayama, Sugishita, Akiguchi, & Kimura, 1994) and TMS (Sandrini & Rusconi,

2009) studies verify that this region plays a causal role in human number processing. For example, 15 minutes of offline 1Hz rTMS to left IPS slowed down response time in a nonsymbolic number comparison task (Dormal, Andres, & Pesenti, 2008), online triple-

118

pulse TMS to the right IPS impaired both automatic and intentional number processing

(Kadosh, Bien, & Sack, 2012). The IPS also seems to underlie numerical processing in rhesus macaques. Single-unit electrophysiology studies have reported that individual neurons in macaque ventral and lateral intraparietal areas represent numerosity in either a peaked (Nieder & Miller, 2004a) or monotonic (Roitman, Brannon, & Platt, 2007) fashion (see also (Nieder, 2012b). Area 5 in the anterior IPS contains neurons that are involved in producing a specific number of actions (Sawamura, Shima, & Tanji, 2002), and their inactivation impairs the production of the correct number of actions

(Sawamura, Shima, & Tanji, 2010). However, the necessity of macaque IPS for numerical processing distinct from action tracking has not yet been established. By manipulating rhesus macaques’ performance in a number task with rTMS to the IPS, we would demonstrate the causal contribution of this brain region to macaque numerical cognition. Such a finding would further validate the selection of this cognitive domain and brain region for an rTMS electrophysiology study in monkeys.

In order to assess the effects of rTMS to the IPS on number processing specifically, and not more general task demands, a structurally similar control task will be required that presumably relies on a different brain area. In the macaque, two functionally distinct visual processing pathways have been identified: the dorsal stream and the ventral stream (Milner & Goodale, 2008; Mishkin et al., 1983). The dorsal stream projects from the occipital to the parietal cortex and is specialized for spatial

119

information, while the ventral stream projects from the occipital to the temporal cortex and is specialized for object identification. Given that numerical cognition relies on the parietal cortex in humans, and that number is not a property of a single object but rather involves parsing several items distributed through space, number processing may be considered a dorsal stream function. Thus a process calling on the ventral stream would be desirable for a control task. One such function is color perception, which is carried out by occipital areas V1 and V2 as well as regions of the inferior temporal cortex

(Conway, 2009). Manipulating the parietal cortex therefore should not affect color processing, making color comparison a suitable control condition to isolate the impact of

IPS-rTMS on number comparison.

The combination of rTMS with intracranial electrophysiological recordings from an awake, behaving animal has, to our knowledge, never previously been accomplished.

There are several obstacles to such a combination, which our team has overcome

(Mueller et al., 2014). First, the circuitry and design of the recording headstage have been modified so it will not be damaged by the TMS-induced electric field, and so that the artifact produced by each TMS pulse in the recorded voltage trace will be minimized.

Importantly, the current induced by TMS in the recording electrode has been reduced so that it is well below the threshold for direct electrical stimulation of neurons. Previous in vivo experiments with TMS (Allen et al., 2007; Tischler et al., 2011; 2012) did not account for this possibility. Second, a novel figure-of-eight TMS coil has been constructed in

120

which the two loops are separated to allow space for the subject’s recording chamber, and angled downward to focus more precisely the induced electric field (Figure 19). This coil also includes independent switches to control the current direction in each of the two loops, enabling a realistic sham or “placebo” condition in which the net induced electric field under the coil’s center is null, but the auditory and tactile sensations produced by each pulse are comparable to regular stimulation. Third, we have developed a sequence of stages to habituate monkeys to rTMS so that they remain still during its application and are willing to perform an experimental task immediately afterwards. Fourth, we have adapted thermoplastic masks as a means for head restraint, allowing easier maneuvering of the coil on the head in the absence of an obstructive head-post (Drucker, Carlson, Toda, DeWind, & Platt, 2015a) (see Appendix). These methodological advances will enable us to assess behavioral performance in a numerical task and record neuronal activity from macaque IPS before, during, and after rTMS.

121

2.3.1 Experiment 1: rTMS and Behavior, Simultanous Presentation of Stimuli

To establish feasibility of influencing monkeys’ cognitive performance using

TMS, we first used a task structure that was already familiar to one of our subjects

(described below) and asked whether offline rTMS would alter her subsequent behavior.

We predicted that active, but not sham, low-frequency rTMS over parietal cortex would reduce accuracy or speed of responses during numerical decisions. To tease apart whether rTMS altered the subject’s representation of numerosity, or more general

122

decision- or task-related factors, we included a control task in which decisions were based on color.

Methods

Subject

The subject was one adult female rhesus macaque (Macaca mulatta, Monkey S).

She had previously been trained to use a touch-screen, and had participated in multiple studies involving numerical judgments (Cantlon & Brannon, 2007b; Jones et al., 2010;

Jordan & Brannon, 2006; Merritt, Rugani, & Brannon, 2009). She was socially housed, and was kept on a water-restricted diet to increase motivation to earn liquid reward in the experimental task.

Task

The subject sat in a primate chair (Crist Instruments, Damascus, MD) with an open slot in front allowing her arms to emerge. She was head-restrained using a thermoplastic mask (Drucker et al., 2015a) (see Appendix) during training and experimental sessions. Visual stimuli were presented on a 15-inch LCD touch-sensitive monitor (Elo TouchSystems, Menlo Park, CA) within her arms’ reach. The subject was trained to perform self-paced, interleaved number and color comparison tasks (Figure

20), which ran on a custom-built Matlab (Mathworks, Natick, MA) script using they

123

Psychophysics Toolbox add-on (D. H. Brainard, 1997) (https://psychtoolbox.org). To initiate each trial, the subject touched a green square at the bottom-center of the otherwise black screen. Immediately following this touch, two arrays of dots appeared at the upper left and right of the screen.

On half the trials, all the dots were yellow, which cued the monkey to choose the array with the greater number of dots (number comparison task). Numbers ranged from

8 to 32, and included the nearest whole numbers to each value of 2n/6, where n took on all whole numbers from 18 to 30, thus yielding twelve different approximate ratio values between any two numbers. The size and spacing of dots in the arrays was varied systematically so that visual parameters that typically covary with number, such as total surface area, were not reliable cues to numerosity (DeWind, Adams, Platt, & Brannon,

2015). Dot arrays were generated randomly each day according to number, size, and spacing parameters in order to minimize the opportunity for memorization of specific stimulus pairs.

On the other half of trials, the dots were a different color on each side ranging from red to blue, which cued the monkey to choose the array with the redder dots (color comparison task). Six different colors were used, and all were equiluminant within 0.1 foot-Lamberts, as measured by a Spyder calibration device (Datacolor, Lawrenceville,

NJ). The same array parameters (number, size, and spacing) were used to generate stimuli for the color comparison as for the number comparison task

124

In both tasks, stimuli stayed onscreen until the subject touched one of the arrays.

If she chose correctly, i.e. the array with the greater number or redder color of dots, she received a drop of juice and the screen turned red for a one-second inter-trial interval

(ITI); if she chose incorrectly, she received no juice and the screen turned blue for a one- second ITI. Trials from the two tasks were randomly intermixed, and the side of correct response was counterbalanced.

Figure 20: Simultaneous-presentation interleaved number and color comparison tasks, performed on a touch-screen. See text for additional description.

TMS

Each day the subject first completed one hundred trials of the task as a baseline.

Then the task was paused and rTMS was delivered over either left or right parietal cortex. A custom-built figure-eight coil (Mueller et al., 2014) was held tangentially to the

125

scalp such that one wing was more anterior than the other, and the coil’s wires extended dorsolaterally away from the monkey. The initial site of stimulation was approximated by lab members familiar with the location of macaque IPS, and marks were made on the subject’s thermoplastic mask to attempt to replicate this position each day. Pulses were delivered at 1 Hz for 10 minutes (600 pulses total) at 60% stimulator output from a

Magstim Super Rapid2 unit. Stimulation could be either active or sham: during sham stimulation, current around one of the coil’s wings was run in the opposite direction from active stimulation, so that the electric fields induced by each wing beneath the coil’s center would cancel each other out, producing negligible electrical stimulation but a similar noise and vibrotactile sensation at the same site on the head. Following rTMS, the task was resumed for one hour. Twelve total sessions were completed, with three in each condition of the 2 X 2 (stimulation type: active/sham; hemisphere: left/right) design.

Analysis

On every trial, the timing and side of the subject’s choice was recorded. The post- treatment session was divided into five-minute epochs, and for each session, a normalized accuracy (proportion of correct choices in that epoch for each task, divided by the proportion of correct choices during the pre-stimulation baseline for that task) and normalized response time (RT; mean latency between the appearance of the dot arrays and the subject’s selection for each task, divided by the pre-stimulation baseline

126

mean RT in each task) was computed for each epoch. An ANOVA was run in each epoch on proportion of correct choices (accuracy) by stimulation type and task (number or color); another ANOVA was run on response time (RT, the latency between appearance of the dot arrays and the subject’s selection) by the same factors. (Due to the small number of sessions and the lack of expectation that side of stimulation should exert an effect, data from both left and right hemisphere stimulation were combined.)

Results

During the first five-minute epoch following stimulation, accuracy was lower in the number task following active as compared with sham rTMS, but higher in the color task following active as compared with sham rTMS (Figure 21). The interaction between task and stimulation type was significant (p = 0.022); there were no main effects of stimulation type or task (p>0.1). There were no main effects of stimulation type or task, nor an interaction, on RT. Although the pattern of reduced accuracy in the number task following active rTMS persisted through the second, third, and fourth epochs, there were no main effects of stimulation type or task, nor any ineractions, on accuracy or RT in any subsequent epoch. We additionally examined whether accuracy or RT varied by the ratio between the two stimuli during the number task. As expected, accuracy increased and RT decreased as the ratio increased. At the hardest ratios, accuracy was

127

lower following active rather than sham rTMS, but with the relatively small number of sessions reported here, this interaction failed to reach statistical significance.

Figure 21: Choice accuracy and response time in the simultaneous-presentation number and color comparison tasks following low-frequency offline rTMS.

Discussion

We found that low-frequency rTMS delivered to parietal cortex could impair performance in a number comparison task. For a period shortly after offline rTMS delivery, accuracy in the number task was lower following active as compared with sham stimulation. In contrast with several findings in humans, where low-frequency rTMS during number processing slowed down responses (Andres, Pelgrims, Michaux,

Olivier, & Pesenti, 2011; Dormal et al., 2008; Salillas, Basso, Baldi, Semenza, & Vecchi,

2009; Sandrini & Rusconi, 2009), response times did not differ significantly following active and sham stimulation in our task. An effect on accuracy alone has been reported in some human studies (Dormal, Andres, & Pesenti, 2012), and we suspect that task

128

demands and motivational states may shift the speed-accuracy tradeoff. Our monkey subjects tend to respond as quickly as possible in order to earn juice rewards at a high rate; thus it might be expected that, in the face of impaired number processing, they would continue to attempt to respond at the same speed but would respond less accurately.

Surprisingly, this effect was reversed for the color comparison task: accuracy was lower following sham as compared with active stimulation. It may be that sham stimulation—in which an electric field is induced below the periphery of our TMS coil— targeted a region of the processing stream that is crucial for color perception. This hypothesis will need to be tested in future studies where regions of the ventral processing stream are targeted, and/or where neuronal activity is directly measured in different areas under the coil during active and sham stimulation. Regardless, this finding speaks to the nature of the impairment in the number task. If rTMS over parietal cortex interfered with general task processes, such as attention to or vision of the screen, or with the ability to translate a decision about two alternatives into a motor action, we would have expected a reduction in accuracy in the color task following active stimulation. Thus the impairment we observed during the number task points toward an interference with numerical processing in particular.

129

2.3.2 Experiment 2: rTMS and Behavior, Sequential Presentation of Stimuli

The pilot study reported in Experiment 1 indicated that offline low-frequency rTMS can influence subsequent performance in a cognitive task in monkeys. In particular, when delivered over parietal cortex, active stimulation reduced accuracy in a number comparison task, whereas sham stimulation reduced accuracy in a color comparison task. Motivated by these results, we altered the task paradigm to be more amenable to electrophysiological recordings, so that we could investigate the neuronal changes involved in these behavioral changes. In particular, we required subjects to fixate their gaze centrally during stimulus presentation so that eye position would not influence the signal recorded from parietal cortex; and we presented stimuli sequentially and introduced a delay before the subject’s response so that we could investigate the encoding of each stimulus and their comparison separately, as in previous studies of number comparison (Nieder & Miller, 2004b) and the vibrotactile frequency experiments discussed in the Introduction. We predicted that, active rTMS would decrease accuracy and/or increase response time in the number task as compared with sham rTMS or no stimulation, and it would either leave the color task unaffected, or improve performance

(increase accuracy or decrease response time) as in Experiment 1. We also recorded where the subject touched the screen for each response as a proxy for arm movement, and predicted that rTMS would not affect arm movements in either condition.

130

Methods

Subject

The subject was one adult female rhesus macaque (Macaca mulatta, Monkey F).

She has previously participated in multiple studies involving numerical judgments using a touch-screen (Cantlon et al., 2016; Cantlon & Brannon, 2005; 2006b; 2006a; 2007a;

2007b; Jordan & Brannon, 2006), and had received training to make eye movements to earn a reward, as well as training in a sequential-presentation numerical match-to- sample task with saccadic responses. She was socially housed during part of the study, and was kept on a water-restricted diet to increase motivation to earn liquid reward in the experimental task. During a sterile surgery (see section A.2.5) she had been implanted with a recording chamber centered over her right IPS (Figure 22).

Figure 22: MRI of Monkey F showing the intraparietal sulcus (within green circle) and the implanted recording chamber (white cylinder above the green circle).

131

Task

23), which ran on a custom-built Matlab (Mathworks, Natick, MA) script using they

Psychophysics Toolbox add-on (D. H. Brainard, 1997) (https://psychtoolbox.org). Gaze position was sampled at 1000 Hz using an Eyelink 1000 infrared eye-tracker (SR

Research, Mississauga, ON, Canada).

Before each trial began, the background of the screen was either yellow or teal, indicating which task—number or color, respectively—was forthcoming. A white square that was 2 degrees of visual angle per side was at the top-center of this colored screen.

To initiate a trial, the subject fixated her gaze within 4 degrees of this square. After 100 milliseconds (ms), the fixation square shrank to 1 degree per side, and the background turned black for another 200ms. While the subject maintained central fixation, an array of dots appeared to the left or right of the fixation point for 250ms, then disappeared for

100-250ms before another array of dots appeared on the other side of the fixation point

132

for 250ms. When the second array disappeared, the subject had to maintain fixation for another 100-250ms before two green squares appeared at the left and right of the screen.

At this point the subject could break her gaze fixation and touch one of the squares with her hand to indicate her response. If she touched the correct side (described below), she received a juice reward and the screen went black for a one-second ITI before the next trial could begin. If she touched the incorrect side, no juice was delivered and the screen turned gray for one second to signal the error, and then went black for the one-second

ITI.

As in Experiment 1, in the number comparison task, the subject was required to choose the side where the greater number of dots had appeared. Dot stimuli were generated according to the same parameters as in Experiment 1, with numbers ranging from 8 to 32 and non-numerical visual features varied so as not to be reliable cues to numerosity. Unlike in Experiment 1, dots in the number trials took on the same colors as in the color trials, in order to make the appearance of each individual stimulus as similar as possible between the two tasks, but both arrays within the trial were the same color.

We also restricted stimulus selection to four numerical ratios, approximating as closely as possible the ratios 1:211/6, 1:2, 1:21/2, and 1:21/6. During the color comparison task, the subject had to select the side where the redder dots had appeared; both arrays used the same layout (i.e. the same number, size, and spacing) of dots. Six equiluminant colors ranging from red to blue were used. Note that the two colored backgrounds used as

133

task-type cues at the beginning of each trial were also equiluminant. Trials from the two tasks were randomly intermixed, and the side of correct response as well as whether the first or second stimulus was correct were counterbalanced.

Figure 23: Interleaved sequential-presentation number and color comparison task. See text for additional description.

TMS

Each day the subject first completed fifteen minutes of the task as a baseline.

Then the task was paused for twelve minutes, during which time active rTMS, sham rTMS, or no treatment (“pause” condition) was delivered. The custom-built figure-eight coil (Mueller et al., 2014) was affixed around the subject’s recording chamber, which was centered over her right intraparietal sulcus, using a Flexbar (Long Island, NY)

134

positioning arm attached to the primate chair. The coil was oriented similarly to

Experiment 1, with the wings anterior and posterior to one another, and the wires extending medially from the subject. A mark was made on the subject’s recording chamber to facilitate aligning the coil similarly each day. Pulses were delivered at 1 Hz for 10 minutes (600 pulses total) at 60% stimulator output from a Magstim Super Rapid2 unit. Following rTMS, the task was resumed for one hour. Twenty total sessions were completed in a randomized order: ten with active stimulation, five with sham stimulation, and five with no stimulation.

Analysis

On every trial, the timing and side of the subject’s choice was recorded, as well as the precise location of her touch on the screen. The post-treatment session was divided into two seven-minute epochs, “early” and “late”, since the effect of ten minutes of low- frequency rTMS is expected to drop off within ten minutes post-stimulation (Sandrini et al., 2011), and roughly 2-3 minutes elapsed between the end of the stimulation train and resumption of the task. We ran a generalized linear mixed-effects model separately for each of the four outcome variables: accuracy (correct/incorrect, using a binomial model), response time (RT), horizontal touch location, and vertical touch location; with the random-effects predictor session number, and the fixed-effect predictors condition

(active rTMS/sham rTMS/pause), side of correct response (left/right), time in session

135

(early/late), and—only for analyses of accuracy and RT—task (number/color) and difficulty (1-4, 4 being the hardest, determined by numerical ratio or the ordinal distance between colors). We excluded these last two factors from analyses of touch location because we did not have any predictions about, or interest in, an interaction between touch location and the specific task at hand; rather, the purpose of these analyses was to determine whether rTMS altered motor output. For analyses of RT and touch location, only correct trials were included.

Results

Contrary to our predictions, condition did not have a significant main effect on accuracy (Figure 24), nor any significant interactions with task type, time in session, or difficulty (all p’s > 0.2). There was a significant main effect of task type on accuracy (beta

= 3.213, p < 0.0001), indicating that the subject performed better in the number than the color task; a significant main effect of difficulty on accuracy (beta = -0.196, p = 0.017), and an interaction between task type and difficulty (beta = -0.762, p < 0.0001), indicating that, in the number task, accuracy decreased as difficulty increased. There were also significant interactions between task type and correct side (beta = -2.173, p = 0.0028) and among task type, correct side, and difficulty (beta = 0.473, p = 0.021), and a marginally significant interaction between correct side and time in session (beta = 1.262, p= 0.089).

There were no significant main effects of correct side or time in session.

136

Condition did have a significant main effect on response time (Figure 25), but only between the pause condition and active rTMS (beta = -0.182, p = 0.029). Response times were not significantly different between active and sham rTMS (beta = 0.0384, p =

0.68). Similarly, there was a condition by task type interaction, but only between pause and active rTMS (beta = 0.335, p = 0.0014), and not between active and sham rTMS (beta

= -0.0207, p = 0.86). Several of the higher-order interactions involving condition were also significant (condition X difficulty: beta = 0.046, p = 0.035; condition X task type X difficulty: beta = -0.084, p = 0.0082; condition X task type X side correct: beta = -0.407, p =

0.0080; condition X correct side X difficulty: beta = -0.0651, p = 0.040; condition X task type X correct side X difficulty: beta = 0.121, p = 00094) or marginally significant 137

(condition X task type X correct side X time in session: beta = 0.487, p = 0.088; condition X correct side X time in session X difficulty: beta = 0.109, p = 0.068; condition X task type X correct side X time in session X difficulty, beta = -0.145, p = 0.092), but again, only for the difference between pause and active rTMS, and not for the difference between sham and active rTMS (all p’s > 0.2). There were also significant main effects of task type (beta = -

0.1223, p = 0.047), indicating quicker responses in the number task, and correct side (beta

= 0.381, p < 0.0001), indicating quicker responses to the right. In addition, there was a significant interaction between task type and difficulty (beta = 0.0375, p = 0.045) and a marginally significant interaction between task type, correct side, and difficulty (beta = -

0.047, p = 0.082). There were no signifcant main effects of time in session (beta = 0.0477, p

= 0.58) or difficulty (beta = -0.00994, p = 0.45) on RT.

Figure 25: Mean response time in the number and color tasks as a function of trial difficulty (4 being the hardest) following treatment. Colors represent the different

138

treatments (see legend), points show actual data and lines show fits from the GLM, and the different plots show performance in each task early and late in the session for trials where the left or right stimulus was correct (see titles).

Touch location, however, was significantly affected by condition (Figure 26). For vertical touch location, there was a significant main effect between active and sham rTMS (beta = 8.303, p = 0.020), although not between active rTMS and pause (beta =

1.916, p = 0.59). All other main effects and interactions were also significant: main effect of correct side (beta = 3.255, p < 0.0001), main effect of time in session (beta = 7.791, p <

0.0001), interaction between active v. sham rTMS and correct side (beta = -10.27, p <

0.0001), interaction between active rTMS v. pause and correct side (beta = -7.204, p <

0.0001), interaction between active v. sham rTMS and time in session (beta = -10.63, p <

0.0001), interaction between active rTMS v. pause and time in session (beta = -4.172, p =

0.0074), interaction between correct side and time in session (beta = -8.585, p < 0.0001), interaction between active v. sham rTMS, correct side, and time in session (beta = 5.999, p = 0.0083), and interaction between active rTMS v. pause, correct side, and time in session (beta = -8.342, p = 0.00014). For horizontal touch location, there was a marginally significant main effect between active rTMS and pause (beta = -6.495, p = 0.084), but no main effect between active and sham rTMS (beta = 3.356, p = 0.38). However, there was a significant interaction among active v. sham rTMS, correct side, and time in session (beta

= 11.47, p = 0.00082), and among active rTMS v. pause, correct side, and time in session

(beta = 10.12, p = 0.0022). There was also a significant main effect of time in session (beta

139

= 9.880, p < 0.0001), and significant interactions between active rTMS v. pause and correct side (beta = 6.171, p = 0.00056) and between correct side and time in session (beta

= -6.615, p = 0.00070). The other three interactions (sham v. active rTMS with correct side; sham v. active rTMS with time in session; and active rTMS v. pause with time in session) were not significant (p > 0.2).

140

Discussion

In our sequential-presentation number and color comparison tasks, we found no impact of offline low-frequency active rTMS, as compared with sham rTMS, over the intraparietal sulcus on choice accuracy or response time. We did find that response times were slower following active rTMS than no treatment (the pause condition), but since there was no significant difference between response times following active and sham rTMS, this impairment likely results from a general response to the experience of rTMS—i.e. having the coil placed on one’s head; hearing a clicking noise; feeling a vibration—rather than from the electric field induced by rTMS.

The lack of impact of rTMS on choice accuracy stands in contrast to the results of

Experiment 1. There are several differences between the methods employed in these two experiments that might contribute to these varying results. First, the task now requires gaze fixation at a central point of the screen, so the subject can no longer foveate the dot arrays. However, since the subject was still able to perform the task with a high degree of accuracy, and since she showed a normal ratio effect in the number task (i.e. accuracy decreased as the ratio between the numbers decreased), we believe she was able to fully perceive the dot stimuli and this change should not have altered the effect of rTMS.

Second, stimuli are now presented sequentially rather than simultaneously, and responses occur after a delay when the stimuli are no longer visible. This introduces a working memory component to the task: the subject must hold some representation of

141

the first stimulus in working memory, and then compare the second stimulus to this representation of the first stimulus, and hold this comparison in working memory until the response targets appear. However, these additional cognitive components should, if anything, provide more opportunities for rTMS to disrupt processing, rather than fewer.

A third methodological change that we believe is more likely to have produced the pattern of results observed here is that the subject now had a cranial implant. That is, a cylindrical recording chamber was placed over her intraparietal sulcus, and bone cement was placed around this chamber to hold it in place, anchored by titanium screws in her skull. Titanium is weakly magnetic and might exert a small influence on the TMS- induced electric field. More importantly, the bone cement introduced a displacement of over a centimeter between the TMS coil and the subject’s scalp (whereas in Experiment

1, the only displacement was a few millimeters due to the thermoplastic mask). The induced electric field drops off with distance from the TMS coil (Figure 27). Thus it is probable that the TMS-induced electric field in our set-up did not reach the deep parts of the IPS (namely, LIP, which is at a minimum about 6-7mm deep, and VIP, which is at a minimum about 8-9mm deep) thought to be involved in numerical representation or in decision-making.

142

Figure 27: Electric field induced by custom-built chamber-centric TMS coil at 100% stimulator output, according to a model (solid line) and measurements in a spherical saline flask (dotted line). Adapted from Mueller et al. (2014) with permission from the publisher.

Further support for this idea comes from the fact that rTMS did have an effect on touch location. More dorsal parts of the parietal cortex, namely the parietal reach region

(PRR) which encompasses areas 5v, MIP, and V6A, are involved in planning arm movements toward a target (Battaglia-Mayer & Ferraina, 2000; Galletti, Fattori, Kutz, &

Battaglini, 1997; Hwang, Hauschild, Wilke, & Andersen, 2012; Snyder, Batista, &

Andersen, 1997). It seems that rTMS did influence the activity of these reach-related regions during our task. Note that we stimulated the right hemisphere, and this subject used her right hand to respond. While it may seem unexpected, therefore, that stimulation would have had any effect on arm movements, this effect could have occurred due to network-level effects of rTMS (Reithler et al., 2011). For instance, perhaps a decrease in activity of the right PRR caused a compensatory increase in 143

activity of the left PRR. If activation of the left PRR increased following rTMS to right

PRR, this could explain why reaches with the right hand toward the left target were actually hypermetric in the early phase following stimulation, rather than hypometric as is typically observed in optic ataxia resulting from parietal lesions (Blangero et al., 2010;

Hwang et al., 2012). Such an interhemispheric activity pattern is currently speculative, however, and would need to be confirmed in future electrophysiological studies in which rTMS is applied to one hemisphere while activity is recorded from the corresponding region in the opposite hemisphere.

2.3.3 Experiment 3: rTMS and Electrophysiology

In Experiments 1 and 2, we demonstrated that offline low-frequency rTMS applied to macaque parietal cortex could alter behavior, albeit differently in the two paradigms. These results raised the question of how rTMS was changing neuronal activity in order to produce changes in behavior. Thus in experiment 3, we recorded single-unit and multi-unit activity, as well as local field potentials, in the IPS before and after low-frequency active or sham rTMS. Unfortunately, due to an unexpected failure of the TMS coil, only a small number of recordings were completed. We present those recordings here as pilot data, but future studies will be needed to fully explore how rTMS affects neuronal activity.

144

Methods

Subject

The subject was the same adult female rhesus macaque that had participated in

Experiment 1, Monkey S. During this experiment she was singly-housed, and on water restriction to increase motivation to perform calibrating eye movements and to undergo training in the task described in Experiment 2. In a sterile surgery (see section A.2.5) she had been implanted with a recording chamber centered over her right IPS (Figure 28).

Figure 28: MRI of Monkey S showing the intraparietal sulcus (within green circle) and the implanted recording chamber (white cylinder above the green circle).

Neurophysiological Recordings and TMS

The subject sat in a primate chair (Crist Instruments, Damascus, MD) and was head-restrained using a thermoplastic mask (Drucker et al., 2015a) (see Appendix). At

145

the beginning of each session, the custom-built figure-eight TMS coil (Mueller et al.,

2014) was affixed around the subject’s recording chamber, which was centered over her right IPS, using a Flexbar (Long Island, NY) positioning arm attached to the primate chair. The coil was oriented orthogonally to Experiments 1 and 2, with the wings medial and lateral to one another, and the wires extending anteriorly from the subject. This change in procedure was necessary due to the geometry of the recording set-up; between the subject’s implant, which involved a thick layer of bone cement, the A mark was made on the subject’s recording chamber to facilitate aligning the coil similarly each day. Each day an electrode—either single contact or 8-contact in a two-tetrode configuration—was introduced into the subject’s parietal cortex using sterile procedures

(see section A.2.5). The electrode was advanced until clear isolation was achieved for a single neuron on at least one channel. The spiking activity of any isolated neurons, as well as the background multi-unit activity and the local field potential on every contact, were recorded for one hour. Then the recording was paused, the amplifier was turned off, and the wires for signal, reference, and ground were unclipped from the electrode to avoid creating an inductive loop. Then active or sham rTMS was delivered at 1 Hz for 10 minutes (600 pulses total) at 60% stimulator output from a Magstim Super Rapid2 unit.

Following rTMS, the recording was resumed for at least one hour. Six sessions of active rTMS and three sessions of sham rTMS were completed.

146

Analysis

Waveforms from any well-isolated single unit were re-sorted offline, resulting in

9 single units from 4 distinct sessions of active rTMS and 3 single units from 2 distinct sessions of sham rTMS. For sessions using a multi-contact electrode, only one MUA signal and one LFP signal were analyzed, chosen based on the deepest channel from which a clear signal was recorded, resulting in 6 multi-unit and LFP signals from 6 distinct sessions of active rTMS and 3 multi-unit and LFP signals signals from 3 distinct sessions of sham rTMS. Firing rates from the single- and multi-unit signals were examined. LFP’s were transformed using Fourier analysis and the average power in two-minute time blocks before and after stimulation were computed for active and sham rTMS in each of seven frequency bands: delta (0.5-4Hz), theta (4-7Hz), mu (8-12Hz), alpha (12-15Hz), beta (16-31Hz), low gamma (32-60Hz), and high gamma (60-90Hz).

Power was normalized to the pre-stimulation mean, and an ANOVA was run on each frequency in each block to determine whether its power differed from zero (baseline).

Results and Discussion

No consistent effect of active versus sham rTMS was observed in the multi- or single-unit firing rates (Figure 29 and Figure 30). In individual sessions, however, large increases or decreases of firing rate were observed (Figure 31 and Figure 32). At this point there are not enough data points to draw conclusions about what may be driving

147

these variable effects. However, we suspect that depth of recording or day-to-day variation in the precise placement of the TMS coil may play a role. Additional recordings will be needed to explore these possibilities.

Figure 29: Multi-unit activity following active (red) and sham (blue) low-frequency rTMS, averaged across sessions.

Figure 30: Single-unit activity following active (red) and sham (blue) low-frequency rTMS, averaged across sessions.

148

6.73 3.365 3.62 2 1.4 1.3

1.2 1.5 1.2 1.1 1 1 1 0.5 0.8 0.9

0 0.8 0 50 100 0 50 100 0 50 100

5.493 0.953 3.1294 2 2.5 1.3

2 1.2 1.5 1.5 1.1 1 1 1 0.5 0.5 0.9

0 0 0.8 0 50 100 0 50 100 0 50 100

2.8176 4.453 7.2515 6 4 2.5

2 3 4 1.5 2 1 2 1 0.5

0 0 0 0 50 100 0 50 100 0 50 100

149

3.62 3.62 3.62 5.493 0.953 2 3 3 3 3

1.5 2 2 2 2 1 1 1 1 1 0.5

0 0 0 0 0 0 50 100 0 50 100 0 50 100 0 50 100 0 50 100

0.953 3.1294 3.1294 3.1294 2.8176 4 4 6 6 2

3 3 4 4 1.5 2 2 2 2 1 1 1

0 0 0 0 0.5 0 50 100 0 50 100 0 50 100 0 50 100 0 50 100

4.453 4.453 7.2515 4 6 3

3 4 2 2 2 1 1

0 0 0 0 50 100 0 50 100 0 50 100

In the local field potentials, it seems that power in the lower frequencies (0.5-

10Hz) may have been slightly higher following real versus sham stimulation (Figure 33).

However, again, there are too few data points at present to draw firm statistical conclusions from these results. When examining power in each frequency band individually (Figure 34), it appears that power in the lower frequencies (delta, theta, and mu bands) slightly decreased following active stimulation, whereas power in the higher frequencies (beta, low gamma, and high gamma) slightly increased following sham stimulation, but in the current dataset these results are not statistically significant.

150

Additional recordings will need to be completed to confirm these effects and determine their dependence on such factors as recording depth and coil orientation.

151

Figure 34: Mean normalized power in each frequency band (colors as in legend) before and after active (top) and sham (bottom) low-frequency rTMS.

2.3.4 General Discussion

In the three experiments presented here, we have laid the groundwork for important investigations into the neuronal changes underlying TMS-induced behavioral effects. In Experiment 1, we demonstrated that offline low-frequency active, as compared with sham, rTMS of macaque intraparietal sulcus can impair performance in a number comparison task, while having the opposite effect on a control color comparison task. This study established the feasibility of using the macaque model to investigate rTMS effects in a complex cognitive task.

In Experiment 2, altered the cognitive task to be more amenable to neurophysiological recording and analysis. In this new paradigm, there were no

152

significant differences between the effects of active and sham rTMS on choice accuracy or response time. There was, however, a small but significant effect of rTMS condition on the location of where the subject touched the screen to make her responses. In other words, active rTMS over posterior parietal cortex altered reaching behavior. We suspect that the subject’s cranial implant may have introduced too much additional distance between the TMS coil and her head for a strong enough electric field to reach the areas of the IPS (namely, LIP and VIP) thought to be relevant for numerical processing and for decision-making. However, the induced electric field was strong enough to alter activity in regions closer to the surface of the brain involved in arm reaches. This finding further demonstrates that rTMS can produce measurable changes in rhesus macaque behavior.

The results of Experiment 2 also help to establish guidelines for selecting brain regions and tasks to target with TMS in the macaque model. When stimulation is applied around a recording chamber and thus over an implant, with currently-available equipment, effective stimulation may be limited to very dorsal areas. However, there may be ways to target deeper areas; for instance, relatively lateral regions may be accessible if the TMS coil is held at the side of the head and the recording chamber is implanted at the top. Moreover, there is much to be learned from stimulating one area and recording in another; it is currently unknown how TMS to one hemisphere affects neuronal activity in corresponding regions of the other hemisphere, or how TMS to one region affects neuronal activity in interconnected regions that lie earlier or later in the

153

processing stream. For instance, it would be interesting to stimulate the IPS and record from PFC, or vice versa, while subjects perform the interleaved number and color comparison tasks used here, as these two regions have been implicated both in numerical cognition (Nieder, 2012a; Nieder & Dehaene, 2009) and in decision-making

((Andersen & Cui, 2009); also see Introduction).

Additionally, new methods of TMS delivery are currently being developed which may enable deeper targeting or show even greater cognitive effects. For example, the Hesed coil (H-coil) can activate regions of motor cortex up to 5.5 cm deep, as compared with 2 cm by a standard figure-eight coil (Zangen, Roth, Voller, & Hallett,

2005), although in general there is a tradeoff between depth and focality of stimulation regardless of coil type (Deng, Lisanby, & Peterchev, 2013). Pulse shapes other than the standard cosine biphasic pulse can induce larger changes in cortical excitability (Goetz et al., 2016). Theta burst stimulation, in which brief trains of very high-frequency pulses are delivered either continually in short succession or intermittently with longer breaks in between sets of trains, may also have stronger or longer-lasting effects than simple rTMS

(Wischnewski & Schutter, 2015). Adapting the stimulator and chamber-centric coil to reach deeper targets, produce novel pulse shapes, or deliver theta bursts could expand the range of behaviors that may be explored in the macaque TMS model.

In Experiment 3, an initial set of electrophysiological recordings were taken in macaque IPS before and after active or sham rTMS. Although equipment malfunction

154

introduced a roadblock and not enough data were ultimately available to reach general conclusions about the neuronal impact of rTMS, stimulation did seem to induce a change in activity each day. For both active and sham low-frequency rTMS, single-unit and multi-unit firing rates sometimes increased and sometimes decreased. Additionally, power in the lower frequency bands seemed to be altered during a short time window following treatment, but day-to-day variability obscured any statistically significant effects. The heterogeneity in these effects suggests that several factors which may modulate neural responses to TMS should be carefully controlled or recorded in future experiments. For instance, coil orientation can modulate the effect of TMS (Richter,

Neumann, Oung, Schweikard, & Trillenberg, 2013; Thomson et al., 2013) and likely interacts with neuronal orientation (Pell et al., 2011), so it will be important to systematically vary the orientation of the coil and also record from precise locations in the brain region of interest where neurons may be oriented slightly differently with respect to the scalp. Ample recordings should also be made at different depths, since the induced electric field is known to decrease with increasing depth (Figure 27). Finally, different types of neurons may respond differently to TMS (Radman, Ramos, Brumberg,

& Bikson, 2009). While it is difficult to ascertain cell type from single-unit recordings in an intact animal, waveform and firing properties may be used to identify, for example, pyramidal neurons versus interneurons in the primate cortex (Constantinidis &

Goldman-Rakic, 2002).

155

Together these studies inform the way forward for uncovering the neuronal basis of the effects of TMS. Determining the precise impact of different types of TMS on different brain regions, and how these effects relate to cognitive and behavioral changes, will be important in optimizing the application of this non-invasive treatment in neurology, psychiatry, and perhaps cognitive enhancement.

156

3. Competitive social decision-making

Deception is commonplace in human social interactions. In business, politics, sports, law, and even our friendships and family relationships, there are many reasons to lie or mislead. However, deception can be costly to society; organizations lose a substantial amount of money to fraud (Gee & Button, 2015), and falsehoods in the courtroom can thwart law enforcement. Excessive lying is characteristic of various psychiatric conditions, including borderline personality disorder, malingering, and factitious disorder (Dike, Baranoski, & Griffith, 2005). At the same time, individuals with other conditions such as Autism Spectrum Disorder often have difficulty telling lies

(Baron-Cohen, 1992; Sodian & Frith, 1992; Yirmiya, Solomonica-Levi, & Shulman, 1996).

Despite the sociological and clinical importance of untruthful behavior, the biological origins and precise brain mechanisms underlying deception remain poorly understood.

Deception may have evolved in social animals as a tactic to compete over food, mates, or other resources (Amici, Call, & Aureli, 2009; C. F. Bond Jr & Robinson, 1988).

Whether the kind of intentional deception practiced by humans is used by other animals has been controversial. Animals unquestionably employ involuntary deception, such as camouflage or false markings that can make harmless prey appear poisonous or dangerous predators appear safe (Caro, 2014; A. K. Miller, Maritz, McKay, Glaudas, &

Alexander, 2015; Stevens, 2007; Théry & Casas, 2002). Animals also deploy behaviors that function to fool conspecifics, as discussed below. Importantly, in all these cases, it is

157

unclear whether any such acts are planned attempts to dupe another individual, rather than pre-programmed or perhaps learned behaviors that the animal automatically employs in a given situation.

Early attempts to systematically study deception in other animals involved chimpanzees (Pan troglodytes). Notably, Woodruff & Premack (Woodruff & Premack,

1979) designed an experiment in which subjects interacted repeatedly with each of two differently-dressed human trainers, one who turned out to be cooperative, the other competitive. While the chimpanzee was watching, a neutral aide placed food under one of two buckets out of the subject’s reach. Then one of the human trainers, who had been instructed to search under one bucket in an attempt to find the food, entered the room, and the chimpanzee was free to communicate with her. If the cooperative trainer searched under the correct bucket, she would give the food to the chimpanzee, but if she searched under the incorrect bucket, the subject received no food. If the competitive trainer searched under the correct bucket, she kept the food for herself, but if she searched under the incorrect bucket, the chimpanzee was then allowed to take the food from the other bucket. The four tested chimpanzees quickly developed methods to indicate the correct bucket to the cooperative trainer, and did eventually withhold information from or actively mislead the competitive trainer. However, it took hundreds of interactions over at least a year for the subjects to become successful with the competitive trainer. This may be due to the fact that the behaviors needed to mislead the

158

competitive trainer – for instance, orienting away from the food – were less natural for the chimpanzees. Thus, while this experiment shows that chimpanzees can learn to provide or withhold signals differentially based on the recipient, it is possible that chimpanzees engaged in trial-and-error learning of how to obtain food in the presence of each experimenter, and then proceeded via stimulus-response associations. Similarly,

Menzel (Menzel, 1974) (see also (Hirata & Matsuzawa, 2001)) found that chimpanzees can develop tactics and countertactics that function to deceive conspecifics, but such behaviors could have been learned by trial and error.

To widen the scope of species and forms of deception under investigation,

Whiten and Byrne (Whiten & Byrne, 1988) solicited anecdotes from field researchers describing deceptive behavior in non-human primates. They defined tactical deception as “acts from the normal repertoire of an individual, used at low frequency and in contexts different from those in which it uses the high frequency (honest) version of the act, such that another familiar individual is likely to misinterpret what the acts signify, to the advantage of the actor” (Byrne & Whiten, 1985). Many of the submitted examples are compelling. For instance, a subadult chacma baboon (Papio ursinus), when pursued by several aggressive adults, suddenly stopped and stared into the distance as though he had caught sight of predators or adversaries (although none seemed to be present), successfully distracting the adults from their chase. A gorilla (Gorilla beringei beringei) noticed a desirable type of vine obscured in a tree, then sat at the side of the trail and

159

groomed herself while her groupmates walked by, jumping up to snatch the vine only once they had passed. Chimpanzee Y hurt his hand in a fight with groupmate N, and for a week afterward, while N could see Y, Y hobbled pitifully, but when Y was out of N’s field of vision, Y walked normally.

In these isolated observations, deception does seem to have occurred, but its intentionality cannot be determined. It is impossible to know whether the actions were performed for some other reason, perhaps accidentaly resulting in deceipt (e.g. did that particular gorilla often pause to groom herself anyway, and one day it happened to result in her exclusive access to a vine?), and whether the actor represented the beliefs of the observing animals or simply learned through trial and error to behave a certain way in a given situation.

Following Whiten & Byrne’s (1988) report, more controlled studies have begun to shed light on the occurrence and nature of deceptive behavior in multiple species. For example, male cuttlefish (Sepia plangon) quickly change their shape and color so as to display male courtship patterns on the side of their body facing a female while displaying a female pattern on the side of their body facing a rival male, in order to avoid the rival male interrupting their courtship (C. Brown, Garwood, & Williamson,

2012). False behaviors or hiding related to attracting or retaining mates have also been reported in geladas (Theropithecus gelada) (le Roux, Snyder-Mackler, Roberts, Beehner, &

Bergman, 2013), macaques (Macaca mulatta and Macaca fascicularis) (Overduin-de Vries,

160

Spruijt, De Vries, & Sterck, 2015) antelopes (Damaliscus lunatus) (Bro Jørgensen & Pangle,

2010), poison frogs (Dendrobates auratus) (Summers, 2014), and mantids (Pseudomantis albofimbriata) (Barry, 2014).

Other species seem to use tactical deception to monopolize food resources. For example, ravens alter the way they store food (Bugnyar & Kotrschal, 2002) and also the way they search for existing caches (Bugnyar & Heinrich, 2006; Bugnyar & Kotrschal,

2004) based on the presence and identity of other ravens, in order to conceal information regarding the location of food or their own knowledge of that location. Scrub-jays similarly alter their caching behavior based on whether others can see them, and take efforts, such as repeated re-caching, to conceal their food caches and invalidate the knowledge of observers (Dally, Clayton, & Emery, 2006a; Dally, Emery, & Clayton, 2005;

2006b; Emery, Dally, & Clayton, 2004). Such strategies are not limited to birds; a food- caching rodent, the eastern grey squirrel (Sciurus carolinensis), will cover empty sites while it is caching food, especially when a conspecific is nearby, in order to reduce cache pilferage by observers (Steele, Halkin, Smallwood, & McKenna, 2008). Relatedly, in various monkey species, including mangabeys (Cercocebus torquatus torquatus) (Coussi-

Korbel, 1994);, Tonkean macaques (Macaca tonkeana) (Ducoing & Thierry, 2003);, and spider monkeys (Ateles geoffroyi), capuchin monkeys (Cebus apella), and long-tailed macaques (Macaca fascicularis) (Amici et al., 2009), subordinate individuals can learn to

161

take indirect routes toward hidden food or delay their approach toward food when they alone have knowledge of the food’s location and a dominant individual is present.

These examples demonstrate that diverse species employ tactical deception to gain an advantage over conspecifics (although it remains unclear whether individuals represent the beliefs of others and act intentionally to manipulate them). This raises the question of how deception evolved, and in particular, what biological mechanisms support it.

An initial clue came from the records of deception in primates collected by

Whiten and Byrne (1990). Even when controlling for group size and the human observers’ effort, Byrne & Corp (Byrne & Corp, 2004) found that neocortex size – but not the size of the rest of the brain – scaled with deception rate in seventeen primate species, including prosimians, New World monkeys, Old World monkeys, and great apes. This pattern suggests that neural architecture in the neocortex of primates (and perhaps the associated neopallium in birds) is involved in deceptive behavior.

More precise localization of the parts of neocortex most relevant for deception has recently become possible using functional neuroimaging. In the first fMRI study of lying, subjects answered yes-or-no autobiographical questions presented either on a screen (visual) or from a speaker (auditory), and either lied or told the truth depending on the color of the yes/no response text (Spence, Farrow, Herford, & Wilkinson, 2001). In both the visual and auditory tasks, activity was greater in the bilateral ventrolateral

162

prefrontal cortex during lies, suggesting that this area plays a role either in producing lies or withholding the truth. However, note that in this non-competitive task, subjects did not make their own decisions about when to lie, and there was no advantage to be gained from lying.

Shortly thereafter, Ganis and colleagues (Ganis, 2003) used fMRI to examine whether different brain processes support different types of lies. In particular, they had subjects produce lies that were spontaneous and isolated, or lies that needed to fit in to a memorized scenario. Compared with telling the truth, both of types of lying activated the bilateral anterior middle frontal gyrus, the bilateral fusiform/parahippocampal gyrus, the right precuneus, and the left cerebellum. Other regions were more involved during one type of lie versus the other: spontaneous-isolated lies activated the anterior cingulate, the left precentral gyrus, the right precentral/postcentral gyrus, and the right cuneus, whereas memorized-scenario lies activated the right anterior middle frontal gyrus. Different forms of deception might therefore rely on different brain networks.

Building on this idea, as well as several intervening studies implicating the prefrontal cortex (Davatzikos et al., 2005; Kozel, Padgett, & George, 2004a; Kozel, Revell,

& Lorberbaum, 2004b; Langleben et al., 2005; 2002; T. M. C. Lee et al., 2002; Nunez,

Casey, Egner, Hare, & Hirsch, 2005; Phan et al., 2005) and/or the anterior cingulate cortex (Johnson, Barnhardt, & Zhu, 2004; Kozel, Padgett, & George, 2004a; Kozel, Revell,

& Lorberbaum, 2004b; Langleben et al., 2002; 2005; Nunez et al., 2005) in deception, Abe

163

and colleagues (Abe et al., 2006) used PET to determine whether two types of lies – pretending to know, i.e. deception based on experienced events; or pretending not to know, i.e. deception based on unexperienced events – were accompanied by different activity patterns in the prefrontal and anterior cingulate cortex. They found that the left dorsolateral prefrontal cortex, right ventrolateral prefrontal cortex, and right medial prefrontal cortex are involved in producing either type of lie, whereas the right anterior cingulate cortex is only involved in pretending not to know. These results highlight the involvement of the prefrontal cortex in deception in general, while also emphasizing that different types of lies can engage distinct brain networks.

Abe and colleagues (Abe, Suzuki, Mori, Itoh, & Fujii, 2007) went on to evaluate how social context modulates how different brain areas are involved in lying. During

PET imaging, subjects were instructed by an interrogator to lie or tell the truth about their past memories, but were sometimes also told secretly by a second experimenter to attempt to deceive the interrogator (i.e. tell the truth when the interrogator asked them to lie, or lie when the interrogator requested the truth). They found that the right anterior prefrontal cortex showed a main effect of lying, and also a main effect of deceiving the interrogator, while the dorsolateral prefrontal cortex, bilateral precuneus, and left cerebellum showed only a main effect of lying, and the left ventromedial prefrontal cortex and left amygdala showed only a main effect of deceiving the interrogator. Thus, when instructed to lie generally, or to defy another person either by

164

telling the truth or a lie, distinct parts of the prefrontal cortex (along with other brain regions) were activated.

Since then, numerous other investigations using diverse experimental paradigms have recapitulated and further clarified the prefrontal cortex’s role in deception (Abe et al., 2009; S. Bhatt, Mbwana, Adeyemo, Sawyer, & Hailu, 2009; Browndyke, Paskavitz,

Sweet, & Cohen, 2008; Christ, Van Essen, Watson, Brubaker, & McDermott, 2009; Q. Cui et al., 2014; Fullam, McKie, & Dolan, 2009; Greene & Paxton, 2009; Hakun et al., 2008;

Kaylor-Hughes et al., 2011; Kireev, Korotkov, & Medvedev, 2012; T. M. C. Lee, Lee,

Raine, & Chan, 2010; Marchewka et al., 2012; Nose, Murai, & Taira, 2009; Spence,

Kaylor-Hughes, Farrow, & Wilkinson, 2008), including intentionally faking responses about previously-viewed words (T. Lee, Au, Liu, Ting, & Huang, 2009) or numbers

(Vartanian et al., 2013), or about having performing certain everyday tasks; lying about the valence of affectively positive or negative images (T. M. C. Lee et al., 2010); feigned hearing loss (Liang et al., 2012; McPherson, McMahon, Wilson, & Copland, 2011); feigned memory deception (Z. X. Chen et al., 2015); strategic lying in a competitive computer-simulated dice game (Sip et al., 2010) or a game in which subjects attempted to fool a computer about which direction an arrow was pointing (Kireev, Korotkov,

Medvedeva, & Medvedev, 2013); in a situation where the a human recipient was aware of the subject’s attempt to deceive them, so the subjects needed to intersperse lies with truth in order to successfully fool the recipient on successive trials (X. P. Ding, Sai, Fu,

165

Liu, & Lee, 2014); and in a strategic repeated interaction game, during lying as well as during truth-telling with the intention to deceive (Volz, Vogeley, Tittgemeyer, Cramon,

& Sutter, 2015).

Many of these studies have specifically implicated the dorsolateral (Abe et al.,

2006; 2007; 2014; Baumgartner, Fischbacher, Feierabend, Lutz, & Fehr, 2009; S. Bhatt et al., 2009; Browndyke et al., 2008; Kireev et al., 2012; Liang et al., 2012; Nunez et al., 2005;

Phan et al., 2005; Sai, Zhou, Ding, Fu, & Sang, 2014; Spence et al., 2001; Yin, Reuter, &

Weber, 2016) and sometimes dorsomedial (Browndyke et al., 2008; Langleben et al.,

2005; Liang et al., 2012; Phan et al., 2005; Spence et al., 2001) prefrontal cortices in deception. For example, when subjects intentionally misidentified previously known faces in a line-up, regions in the frontal and parietal cortices were selectively activated, including the left precuneus, right middle frontal gyrus, right inferior frontal gyrus, right supremarginal gyrus, right superior frontal gyrus, and right dorsalateral prefrontal cortex (S. Bhatt et al., 2009). Malingered (feigned cognitive impairment) errors of omission activated the dorsolateral and dorsomedial prefrontal cortex, while malingered errors of inclusion activated the bilateral ventrolateral prefrontal cortex (Browndyke et al., 2008). Deciding to break a promise was associated with increased activation of the dorsolateral prefrontal cortex, anterior cingulate cortex, and amygdala (Baumgartner et al., 2009). Another study of spontaneous deception revealed that the dorsolateral prefrontal cortex and frontopolar cortex, as well as the anterior cingulate cortex and

166

supramarginal gyrus, were more active during deliberate deceptive and manipulative honest responses than control truthful responses (Kireev et al., 2012). The decision to tell a lie or the truth that would harm or help another person involved the dorsolateral prefrontal cortex, and when scenarios involved harm to the other person, the right temporoparietal junction and right medial frontal cortex distinguished dishonest from honest decisions (Abe et al., 2014). During a gambling game, the subgenual anterior cingulate cortex, dorsalateral prefrontal cortex, ventrolateral prefrontal cortex, and inferior parietal lobule distinguished between spontaneous lying and truth-telling, whereas a different network, including the right inferior frontal gyrus, left supplementary motor area, anterior cingulate cortex, inferior parietal lobule, and superior frontal gyrus, was engaged for instructed lying and truth-telling (Yin et al.,

2016). Finally, in a mock crime scenario, guilty participants showed increased activity in the dorsolateral prefrontal cortex and supplementary motor cortex when responding to probe items related to the crime—that is, concealing information about the crime—than when responding to irrelevant items (Sai et al., 2014).

While neuroimaging studies can correlate certain brain regions with deception, loss-of-function studies enable the causal implication of these brain regions. Case reports relating to the neurobiological basis of deception have been diverse, both in terms of the brain regions implicated and the behavioral phenotype (see (Abe, 2009) for a review).

Notably, one recent study found that patients with focal lesions of the DL-PFC, as

167

compared to patients with OFC lesions and healthy controls, made decisions that maximized their own economic gain even when that meant decreasing another person’s gain by sending them a false message (L. Zhu et al., 2014), indicating that DL-PFC is necessary for mediating between honesty and self-interest.

Transient manipulations of brain activity can also shed light on the causal involvement of particular regions in deception (Luber, Fisher, Appelbaum, Ploesser, &

Lisanby, 2009). For example, the suppression of activity in the anterior prefrontal cortex

(aPFC) via cathodal transcranial direct current stimulation (tDCS) increased the ease with which, following a thief role-play, subjects lied in an interrogation about facts that would implicate them in having stolen money, perhaps implicating aPFC in the socio- emotional components of the decision to deceive (Karim et al., 2010). On the other hand, the disruption of activity in the DL-PFC via anodal tDCS substantially slowed response times when subjects lied about experienced events, but not new events (Priori et al.,

2008), confirming the role of DL-PFC in producing at least certain types of lies. Applying tDCS to DL-PFC has also been shown to decrease response times during various types of lies (Fecteau, Boggio, Fregni, & Pascual-Leone, 2013).

Transcranial magnetic stimulation (TMS) was first used over motor cortex along with the measurement of motor-evoked potentials to demonstrate increased cortical excitability when subjects lie (Hadar, Makris, & Yarrow, 2012; Lo, Fook-Chong, & Tan,

2003), and more importantly, has been used to selectively manipulate activity in certain

168

prefrontal regions to modulate lying behavior. In particular, low-frequency repetitive

TMS (rTMS) to the right DL-PFC decreased the propensity to lie spontaneously about the color of a presented object, whereas rTMS to the left DL-PFC increased this propensity (Karton & Bachmann, 2011) (though see (Karton, Palu, Jõks, & Bachmann,

2014a)). Conversely, high-frequency rTMS to the right DL-PFC increased the propensity to lie spontaneously about the shape of an object (Karton et al., 2014a; Karton, Rinne, &

Bachmann, 2014b). Together these lesion and stimulation studies support the causal involvement of DL-PFC in deceptive behavior.

All these studies leave unanswered the question of how individual neurons are involved in producing deceptive behavior. Therefore we sought to uncover signals related to deception in neurons of the macaque DL-PFC and DM-PFC, regions that have been implicated in deception (as discussed above) as well as other strategic behaviors during social interactions in humans (M. A. Bhatt, Lohrenz, Camerer, & Montague, 2010;

Coricelli & Nagel, 2009; Steinbeis, Bernhardt, & Singer, 2012). Additionally, in macaques, neurons in the lateral PFC have been shown to distinguish between winning and losing in a competitive game, and also between whether the subject monkey was competing against the computer or another monkey (Hosokawa & Watanabe, 2012), while neurons in the DM-PFC have been implicated in strategy adjustment in accordance with the inferred strategy of a computer opponent (Seo et al., 2014).

169

We trained monkeys to play a zero-sum two-player computer game called

Penalty Shot, in which one player—the “shooter”—was motivated to score a goal with a ball, while the other player—the “goalie”—was motivated to block the ball. Following previous attempts to define deception in nonhuman animals (Byrne & Whiten, 1985;

McWhirter, 2015; Searcy & Nowicki, 2005), here we will consider deception to occur when a sender misuses a message or signal in a particular context to the sender’s benefit and the receiver’s detriment. In other words, if the shooter moves during competitive play in a way that conceals information so that he may win the game and the goalie may lose, and he would not move that way in a similar situation without a live opponent present, then the shooter’s behavior can be considered deceptive. We recorded single- unit activity in the shooter’s DL-PFC and DM-PFC to identify a neuronal signature of deception, and we pharmacologically inactivated each of these regions to determine its causal contribution to behavior in this social, strategic game.

3.1 Methods

3.1.1 Subjects

Three adult male rhesus macaques (monkeys O, Y, and E) played as shooters in the penalty shot game described below, and these three plus one other (monkey D) played as goalies. All subjects were singly housed, and were kept on a water-restricted diet to increase motivation to earn liquid reward in the experimental task.

170

3.1.2 Task (Penalty Shot)

Two monkeys sat in primate chairs (Crist Instruments, Damascus, MD) modified to allow access to joysticks (IP66s, CTI Electronics, Stratford, CT) affixed in a custom- built acrylic case at the front. All axes of the joystick were calibrated with a MATLAB

(Natick, MA, USA) script to ensure that every direction yielded equal speed. Horizontal and vertical eye positions of the shooter were sampled at 1000 Hz by an infrared camera

(Eyelink 1000, SR Research, Kanata, Ontario, Canada). The task was presented and data were collected via custom-built scripts in MATLAB using the Psychophysics and

Eyelink Toolbox extensions ((D. H. Brainard, 1997; Pelli, 1997) (see http://psychtoolbox.org/). Primate chairs were oriented at a ninety-degree angle to one another such that each monkey could see the other to his left or right. Each monkey also faced his own computer monitor displaying the task; the display was the same on both screens.

Each trial was preceded by a 300ms period during which the monkeys each had to center their joysticks. Then a green circle, the ball, appeared at the left and vertical center and a long red rectangle, the goalie bar, appeared at the vertical center near the right of the screen. Further to the right, a gray line, the goal, extended from the top to the bottom of the screen. The screen was otherwise black (Figure 35). The “shooter” monkey’s joystick controlled the movement of the ball, which was unconstrained in all

171

directions except that it could not go beyond the borders of the screen, and the “goalie” monkey’s joystick controlled the movement of the bar, which was unconstrained vertically other than not extending beyond the screen, but did not move horizontally.

The monkeys’ roles were fixed throughout an experimental session. Each trial ended when the ball reached the goal, when the ball hit the goalie bar, or when ten seconds had elapsed with no movement from the shooter. The horizontal and vertical location of the ball, as well as the vertical location of the goalie bar, were recorded throughout the trial at roughly 60 Hz. If the ball reached the goal, the shooter was considered to have won and he received a drop of juice in his mouth; while if the ball hit the goalie bar, the goalie was considered to have won and he received a drop of juice in his mouth. Juice volume on each successful trial was similar for both monkeys and constant throughout the session. Following juice delivery, the screen went blank for a 1.5 s inter-trial interval before the next trial began. The speeds of the ball and goalie bar were titrated by the experimenter between gaming blocks to achieve a roughly 50% win rate for each player in each session. Monkeys were first trained against a computer opponent performing simple movements, and subsequently confronted monkey opponents.

172

Gaming blocks of about 200 trials were performed in alternation with control movement blocks of about 110 trials. These control trials took one of several forms, which were randomly interspersed. In the first type (direction controls), the ball began at the center of the screen and the shooter had to navigate to a circular target at one of eight peripheral locations evenly spaced around a circle. In the second type (movement sequence to targets), the ball began at the left of the screen and the shooter had to navigate to two circular targets in sequence, where the first was at the horizontal center and near the vertical top (or bottom, respectively) of the screen, and the second was toward the right of the screen (where the goal-line would be during competitive trials) and near the vertical bottom (or top). In the third type (movement sequence around obstacles), the ball began at the left of the screen and the shooter had to navigate to the goal-line at the right

173

of the screen while avoiding one, two, or three stationary goalie bars positioned at random.

In addition to competing against a live monkey opponent, monkey O also completed five sessions as a shooter playing against replays of goalie movements from randomly-selected previous trials (goalie replay condition).

3.1.3 Neurophysiological Recordings

In an initial sterile surgical procedure, a head restraint prosthesis was implanted using standard techniques (see section A.2.3). After a six-week recovery period, animals were habituated to head restraint and trained to perform oculomotor tasks for juice rewards. In a second sterile surgical procedure, a Cilux recording chamber (Crist

Instruments, Hagerstown, MD) was implanted (see section A.2.5) over the prefrontal cortex using a stereotax for positioning. After all surgeries, animals received analgesics for three days and antibiotics for ten days. The chamber was kept clean with regular saline/betadine washes and sealed with replaceable sterile caps (Crist Instruments).

For microelectrode recording, a sterile, stainless steel guide-tube was lowered 1-

3mm past the dura mater. Single tungsten electrodes (250 µm diameter, impedance 1-10

MΩ; FHC, Bowdoin, ME) or an eight-contact serial U-Probe (15 µm electrode diameter,

150 µm inter-electrode spacing, impedance 275 ± 50 kΩ; Plexon Inc, Dallas, TX) were lowered using a hydraulic microdrive (David Kopf Instruments, Tujunga, CA) until the

174

target area, either DL-PFC or DM-PFC, was reached. After 10-20 minutes of rest, the electrode was slowly advanced until the waveforms of 1 to 3 neurons per contact were isolated. Individual action potentials were identified online by standard criteria, and isolated and recorded on a Plexon system.

Neurons were selected for recording on the basis of isolation quality only, and not on task-dependent properties. Recording sites were determined by grid coordinates, recording depth, gray/white matter electrophysiological signatures, and structural MRIs

(Figure 36 & Figure 37). Structural MRIs were acquired on a 3T Siemens Trio (Malvern,

PA) (T2 scans with TE=114, TR=4300, FOV=256, 2mm slice thickness) with recording grid and fiducial markers in place. Recording depth and trajectories were calculated using

Osirix software (Rosset, Spadola, & Ratib, 2004).

Figure 36: Coronal MRI showing locations of DL-PFC and DM-PFC recordings in Monkey O. Injections and additional recordings were completed in analogous areas of the right hemisphere.

175

Figure 37: Coronal MRI showing DL-PFC and DM-PFC injection and recording locations in Monkey Y.

3.1.4 Pharmacological Inactivation

The GABAA receptor agonist muscimol diluted in saline (5mg/ml) or saline was injected using a sterile microinjection canula (Crist Instruments). The target area was first identified with a recording electrode on a previous day, and the canula was then driven using the same hydraulic microdrive and grid system. The injections were extended over 20-30 minutes. Each of three volumes of muscimol solution or saline—3,

4, and 5 µL—was injected in the DL-PFC during one session and in the DM-PFC during another session for each of two monkeys (O and Y) playing as the shooter, resulting in

24 (2 solutions x 3 volumes x 2 sites x 2 subjects) total injections. Muscimol has previously been used to reversibly inactivate portions of the macaque frontal cortex and impair performance in a variety of tasks (Dias, Kiesau, & Segraves, 1995; Iba &

Sawaguchi, 2003; Nakamura, Sakai, & Hikosaka, 1999; Sawaguchi & Iba, 2001; Mototaka

176

Suzuki & Gottlieb, 2013; E. A. West, DesJardin, Gale, & Malkova, 2011). Behavioral deficits resulting from muscimol injections can begin to appear anywhere from one to thirty minutes following the injection, and generally persist for at least ninety minutes, with gradual recovery thereafter. The activity of neurons within a radius of 1-3mm of the injection site is typically suppressed (Arikan et al., 2002; Edeline, Hars, Hennevin, &

Cotillon, 2002; J. H. Martin & Ghez, 1999).

3.1.5 Data Analysis

Behavior During the Game

As a behavioral metric of whether shooters plan their end goals in advance, we examined saccadic eye movements before and during the trial. We used a velocity

(25°/s) and a duration (5ms) threshold to detect saccades in the eye-tracking data. We included all saccades where the end-point fell within 5° horizontally of the goal-line as

“saccades to the goal-line,” and considered the vertical locations of the end-points of these saccades. We slid a 200 ms window in 50 ms steps from 1000ms before the trial began until the end of the trial, and recorded the vertical end-point of the final saccade to the goal-line that occurred within each window (if there was one). We will call this resulting variable the “saccade end-point.” We fit a generalized linear mixed-effects model in each window predicting the final vertical location of the ball—either when it crossed the goal-line or collided with the goalie bar—on each trial from the fixed effect

177

saccade end-point and the random effects subject and session. To determine whether the shooter’s saccades might be better explained by the location of the goalie bar, in each window beginning at the start of the trial (because there was no goalie bar before trial onset), we fit a similar generalized linear mixed-effects model predicting mean goalie vertical position from the fixed effect saccade end-point and the random effects subject and session. Finally, to compare how well the shooter’s saccades predicted his final goal with an overt metric of his behavior in the task, in each window beginning at the start of the trial we fit a similar generalized linear mixed-effects model predicting the final vertical location of the ball from the fixed effect ball projection (the location where the ball would be geometrically projected to end on the goal-line given its current position and direction, reduced to the top or bottom of the screen in cases where the projection landed beyond the screen) and the random effects subject and session.

Behavior During Goalie Replays

One subject, monkey O, competed for five consecutive sessions against randomly-selected replays of past goalie behavior. To assess whether the shooter’s behavior changed in this control condition, we considered the five previous and five subsequent sessions where this subject played as the shooter. We first fit generalized linear mixed-effects models in 200 ms windows slid in 50ms steps (as above) predicting final ball position from the fixed effect ball projection and the random effect session

178

number, separately for the prior five sessions, the five replay sessions, and the subsequent five sessions. To assess whether the replay condition significantly affected this relationship, in each window we also fit a generalized linear mixed-effects model predicting final ball position from the fixed effects ball projection, session type (replay or live opponent), and their interaction, and the random effect session number. We also performed these analyses using saccades to the goal-line instead of ball projection. To determine which windows showed a significant interaction between session type (replay or live) and either ball projection or saccades, we limited analysis to the first 3 seconds of each trial and used a conservative threshold p-value of 0.05 / 60 = 0.000833 since there were 60 different windows.

Neuronal Firing Rates

Because we found that the relationship between the ball’s projection on the goal- line and its final endpoint distinguished the shooter’s behavior in replay trials and live- opponent trials, we defined a related moment-to-moment metric, the “deception index”

(DI), as the absolute value of the difference between the final vertical position of the ball at the end of the trial, and the projection of the ball on the goal-line given its position and direction at any point in the trial (reduced to the top-most or bottom-most point on the goal-line if the projection would land beyond the edge of the screen): in other words, the difference between where the shooter appears to be heading and where he is actually heading.

179

Because we found that changes in DI characterized the difference between play against a live opponent and the goalie replay condition, we predicted that variations in

DI would be reflected by variations in neuronal firing rates. We further predicted that neurons would track the movements of the goalie bar. To test these predictions, we used a sliding-window generalized linear model predicting the firing rate of each neuron from the five factors DI, goalie bar position, shooter’s horizontal gaze position, shooter’s vertical gaze position, and success in the trial (win or loss). We fit this model in 100 ms windows slid in 20 ms steps throughout the trial, from the time movement began (onset) to the time the shooter collided with the goalie bar or arrived at the goal line (offset). To assess statistical significance, we then ran a permutation test with 100 iterations, in which the trial to which each sequence of firing rates was assigned was shuffled randomly on each iteration. For each of the five factors, we determined the longest run of consecutive windows in which the p-value associated with that factor was less than

0.05 on each iteration. This provided a 100-value null distribution of how many consecutive windows would be expected to have p < 0.05 for that factor by chance. We used the cutoff for the top 5% of this distribution as the threshold for the number of consecutive windows with p < 0.05 for each factor the neuron would need to exhibit for it to be considered significantly modulated by that factor. We compared the length of the longest run of windows with p < 0.05 in the actual data for that neuron with these cutoffs to determine whether the neuron was significantly modulated by each factor,

180

and used the earliest window that was part of a run longer than the cutoff to determine the onset timing of the neuron’s modulation. We performed this analysis both when aligning to trial onset and to trial offset.

We performed the same analyses for firing rates during control trials where the shooter had to move from the left to the right of the screen in a curvilinear path (i.e. movement sequence to targets and movement sequence around obstacles controls), except that the factors goalie position and success (win/loss) were not included as they were not present in these trials. These controls were obtained for 132 of the 171 DL-PFC cells, and

45 of the 98 DM-PFC cells.

To extend our analysis to the time leading up to and directly following the trial, we assigned each game trial to a low, medium, or high DI bin based on the mean DI across the trial and the lowest, middle, and highest equal-size (based on number of trials) thirds of the distribution of this value. We ran a sliding-window ANOVA on normalized firing rates (the firing rate divided by the mean firing rate in the 50 preceding and 50 following trials) by the two factors DI bin and trial outcome (win/loss) in 300 ms windows moved in 150 ms steps throughout the trial. As above, we used a permutation test to determine runs of consecutive windows with p < 0.05 for each factor that were longer than expected from a null distribution, with 20 iterations. We also ran a sliding-window ANOVA and permutation test in the control conditions on normalized firing rate by direction in the movement direction controls; by up-down versus down-up

181

trajectory in the movement to targets control; and by up-down versus down-up trajectory in the movement around obstacles control. We aligned control and game trials to movement onset and offset. For game trials, we also aligned to the penultimate move and final move, as derived from a behavioral model described as follows:

We fit a mixture of hidden Markov models to assign a single "strategy" label to each trial. Both the clustering model and the HMM are non-parametric in the sense that both the number of strategies/clusters and the number of hidden states are discovered from the data by using a finite approximation for inference with Dirichlet processes. The observation model that links discrete states to observed position and velocities is assumed to be multivariate Gaussian. Parameter priors for the observation are standard

Normal inverse Wishart. Prior strength on the mean corresponded to one observation while prior strength on the covariance matrix corresponded to d+1 observations, where d is the dimension of the observations (d=6, including the horizontal and vertical positions and velocities of the shooter, and the vertical position and velocity of the goalie). For both the mixture components and the transition probabilities, the Dirichlet process parameter alpha was set to one.

Muscimol Inactivation

We limited our analysis to the first 60 minutes of post-injection gameplay (i.e. roughly 15-75 minutes after the injection), since data were reliably collected up to that

182

time each day. To determine whether muscimol inactivation of either region impacted the behavior of the shooter in this task, we examined several aspects of the shooter’s performance. First we examined win rate by fitting a binomial generalized linear mixed- effects model including all post-injection game trials with injection type

(muscimol/saline) as a fixed effect, and subject and session number as random effects.

Next, to determine whether the impact of muscimol differed by brain region, we fit a binomial generalized mixed-effects model including all post-injection game trials with injection type (muscimol/saline) and location (DL-PFC/DM-PFC), as well as their interaction, as fixed effects, and subject and session number as random effects. Finally, to determine whether these effects varied over time, we fit the same model including the three-way and two-way interactions with time in session as a fixed-effect parameter.

Similar to our analysis of the replay sessions, to examine the projection-endpoint correspondence, we first fit a generalized linear mixed-effects model predicting the shooter’s final position on the goal-line from the projection on the goal-line as a fixed effect and the subject and session number as mixed effects in 200ms windows slid in

50ms steps through the trial, separately for each of the four injection conditions (saline-

DLPFC, muscimol-DLPFC, saline-DMPFC, muscimol-DMPFC). To follow-up on what appeared to be significant differences between the injection conditions (described in the

Results), we next fit a generalized linear mixed-effects model predicting the shooter’s final position on the goal-line from the projection on the goal-line as well as the

183

interaction between the projection, the injection location (DLPFC or DMPFC), and the injection type (DLPFC or DMPFC). We again used a 200ms windows slid in 50ms steps through the first 1500ms of the trial, resulting in 30 windows; consequently we used a threshold p-value of 0.05 / 30 = 0.0017 to determine whether the interaction term was significant in each window. Additionally, we included interaction terms with time in session and followed up on a resulting significant interaction during late-trial windows by performing this analysis both for each of six nonoverlapping 10-minute windows in the session, using a Bonferroni-corrected threshold p-value of 0.05 / 6 = 0.0083, to assess when the effects occurred.

Finally, since we had discovered that pre- and early-trial saccades to the goal-line were predictive of the shooter’s final position in this task, we performed the same analyses as in the previous paragraph, with saccade end-point on the goal-line in place of projection, in order to assess the possibility that muscimol disrupted planning ability or eye movement ability. To maintain the same number of windows and thus the same threshold p-value as in the projection analysis, we slid a 200ms window in 50ms steps from 500ms before the start of the trial until 1000ms into the trial. Again, to determine whether there was a time-varying effect, we fit additional models that included interaction terms with time in session.

184

3.2 Results

3.2.1 Behavior During the Game

First, we note that in the Penalty Shot game, shooters quickly developed curvilinear trajectories to reach the goal-line. The shooters and the goalies each won roughly half of the trials during standard recording sessions, although shooters won slightly more often (out of 52,842 interactions, the shooter won 27,758 (52.53%) and the goalie won 25,084 (47.47%), binomial p < 0.0001).

To determine whether shooters planned their overall trajectories in advance, we examined whether a feature of their behavior that exerted no control on the ball’s current path—their eye movements—was predictive of to where they ended up moving the ball on the goal-line. We found that there was a positive relationship between the vertical end-point of saccades to the goal-line and the final vertical position of the ball throughout the trial (Figure 38). This relationship evolved throughout the trial; it was low but positive at movement onset (β = 0.129, SE = 0.010, p < 0.0001), lower and not significantly different from zero in the window beginning 500 ms into the trial (β = 0.108,

SE = 0.011, p = 0.41), higher in the window beginning 1000 ms into the trial (β = 0.529, SE

= 0.0096, p < 0.0001), and yet higher in the windows beginning 1500 ms (β = 0.822, SE =

0.0055, p < 0.0001) and 2000ms (β = 0.829, SE = 0.0044, p < 0.0001) into the trial. Moreover, in the window from 1000-800 ms prior to the start of movement, the relationship was also significantly positive (β = 0.66, SE = 0.0091, p < 0.0001). These relationships are

185

illustrated as heat-maps in Figure 39, with saccade and movement end-point each discretized into 32 equal-sized bins (according to physical space on the screen). Thus the shooter’s saccades reveal his planned movement endpoint before, at the start of, and during the trial, although there is a period early in the trial where this is not the case.

186

It might be expected that, if the shooter is performing saccadic eye movements to the goal-line, he would look at the goalie bar, a bright object in the area which should carry information relevant to his task performance. As can be seen in Figure 38, although the shooter’s saccades do correlate positively with the position of the goalie bar, this relationship is weaker than the relationship between the shooter’s saccades and his eventual goal. Examining the same time points as above (with the exception of the pre- trial time point, since the goalie bar was not present then), this relationship also evolved throughout the trial; it was lowest at movement onset (β = 0.0025, SE = 0.0006, p = 0.87), low but significantly positive in the window beginning 500 ms into the trial (β = 0.0835,

187

SE = 0.0042, p < 0.0001), higher in the window beginning 1000 ms into the trial (β = 0.103,

SE = 0.0047, p < 0.0001), higher in the window beginning 1500 ms into the trial (β = 0.128,

SE = 0.0045, p < 0.0001), and yet higher in the window beginning 2000 ms into the trial (β

= 0.184, SE = 0.0049, p < 0.0001) into the trial.

We also considered whether the projection of the ball on the goal-line was a better predictor of the ball’s final position. As shown in Figure 38, early in the trial, the projection of the ball on the goal-line was a stronger predictor of its final position than the shooter’s saccades, but starting about one second into the trial, saccades became a better predictor. Although it appeared greater than the relationship for saccades, the relationship between ball projection and final position was not significant at the beginning of the trial (β = 0.142, SE = 0.0046, p = 0.31), but it became significant and increased through the trial (window starting 500 ms into trial, β = 0.221, SE = 0.0044, p <

0.0001; window starting 1000ms into trial, β = 0.418, SE = 0.0038, p < 0.0001; window starting 1500ms into trial, β = 0.567, SE = 0.0030, p < 0.0001; window starting 2000ms into trial, β = 0.620, SE = 0.0034, p < 0.0001).

3.2.2 Behavior During Goalie Replays

To determine how the shooter’s behavior changed when the goalie played randomly rather than reactively, we compared the relationship between the ball’s projection on the goal and the final ball position during play versus a live opponent with

188

play against computer replays of previous goalie moves. We found that this relationship was stronger when playing against goalie replays than a live opponent, beginning at the start of the trial but persisting through the trial (Figure 40). For instance, this pattern held in the window beginning at movement onset (replays: β = 0.108, SE = 0.012, p <

0.0001; previous sessions: β = -0.0746, SE = 0.0377, p = 0.048; next sessions: β = 0.002, SE =

0.0175, p = 0.99; interaction between session type (replay or not) and ball projection: β =

0.119, SE = 0.0254, p < 0.0001), and in the window starting 1000 ms into the trial (replays:

β = 0.312, SE = 0.016, p < 0.0001; previous sessions: β = -0.0837, SE = 0.0195, p < 0.0001; next sessions: β = 0.236, SE = 0.0259, p < 0.0001; interaction between session type (replay or not) and ball projection: β = 0.168, SE = 0.0330, p < 0.0001).

189

In contrast, the tendency to perform saccades that predicted eventual goal-line was not affected by whether the shooter played against a live or replay opponent (Figure

41). (e.g. in the window beginning one second prior to movement onset, replays: β =

0.1038, SE = 0.0317, p = 0.0011; previous sessions: β = 0.183, SE = 0.057, p = 0.0016; next sessions: β = 0.0045, SE = 0.0539, p = 0.93; interaction between session type (replay or not) and saccades: β = 0.021, SE = 0.0689, p = 0.76; beginning at movement onset, replays: β =

0.101, SE = 0.0238, p < 0.0001; previous sessions: β = 0.214, SE = 0.0576, p = 0.0002; next sessions: β = 0.0858, SE = 0.0441, p = 0.052; interaction: β = -0.0242, SE = 0.0577, p = 0.67; beginning 1 s into the trial, replays: β = 0.174, SE = 0.0458, p = 0.0002; previous sessions: β

= -0.0650, SE = 0.0495, p = 0.19; next sessions: β = 0.183, SE = 0.0461, p = 0.0001; interaction: β = 0.0623, SE = 0.0947, p = 0.51). Around 1.5 s into the trial, there was a significant interaction between session type (replay or not) and saccades, but as can be seen in Figure 41, this largely reflected a difference only with the sessions following the replays, and not those that came previously.

190

3.2.3 Neuronal Firing Rates

Using a sliding-window GLM and a permutation test for significance as described in the Methods, aligned to trial onset, we found that 28/171 DL-PFC neurons

(16.4%) were significantly modulated by deception index (DI, the absolute value of the difference between the projection of the ball on the goal-line given its current position and direction, and its actual end-point on that trial) during game trials and not during control trials (which were obtained for 22 of these 28 cells). This modulation began, on 191

average, 0.909 s (SEM = 0.114) into the trial. We also found that 22/98 DM-PFC neurons

(22.4%) were significantly modulated by DI during game and not control trials (obtained for 16 of these 22 cells), with modulation beginning, on average, 1.084 s (SEM = 0.178) into the trial. The difference between timing of onset of significant modulation in the two regions was not significant (t48 = -0.863, p = 0.393).

When aligned to trial offset (i.e. when the ball crossed the goal-line or collided with the goalie bar), we found that 37/171 DL-PFC neurons (21.6%) were significantly modulated by DI during game trials and not during control trials (which were obtained for 32 of these 37 cells). This modulation began, on average, 1.466 s (SEM = 0.076) before the end of the trial. In the DM-PFC, 19/98 cells (19.0%) were significantly modulated by

DI during game trials and not during control trials (which were obtained for 13 of these

19 cells). This modulation began, on average, 1.228 s (SEM = 0.114) before the end of the trial (SEM = 0.114). There was a trend toward a difference in timing of the onset of significant modulation in the two regions (t54 = 1.776, p = 0.0813).

While each individual cell was considered significantly modulated by a given factor based on the number of consecutive windows in which modulation by that factor was significant at the p < 0.05 level (according to a permutation test), we also examined how many cells in the population as a whole were modulated (p < 0.05) by each factor at each point in time. Heat maps showing the p-value for each cell in each window, separated by region and sorted by the the earliest window with p < 0.05, as well as a

192

summary showing what proportion of cells in each region had p < 0.05 in each window through the trial, are given in Figure 41. These plots reveal that early in the trial (i.e. in the first 200 ms from trial onset, and 2000 to 1000 ms before trial offset), more cells in

DL-PFC were modulated by DI, whereas later in the trial (500ms and later from trial onset, and 600 to 300 ms from trial offset), more DM-PFC cells were modulated by DI.

193

To determine whether neurons in DL-PFC and DM-PFC tracked the goalie bar, in the same GLMs we examined whether firing rates were modulated by the vertical position of the goalie bar. When aligning to trial onset, we found that 62/171 DL-PFC neurons (36.3%) were significantly modulated by goalie position, beginning 0.884 s

(SEM = 0.0977) into the trial, and 30/98 DM-PFC neurons (30.6%) were significantly modulated by goalie position, beginning 0.968 s (SEM = 0.126) into the trial (note that comparison with control trials was not relevant here, as there was no moving goalie bar during the control trials). The difference in timing between the two regions was not significant (t90 = -0.503, p = 0.616). When aligning to trial offset, 52/171 DL-PFC neurons

(30.4%) were significantly modulated by goalie position, beginning 1.365 s (SEM =

0.0731) before the end of the trial, and 31/98 DM-PFC neurons (31.6%) were significantly 194

modulated by goalie position, beginning 1.075s (SEM = 0.109) before the end of the trial.

The difference in timing between the two regions was significant (t81 = 2.290, p = 0.025).

Again, in Figure 43we show how these effects evolve through the full populations of

DL-PFC and DM-PFC cells.

195

To assess whether activity in each region reflects anticipation of winning the trial (or anticipation of reward; note that these features cannot be distinguished here), we included whether the trial was ultimately a win or a loss for the shooter in the same GLMs. When aligning to trial onset, we found that 27/171 DL-PFC neurons (15.8%) were significantly modulated by trial outcome, starting on average 0.948 s (SEM = 0.141) into the trial, and 10/98 DM-PFC neurons (10.2%) were significantly modulated by trial outcome, starting 1.603s (SEM = 0.140) into the trial. The difference in timing between the two regions was significant (t35 = -2.639, p = 0.012). When aligning to trial offset,

56/171 DL-PFC neurons (32.7%) were significantly modulated by trial outcome, starting

1.300 s (SEM = 0.081) before the end of the trial, and 26/98 DM-PFC neurons (26.5%) were significnalty modulated by trial outcome, starting 1.143 s (SEM = 0.128) before the end of the trial. The difference in timing between the two regions was not significant (t80

196

= 1.067, p = 0.289). Again, in Figure 44 we show how these effects evolve through the full populations of DL-PFC and DM-PFC cells. In Figure 45 we show how many cells in each region were significantly modulated by each of these three factors (DI, goalie position, and trial outcome) for each alignment, and the overlap among them.

197

198

Figure 45: Venn diagrams showing overlap in number of cells in DL-PFC (top) and DM-PFC (bottom), aligned to trial onset (left) and offset (right), significantly modulated by deception index, trial outcome (win/loss), and goalie position.

Finally, we included the shooter’s horizontal and vertical gaze position in the same GLMs. Many cells in each region were significantly modulated by horizontal and/or vertical gaze position (horizontal gaze position, onset aligned: 49/171 DL-PFC neurons (28.7%) and 27/98 Dm-PFC neurons (27.6%); offset aligned: 49/171 DL-PFC neurons (28.7%) and 27/98 DM-PFC neurons (27.6%); vertical gaze position, onset

199

aligned: 53/171 Dl-PFC neurons (31.0%) and 25/98 DM-PFC neurons; offset aligned:

51/171 Dl-PFC neurons (29.8%) and 26/98 DM-PFC neurons (26.5%)), but we do not show figures for these effects here, as the main purpose of including these variables in the GLMs was to ensure that the effects we observed for DI, goalie position, and trial outcome remained even when the variation in firing rate due to eye position was taken into account.

To examine whether deceptive strategies were planned early in, or even prior to, the trial, as well as whether neurons in each area represented deceptive strategy use and/or trial outcome after the trial, we ran a sliding-window ANOVA on firing rate by

DI bin (low, medium, or high for each entire trial) and by trial outcome (win/loss). As opposed to the momentary DI used in the GLM’s above, these DI bins are constant over a trial, giving a categorization of how deceptive a strategy the shooter used on each trial.

We found that a substantial number of neurons in each region were significantly modulated by each factor for each alignment (trial onset, penultimate move, final move, and trial offset), but in general, a higher proportion of neurons in DM-PFC were modulated by each factor. DM-PFC modulation occurred earlier in the trial than DL-PFC modulation for DI bin, but later in the trial for trial outcome (win/loss). We give details about these results in the next paragraphs, and show sample single-unit peri-stimulus time histograms for DL-PFC in Figure 46 and DM-PFC in Figure 47.

200

For DI bin, when aligning to trial onset, 51/171 (29.8%) of DL-PFC neurons showed significant modulation, with a mean onset time of 4.356 s (SEM = 0.491 s). Six of these were significant in the 2 s prior to movement onset; another 5 were significant within the first 1s of the trial, and another 10 within the first 2s of the trial. In the DM-

PFC, 41/98 (41.8%) neurons were significantly modulated, with a mean onset time of

2.098 s (SEM = 0.506 s). Twelve of these were significant in the 2 s prior to movement onset, another three within the first 1 s, and another eleven within the first 2 s. The timing was significantly different between the two regions (t90 = 3.177, p = 0.0020), as was the proportion of significantly modulated cells (χ21 = 3.994, p = 0.046).

When aligned to the penultimate move, 23/171 (13.5%) of DL-PFC neurons were significantly modulated by DI bin, with a mean onset time of -1.076 s (SEM = 0.232 s); 18 of these cells were modulated prior to the move. In the DM-PFC, 20/98 (20.4%) cells were significantly modulated, with a mean onset time of -1.418 s (SEM = 0.306 s); 16 of these were modulated prior to the move. The timing was not significantly different between the two regions (t54 = 0.902, p = 0.37), nor was the difference between the proportions of modulated cells in each region (χ21 = 2.246, p = 0.13). When aligned to the final move,

30/171 (17.5%) of DL-PFC neurons were significantly modulated by DI bin, with a mean onset time of -0.960 s (SEM = 0.284 s); 22 of these cells were modulated prior to the move.

In the DM-PFC, 26/98 (26.5%) cells were significantly modulated, with a mean onset time of -1.154 s (SEM = 0.342 s); 19 of these were modulated prior to the move. The

201

timing was not significantly different between the two regions (t54 = 0.440, p = 0.66), but the difference between the proportions of modulated cells in each region was marginally significant (χ21 = 3.052, p = 0.081). Finally, when aligned to movement offset, 39/171

(22.8%) of DL-PFC cells and 32/98 (32.7%) of DM-PFC cells were significantly modulated by DI bin, and there was a marginally significant difference between these proportions

(χ21 = 3.109, p = 0.078).

For trial outcome, when aligning to trial onset, 53/171 (31.0%) of DL-PFC neurons were significantly modulated, with a mean onset time of 2.732 s (SEM = 0.327 s); six of these were modulated prior to trial start, another three in the first 1 s, and another nine in the first 2 s. In the DM-PFC, 35/98 (35.7%) of neurons were significantly modulated, with a mean onset time of 2.821 s (SEM = 0.369 s); three of these were modulated prior to trial start, another two in the first 1 s, and another 4 in first 2 s. The difference in timing between the two regions was not significant (t86 = -0.178, p = 0.86), nor was the difference in proportions of significantly modulated cells between the two regions (χ21 = 0.631, p =

0.427).

When aligned to the penultimate move, 50/171 (29.3%) of DL-PFC neurons were significantly modulated by trial outcome, with a mean onset time of 0.114 s (SEM =

0.204s); 22 of these were modulated before the move. In the DM-PFC, 33/98 (33.7%) of neurons were significantly modulated, with a mean onset time of 0.505s (SEM = 0.179s);

8 of these were modulated before the move. The difference in timing between the two

202

regions was not significant (t81 = -1.344, p = 0.183), nor was the difference in proportion of significantly modulated cells between the two regions (χ21 = 0.574, p = 0.449). When aligned to final move, 42/171 (24.6%) of DL-PFC neurons were significantly modulated by trial outcome, with a mean onset time of -0.0893s (SEM = 0.1833 s); 21 of these were modulated before the move. In the DM-PFC, 35/98 (35.7%) of neurons were significantly modulated, with a mean onset time of 0.360s (SEM = 0.1908s); 9 of these were modulated before the move. The difference in timing between the two regions was marginally significant (t75 = -1.690, p = 0.0951), as was the difference in proportion of significantly modulated cells between the two regions (χ21 = 3.793, p = 0.0515). Finally, when aligned to trial offset, 43/171 (25.1%) of DL-PFC neurons and 37/98 (37.8%) of DM-PFC neurons were significantly modulated by trial outcome, and the difference between these proportions was significant (χ21 = 4.740, p = 0.0295).

203

Figure 46: Firing rate modulation by trial-wide Deception Index bin (top) and trial outcome (bottom) for a sample DL-PFC neuron. Shaded areas represent standard error of the mean.

Figure 47: Firing rate modulation by trial-wide Deception Index bin (top) and trial outcome (bottom) for a sample DM-PFC neuron. Shaded areas represent standard error of the mean.

204

3.2.4 Muscimol Inactivation

To determine whether suppressing activity in DL-PFC or DM-PFC altered the shooter’s performance in Penalty Shot, we first examined how frequently the shooter won the game following each injection type. When both regions were considered together, we found that, as compared with saline, muscimol significantly decreased win rate (β = -0.353, SE = 0.107, p = 0.00099) (Figure 48 & Figure 49). However, when location was included in the model, the interaction between injection type and location was not significant (β = -0.213, SE = 0.198, p = 0.28), indicating that the decrease in proportion of trials won following muscimol, as compared with saline, did not differ between DL-PFC and DM-PFC. Additionally, when time in session was included in the model, the two- way interaction between time and injection type was not significant (β = -0.000038, SE =

0.000060, p = 0.52), nor was the three-way interaction between time, injection type, and injection location (β = -0.00019, SE = 0.000082, p = 0.15) (Figure 50).

205

206

Figure 50: Win proportion over the course of the post-injection session. Shaded areas represent standard error of the beta estimates.

Next we examined the relationship between the ball’s projection on the goal and its final location on the goal-line. We found that, early in the trial, this relationship was strongest following muscimol to the DL-PFC, and later in the trial, this relationship was weakest following saline to the DM-PFC (Figure 51). For example, in the window beginning at movement onset, projection most strongly predicted final location after muscimol to the DL-PFC (β = 0.223, SE = 0.0218, p < 0.0001), followed by saline to the

DM-PFC (β = 0.194, SE = 0.0272, p < 0.0001) and muscimol to the DM-PFC (β = 0.189, SE =

0.0225, p < 0.0001), followed by saline to the DL-PFC (β = 0.119, SE = 0.0228, p < 0.0001).

Although the interaction between injection type and location was not significant at our threshold p-value of 0.0017 in this window (β = 0.104, SE = 0.0476, p = 0.0292), we note that the interaction term had a p-value below 0.05 in every 200 ms window slid in 50 ms

207

steps from trial onset through 600 ms into the trial, indicating that there is an early-trial interaction that would be significant if we used only one window around this time period and did not use an overly conservative Bonferroni correction. In the window beginning 1050 ms after movement onset, projection most strongly predicted final location after muscimol to DM-PFC (β = 0.689, SE = 0.0122, p < 0.0001) and muscimol to the DL-PFC (β = 0.685, SE = 0.0121, p < 0.0001), slightly less strongly after saline to the

DL-PFC (β = 0.678, SE = 0.0125, p < 0.0001), and even less strongly after saline to the DM-

PFC (β = 0.530, SE = 0.0150, p < 0.0001), and the interaction between injection type and location was significant (β = -0.152, SE = 0.0265, p < 0.0001). The interaction term was significant at p < 0.0017 for all windows between 850-1300 ms into the trial. These interaction terms indicate that in roughly the first 600 ms of the trial, muscimol (as compared with saline) to the DL-PFC strengthened the relationship between the ball projection and the final goal, whereas later in the trial, from 850 to 1300 ms, muscimol

(as compared with saline) to the DM-PFC strengthened this relationship.

208

The time course for these effects is shown in Figure 52. We examined the effects of injection type and the interaction between injection type and location in each of four non-overlapping fifteen-minute periods through the session, using a Bonferroni- corrected threshold p-value of 0.05/4 = 0.0125. During the early-trial window (250-450 ms into the trial), there was an effect of injection type in both the third and fourth periods (45-60 minutes after injection: β = 0.263, SE = 0.0622, p < 0.0001; and 60-75 minutes after injection: β = 0.190, SE = 0.0740, p = 0.010), and there was an interaction between injection type and location only in the third period (45-60 minutes after injection: β = -0.3225, SE = 0.0955, p = 0.0007). During the late-trial window (1050-1250 ms into the trial), there was a significant effect of injection type only in the fourth period

(60-75 minutes after injection: β = -0.118, SE = 0.0403, p = 0.0035), although during the 209

third period (45-60 minutes after injection) it was marginally significant (β = 0.0927, SE =

0.0400, p = 0.0205), and there was a significant interaction between injection type and location in the second, third, and fourth periods (30-45 minutes: β = 0.129, SE = 0.0503, p

= 0.010; 45-60 minutes: β = 0.1990, SE = 0.0574, p = 0.0005; 60-75 minutes: β = 0.255, SE =

0.0565, p < 0.0001).

210

In the initial analysis, we did not find any significant differences between the impact of muscimol and saline on the relationship between saccades and the final goal position (Figure 53); that is, the interaction between injection type and location was not significant in any tested window (for example, in the window beginning 1 s before movement onset: β = -0.0250, SE = 0.0948, p = 0.79; in the window beginning at movement onset: β = 0.106, SE = 0.0858, p = 0.22). Although we originally used 30 windows from 1000 ms before the trial through 500 ms into the trial to match the threshold p-value with the ball projection data, from examining Figure 53 it appeared that a rise in beta values may occur earlier following muscimol versus saline around

750-900 ms into the trial, so we extended our analysis through 1000 ms into the trial and adjusted our threshold p-value to 0.05 / 40 = 0.0013. At this conservative threshold, we did observe a significant effect of injection type on the relationship between saccades and final goal in the window 800-1000 ms into the trial (β = -0.726, SE = 0.193, p = 0.0002) and the next window 850-1050 ms into the trial (β = -0.645, SE = 0.173, p = 0.0002), and no significant interaction with injection location in either window (p = 0.076 and 0.13, respectively), suggesting that muscimol in either region led the subjects to reveal their intentions with their gaze earlier in the trial.

211

The time course for these effects is shown in Figure 54. Again we examined the effects of injection type and the interaction between injection type and location in each of four non-overlapping fifteen-minute periods through the session, using a Bonferroni- corrected threshold p-value of 0.05/4 = 0.0125. For the pre-trial window (1000-800 ms before trial onset), neither the effect of injection type nor the interaction between injection type and location with saccades was significant in any period (all p’s > 0.2). The same was true for the trial-start window (0-200 ms after trial onset; all p’s > 0.1).

However, for the mid-trial window (800-1000 ms after trial onset)—where we observed a significant effect of muscimol on the relationship between saccades and the final ball position—there was a significant effect of injection type during the first, second, and

212

third periods (15-30 minutes after injection: β = 0.441, SE = 0.158, p = 0.0054; 30-45 minutes after injection: β = 0.705, SE = 0.150, p < 0.0001; 45-60 minutes after injection: β =

0.558, SE = 0.172, p = 0.0012), and there was a significant interaction between injection type and location during the second period only (30-45 minutes minutes after injection:

β = -0.877, SE = 0.222, p = 0.0001).

213

3.3 Discussion

Everyday social life sometimes demands that we mislead another individual.

Particularly in competitive scenarios, it can be to our advantage to behave in a way that conceals our true intentions. Here we showed that rhesus macaques also behave deceptively in a competitive, zero-sum “penalty shot” game, and the DL-PFC and DM-

PFC are part of the brain network that mediates such behavior.

First, we showed that shooters plan their ultimate goal in advance. Regardless of the path taken to arrive at the goal-line, shooters made saccadic eye movements to points on the goal-line that were near the ball’s final location on a trial, even prior to and at the start of the trial. Despite this advance planning, early in the trial there was only a low correspondence between where the shooter appeared to be directing the ball—the projection of the ball on the goal-line—and where the ball landed at the end of the trial.

Notably, when a shooter played against nonreactive computer replays of randomly- selected previous goalie movements, the correspondence between the ball’s projection on the goal-line and its final location increased throughout the trial. This finding indicates that shooters attempted to mislead live goalie opponents by moving in a direction that was at odds with their planned end-point.

Guided by this finding, we defined a moment-to-moment “deception index” (DI) as a measure of the difference between where the shooter appears to be directing the ball at a given point in the trial, and where the ball lands at the end of the trial. We found

214

that neurons in both the DL-PFC and DM-PFC are modulated by DI, with this modulation occurring earlier in DL-PFC than DM-PFC. We also found that neurons in the DL-PFC were modulated by trial outcome earlier in the trial than neurons in DM-

PFC, perhaps indicating an association between use of deceptive tactics and anticipation of winning. Additionally, we found that neurons in both regions were modulated by the position of the goalie bar, and again this modulation was greater earlier in the trial in

DL-PFC than DM-PFC, and greater later in the trial in DM-PFC than DL-PFC. Thus it seems that both of these prefrontal regions participate in generating deceptive behavior and in tracking the opponent’s moves, but DL-PFC is involved earlier in the trial than

DM-PFC. On the other hand, DM-PFC showed stronger and earlier modulation by trial- wide DI bin, and later modulation by trial outcome, suggesting that DM-PFC may be more involved in selecting which overall strategy to use on a given trial and in encoding the reward outcome associated with each strategy, while DL-PFC is more involved in producing moment-to-moment deceptive behavior, particularly early in the trial.

The idea that DL-PFC and DM-PFC participate in generating deceptive behavior early and late in the trial, respectively, was further supported by pharmacological inactivation of each region. As compared with saline injections, muscimol injections into either the DL-PFC or DM-PFC reduced the shooter’s win rate. Although the degree of impairment as measured by trial outcome was similar between the two regions, we found that muscimol, as compared with saline, increased the correspondence between

215

the ball’s projection and its final position—that is, it decreased the shooter’s degree of deception—early in the trial following DL-PFC injections, and later in the trial following

DM-PFC injections.

The correspondence between ball projection and end goal can be thought of as an inverse of behavioral variability. That some brain regions are involved in producing beneficial variability in motor output has been observed in songbirds; specifically, the anterior forebrain pathway, including the lateral magnocellular nucleus of the anterior nidopallium (LMAN) and the basal ganglia, is necessary for modulating song variability, which can contribute to motor learning and adjustment of behavior to changing conditions (Kao, Doupe, & Brainard, 2005; Stepanek & Doupe, 2010; Woolley,

Rajan, Joshua, & Doupe, 2014). Based on our results, we posit that the primate DL-PFC and DM-PFC may be part of a similar network, producing variable motor output when such randomness is beneficial to the present task.

Our design did not enable us to isolate neuronal encoding of the goalie’s intentions from encoding of the shooter’s intentions and of the goalie’s current position.

Exploring where and how another individual’s intentions are represented in the brain will be an important avenue for future research. One recent study found that neurons in the macaque dorsal anterior cingulate cortex (dACC) are involved in representing another individual’s binary decisions in a cooperative scenario (the “Prisoner’s

Dilemma”) (Haroush & Williams, 2015). It remains to be tested whether dACC and

216

other brain regions, such as the DL-PFC and DM-PFC, are similarly involved in representing a dynamic competitor’s intentions.

The results presented here corroborate findings in humans suggesting a role for

DL-PFC (Abe et al., 2006; 2007; 2014; Baumgartner, Fischbacher, Feierabend, Lutz, &

Fehr, 2009; S. Bhatt et al., 2009; Browndyke et al., 2008; Kireev et al., 2012; Liang et al.,

2012; Nunez et al., 2005; Phan et al., 2005; Sai, Zhou, Ding, Fu, & Sang, 2014; Spence et al., 2001; Yin, Reuter, & Weber, 2016) and DM-PFC (Browndyke et al., 2008; Langleben et al., 2005; Liang et al., 2012; Phan et al., 2005; Spence et al., 2001) in deception, and suggest a mechanistic account for each area’s involvement: both regions participate in generating unpredictable behavior that runs counter to the subject’s true goal, but DL-

PFC participates in this process earlier than DM-PFC, suggesting a hand-off of control from DL-PFC to DM-PFC. Additional studies using the paradigm developed here could investigate whether and how neurons in other regions found to be activated during lying in humans, such as the ACC, ventrolateral and ventromedial prefrontal cortex, and striatum (Abe 2011), contribute to deceptive behavior.

Finally, here we examined the interactions of monkeys in a relatively limited number of sessions after they had already learned the experimental task. We suspect that, over repeated interactions, subjects would exhibit an “arms race” with the shooter and the goalie developing progressively more complex tactics and means to conceal their intentions. Potential behavioral markers of deception, such as facial expressions

217

(Brinke, Porter, & Baker, 2012) or saccadic eye movements as found here, would be expected to decrease over time. By carefully controlling which subject plays which role, and which subjects are paired as opponents, over time, we could track the emergence of deceptive strategies and perhaps gain insight into the brain mechanisms involved in generating novel deceptive behaviors.

218

4. General conclusions

In this dissertation, I have presented new evidence that bears on how primates make decisions, as well as how we can modulate the decision-making process. The first study explored whether monkeys can make decisions about the magnitude of discrete ratios, and the second study determined that monkeys’ ordinal decisions are influenced by spatial-numerical associations. The third set of experiments indicated that monkeys’ numerical decision-making may be altered with transcranial magnetic stimulation

(TMS), and laid the groundwork for investigations of how TMS alters neuronal activity to impact decision-making. In the final set of experiments, we investigated how the brain generates strategies that maximize reward in dynamic, competitive, social interactions. Below we review these findings, relate them to the neurobiology of perceptual, abstract, value-based, and social decision-making, and discuss avenues for future research.

4.1 Perceptual and Abstract Decisions

In the study presented in section 2.1 (Drucker, Rossa, & Brannon, 2015b), we trained two rhesus macaques to choose arrays that contained the greater ratio of positive to negative stimuli, regardless of the absolute number of stimuli in each of the two choice arrays. Subjects learned this task, and their performance generalized to novel ratios. Moreover, performance was modulated by the ratio between ratios; subjects

219

responded more quickly and accurately when the ratio between ratios was higher.

Control conditions ruled out the possibility that subjects were relying on surface area, although the ratio between ratios of surface area did seem to influence their choices. Our results demonstrate that rhesus monkeys can compare discrete ratios, demonstrating not only proportional reasoning ability but also the ability to reason about relations between relations.

The decisions monkeys made in the ratio comparison task were both perceptual and abstract. They had to encode the visual arrays of shapes, but then abstract away from that the numerosity of each type of shape, and then abstract one level further to the relative magnitude of each type of shape within each array, and use this abstract information to make a judgment about which ratio was greater. Even if monkeys extracted the total surface areas—a purely visual feature—of each type of shape rather than the numerosities, they still had to abstract away from that the relative magnitudes of the surface areas and use this information to make their decisions. That monkeys succeeded at this task underscores that other primates can reason abstractly about relational information.

Recent studies have begun to uncover the brain areas involved in representing ratios. Several have shown that the network for proportion is largely overlapping with the frontoparietal network for number (Jacob, Vallentin, & Nieder, 2012). For example, one fMRI study in humans found that activity in the intraparietal sulcus (IPS) was

220

modulated by the numerical distance between fractions as a whole, rather than the numerical distance between their numerators or denominators alone (Ischebeck,

Schocke, & Delazer, 2009). Other studies using fMRI adaptation found that the degree of

BOLD signal recovery in the parietal and frontal cortices, and especially in the anterior

IPS, correlated with the numerical distance between a previously-adapted proportion and a newly-presented deviant proportion, whether these proportions were presented as nonsymbolic line lengths, mixtures of different-colored shapes, numerical fractions, or verbal fractions (Jacob & Nieder, 2009a; 2009b). Additionally, neurons in the macaque prefrontal cortex (PFC) and inferior parietal lobule were found to be tuned to particular proportions of line lengths in a delayed-match-to-sample task (Vallentin & Nieder,

2010). Together with our finding that discrete ratio comparison follows Weber’s law, these results raise the possibility that a common network in the brain is engaged when representing any type of magnitude. (Note, however, that additional studies have also implicated the macaque inferotemporal cortex (Vighneshvel & Arun, 2015) and the human parahippocampal place area (Cant & Xu, 2014) in representing relative, as opposed to absolute, values, suggesting that other brain areas may be recruited when processing proportions.)

At the same time, the delayed-match-to-sample paradigm employed by

(Vallentin & Nieder, 2008) using a limited number of line-length proportions raises the question of whether monkeys in that study were encoding an analog magnitude or a

221

category for each stimulus. As discussed in the introduction, neurons in both the PFC and IPS (specifically area LIP) have been implicated in abstract categorization

(Freedman & Assad, 2006; 2011; Freedman & Riesenhuber, 2002). Our finding that monkeys generalize their ratio comparison to novel ratios and novel ratios-between- ratios suggests that they are not simply memorizing ratio “categories.” However, whether the neuronal mechanisms purported to support proportional reasoning using a delayed-match-to-sample task which could be solved by categorization (Vallentin &

Nieder, 2008) also underlie encoding of novel ratios remains to be tested.

In terms of a drift diffusion model, the findings so far speak to the representation of the temporary sensory evidence. That is, they show sensitivity of different brain regions to the ratio that is currently presented. Because these brain regions, and the behavioral signatures, are similar to those of analog numerosity, it is reasonable to speculate that similar evidence accumulation and decision commitment processes would occur for ratios as well. Indeed, the suggestion that there is a general brain circuit for encoding magnitudes (see also (Bueti & Walsh, 2009)) raises the possibility that the mechanisms uncovered for decision-making in the vibrotactile frequency task (e.g.

(Hernandez et al., 2002); see Introduction)—which required comparing the magnitude of two frequencies—may apply more generally to all magnitude-based decisions. Future experiments isolating various stages in the decision process during number and ratio

222

comparison, as well as during comparisons of other magnitudes such as space and time, could test this idea.

The study presented in section 2.2 (Drucker & Brannon, 2014) provided additional support for this possibility. In that study, we showed that monkeys, like humans, associate number with space. We trained four adult male rhesus macaques to select the fourth position from the bottom of a five-element vertical array. Monkeys maintained a preference to choose the fourth position through changes in the appearance, location, and spacing of the vertical array. We next asked whether monkeys show a spatially-oriented number mapping by testing their responses to the same five- element stimulus array rotated ninety degrees into a horizontal line. In these horizontal probe trials, monkeys preferentially selected the fourth position from the left, but not the fourth position from the right. Our results indicate that rhesus macaques map number onto space, suggesting that the association between number and space in human cognition is not purely a result of cultural experience and instead has deep evolutionary roots.

Indeed, as discussed in Drucker & Brannon (2015), in addition to monkeys and humans, chimpanzees (Adachi, 2014) and birds (Rugani, Kelly, & Szelest, 2010; Rugani,

Regolin, & Vallortigara, 2007; Rugani, Vallortigara, Vallini, & Regolin, 2011) associate number with space, and in newborn domestic chicks—as in humans—this mapping is context-dependent, with a number being treated as “small” or “large” depending on its

223

relation to a reference value (Rugani, Vallortigara, Priftis, & Regolin, 2015) (see discussion in (Drucker & Brannon, 2015)). In these studies, the animals—especially the

3-day-old chicks tested by Rugani and colleagues (2015)—had little experience with numerical stimuli, had limited or no exposure to cultural artifacts like keyboards portraying numbers in a line, and definitely did not read written text, largely ruling out the possibility that number-space mappings are learned, as suggested by some authors

(Shaki & Fischer, 2008; Shaki, Fischer, & Petrusic, 2009; Zebian, 2005). Rather, the tendency to map number to space seems to be an intrinsic part of representing number or more generally magnitude. Why might animals map number or magnitudes onto space?

One explanation involves hemispheric asymmetries in brain function. Right hemisphere dominance for numerical processing could produce increased right hemisphere activation—and thus increased leftward attention—when considering numerical information, which could cause animals to begin enumerating from the left

(Vallortigara, 2012). Another possibility is that numerical representations are spatially organized in the brain. Recent functional neuroimaging evidence suggests that there is a topographical arrangement of numerical magnitudes in human parietal cortex (Harvey,

Klein, Petridou, & Dumoulin, 2013). However, this cortical map has only been found in humans, and only for the numbers 1-7.

224

Another idea, as introduced above, is that numerical and spatial cognition rely on common neural circuits (Hubbard, Piazza, Pinel, & Dehaene, 2005). The same regions of parietal cortex—in particular, the lateral and ventral intraparietal areas—play a role in processing both number and spatial attention in primates. Activation of these regions by numerical information could therefore lead to shifts in spatial attention. This hypothesis is attractive from an evolutionary perspective, since navigating space is a fundamental problem faced by all animals. Rather than developing new cognitive systems to deal with abstract concepts of “less” and “more”, it would be more efficient to re-use the system already in place for processing space (Holmes & Lourenco, 2011). Considering numerical processing as an exaptation (Gould & Vrba, 1982) of spatial processing could explain the functional link between these two systems.

These findings have several implications for the study of decision-making. First, they emphasize that the influences on our decisions may be diverse and unexpected.

Without knowledge of the SNARC effect, few people would predict that their response speed when deciding whether a number is odd or even would be faster on the left for smaller numbers and on the right for larger numbers. In the ongoing investigation of how we make decisions, these influences will need to be uncovered and taken into account. Additionally, understanding which cognitive functions are exaptations of each other—for instance, if reasoning about number and perhaps other types of magnitudes such as temporal duration (Vallesi, Binns, & Shallice, 2008), volume of liquid

225

(Kirjakovski & Utsuki, 2012), and even degree of emotional expression (Holmes &

Lourenco, 2011) utilizes the same cognitive architecture as reasoning about space—could simplify the space of possible decision problems. In other words, describing the mechanism for one type of decision could effectively describe the mechanism for all similar types of decisions.

4.2 Value-Based and Social Decisions

In the study presented in section 3, we demonstrated that rhesus macaques mislead an opponent in a dynamic, zero-sum game. We found that shooters’ movements in the task reveal their goals more veridically when playing against a nonreactive goalie

(randomly-selected replays of past goalie movements). Moreover, we found that neuronal firing rates in the dorsolateral and dorsomedial prefrontal cortex (DL-PFC and

DM-PFC) are modulated by the degree of mismatch between the shooter’s revealed behavior and his actual goal, suggesting that these regions play a role in this deception.

Indeed, pharmacologically inactivating either region impaired the shooters’ performance in the game, at least in part by decreasing this mismatch between revealed and actual intentions. In both the recordings and the inactivations, we observed effects in DL-PFC earlier, and in DM-PFC later, in the trial for online generation of deception.

At the same time, we found stronger and earlier modulation in DM-PFC by trial-wide deceptive strategy and later modulation by trial outcome, suggesting that DM-PFC

226

incorporates reward outcome to select a strategy, which is carried out by both DL-PFC early in the interaction and DM-PFC later in the interaction.

Our results emphasize that social context can modulate the strategic decisions that an individual makes. When playing against a computer rather than a live opponent, the shooter behaved differently, concealing less about his forthcoming movements.

Although we did not record neuronal activity during this condition, we hypothesize that neuronal firing rates in DL-PFC and DM-PFC would no longer be modulated by the deception index. We further hypothesize that the ACC, which has been implicated in other-oriented preferences (Chang et al., 2013) and in predicting a cooperator’s decisions

(Haroush & Williams, 2015), as well as the superior temporal sulcus, which may be a macaque analog of the human temporoparietal junction (Mars, Sallet, Neubert, &

Rushworth, 2013) implicated in theory of mind and social decision-making (Behrens et al., 2009; C. D. Frith & Frith, 2006; D. Lee & Seo, 2016; Utevsky & Huettel, 2015), would show modulation during play against a live opponent that would be absent when subjects played against a computer.

Additionally, we found that neurons in both DL- and DM-PFC were modulated during the trial by the eventual trial outcome, and degree of modulation increased with time in the trial. In other words, both regions seemed to anticipate the reward outcome.

As noted in the Discussion of section 3, one interesting avenue for future study will be the learning process involved in the Penalty Shot task. As monkeys play against each

227

other repeatedly, how do they first develop the curvilinear trajectories and deceptive behavior observed here, and will they continue to refine and change their strategies as their opponent catches on? How are these novel strategies generated and then maintained by the brain? Answering such questions would help to fill important gaps in our understanding of value-based decision making, such as how the brain links different components of a complex behavior, such as different moves within a trajectory, to a single outcome, such as a juice reward (or lack thereof) at the end of the trial. Our results suggest that the DL-PFC and DM-PFC play a causal role in generating variable behavior, but they do not yet speak to the issue of how values are assigned to different strategies or moves, or how selection among these strategies occurs, although our trial-level findings point to a larger role for DM-PFC in these aspects of decision-making.

4.3 Manipulating Decision Circuits

One reason to study the neurobiological basis of decision-making is to restore decision-making abilities when they become impaired, and perhaps even to improve decision-making in healthy individuals. Indeed, impaired decision-making characterizes a number of disorders, such as substance abuse and addiction (Bechara & Damasio,

2002; Grant, 2000), multiple sclerosis (Radomski et al., 2015), Parkinson’s disease

(Mimura, Oeda, & Kawamura, 2006), Alzheimer’s disease (Hot, Ramdeen, Borg, Bollon,

& Couturier, 2013), Huntington’s disease, and frontotemporal dementia (Gleichgerrcht,

228

Ibáñez, Roca, Torralva, & Manes, 2010), and decision-making abilities decline following sleep deprivation (Killgore, Balkin, & Wesensten, 2006) and during aging even in the absence of other disorders (Boyle et al., 2012). Recently, non-invasive brain stimulation techniques, particularly transcranial magnetic stimulation (TMS), have emerged as a means to modulate activity in particular neural circuits, yielding amelioration of various clinical disorders (Berlim & Van Den Eynde, 2014; Connolly et al., 2012; Hanlon et al.,

2015; Kamble et al., 2014; J. Lee et al., 2016; Schulz, Gerloff, & Hummel, 2013; Slotema et al., 2014; Wassermann & Zimmermann, 2012) and changes in cognitive function (Nevler

& Ash, 2015; Sandrini et al., 2011). However, the neural mechanisms by which TMS exerts its effects are remain poorly understood.

In section 2.3, we completed a set of pilot studies to establish a rhesus macaque model for the exploration of how TMS affects neuronal activity to influence cognition and behavior. We used a custom-built TMS coil that included a switch between active and sham modes, as well as a space in the center to fit around a primate recording chamber. We first showed that ten minutes of offline low-frequency (1 Hz) active, as compared with sham, rTMS applied over the posterior parietal cortex impaired performance in a number-comparison task when two numerical arrays were presented simultaneously on a touch-sensitive monitor. In the same paradigm, sham as compared with active rTMS impaired performance in a color-comparison task.

229

Next we modified this task to involve central gaze fixation and sequential presentation of number and color stimuli, so that neuronal signals related to encoding each stimulus and comparing them to arrive at a decision could be disentangled. In this paradigm we found no consistent differences between performance following active and sham rTMS, except that the subject’s reaching movements were altered. We suspect that, since stimulation occurred over this subject’s cranial implant, effective stimulation did not reach deeply enough into the parietal cortex to affect numerical decisions, but did reach more dorsal parts of the parietal cortex that are involved in planning arm movements.

Finally, we recorded single-unit and multi-unit neuronal activity, as well as local field potentials, in the intraparietal sulcus before and after active and sham rTMS. While changes in activity were frequently observed, they varied in direction and magnitude from day to day, perhaps due to such variables as the type of neuron recorded from, the depth of recording, and the exact placement and orientation of the coil. Together these studies indicate that rTMS can alter cognition, behavior, and neural activity in the rhesus macaque model, and lay the groundwork for future studies manipulating and recording from putative decision-making circuits.

In addition to optimizing clinical application, understanding how TMS affects neuronal activity will clarify our interpretations of cognitive neuroscience experiments in which TMS is used (Miniussi, Harris, & Ruzzoli, 2013; Parkin, Ekhtiari, & Walsh,

230

2015). If behavior changes in some way following stimulation, then in order to accurately attribute that behavior to a patterns of activity in particular brain regions, it is necessary to know how that particular form of stimulation (e.g. high- or low-frequency rTMS; intermittent or theta burst stimulation) affects regions beneath and distal to the site of stimulation. Thus uncovering the neural impact of TMS can improve our understanding of the cognitive processes, such as decision-making, that it alters.

4.4 Conclusion

Overall, we have shown that when making abstract or social decisions, one nonhuman primate—the rhesus monkey—behaves remarkably similarly to humans.

Monkeys’ ordinal numerical decisions are influenced by spatial associations; they are able to compare fractions, and treat them as analog magnitudes; and they exhibit deception in a competitive social situation. We have uncovered part of the neural circuit involved in deception in the prefrontal cortex, and have found that the posterior parietal cortex may causally contribute to numerical decisions. Finally, we have contributed to a new line of research investigating the neural mechanisms underlying the effects of a non-invasive brain stimulation technique, transcranial magnetic stimulation, which could prove to be a fruitful treatment option for psychiatric disorders and other impairments of decision-making, and might provide a means to improve decision- making even in healthy individuals.

231

Appendix A: Non-invasive primate head restraint using thermoplastic masks

A.1 Introduction

Neuroscientific studies in awake, behaving non-human primates make immense and unique contributions to our understanding of brain function. Intracranial electrophysiological recordings offer unparalleled spatial and temporal precision compared with techniques like functional magnetic resonance imaging (Wurtz and

Sommer, 2006), and are of particular value in the monkey due both to structural and functional homologies with human brains and complex cognitive behavior (e.g. Adams et al., 2012; Chang et al., 2013; Goulas et al., 2014, 2014; Grefkes and Fink, 2005;

Hutchison et al., 2012; Miranda-Dominguez et al., 2014; Tsao et al., 2008). Such recordings often require the animal’s head to be immobilized to ensure stability of the recording electrode (but see (Roy and Wang, 2012; Schwarz et al., 2014) for examples of wireless recordings in freely-moving primates). Head immobilization further facilitates eye-tracking (Kimmel et al., 2012), a crucial component of many neuroscientific studies.

Other techniques facilitated by head fixation include microstimulation (Tehovnik et al.,

2006) and drug delivery via intracranial injection (Kurata and Hoffman, 1994; Roy et al.,

2014; Sommer and Wurtz, 2006) or a nebulizer (Chang et al., 2012). Recent interest in the 232

neural mechanisms underlying transcranial magnetic stimulation (TMS) has also focused attention on electrophysiological recordings in the monkey before and after

TMS (Mueller et al., 2014), and TMS application requires the head to be stationary.

The wide array of neuroscientific methods that depend on primate head fixation has led to the development of several immobilization devices. The majority of these devices involve implantation of screws or bolts into the subject’s skull, as well as the attachment of additional hardware – most commonly a head-post – that protrudes from the head and can be attached to the primate chair during experimental sessions (Adams et al., 2007; Betelak et al., 2001; Evarts, 1968; Foeller and Tychsen, 2002; Porter et al.,

1971). An alternative technique is the halo device, in which an aluminum ring surrounds the skull and is attached by several skull pins which require a smaller amount of skin to be removed and shallower penetrations into the skull (Friendlich, 1973; Isoda et al., 2005;

Pigarev et al., 1997, 2009).

While the stability and biocompatibility of such implants continues to improve, they still carry five main disadvantages. First, the animal typically needs to be placed under general anesthesia for the initial surgical attachment of the device to the skull.

Although routine, anesthesia is a risky procedure, with a wide range of potential harmful side-effects like hypothermia, hemorrhage, aspiration, respiratory insufficiency, cardiovascular emergencies, and death (Thurmon et al., 1996). Second, post-operative recovery can be stressful for the animal since analgesic and antibiotic drugs need to be

233

administered intramuscularly while the animal is awake, and socially-housed animals must be separated from one another. Third, implanted head-restraint devices carry the risk of failure and detachment from the skull, particularly when they are under pressure during repeated head fixation. Implant failure requires an emergency surgical procedure to close the wounds on the animal’s head and/or re-attach the device. Fourth, the protrusion of these devices themselves, as well as the hardware on the primate chair to which they attach, impede the application of TMS. During TMS, the large stimulating coil must lie flush with the scalp, centered over the brain region of interest, and the thick wires attaching the coil to the stimulator must have space to emerge from the set-up.

Finally, metal implants on an animal’s head cast shadows during magnetic resonance imaging, thus limiting brain scan utility.

A recent innovation uses a two-piece plastic head mold and a bar clamp holder to restrain monkeys’ heads during experiments, and does not require surgery or metal implants (Amemori et al., 2015). However, the thick plastic mold and large bar clamp create potential barriers to TMS delivery. The design may also limit intracranial recordings or injections in far posterior and temporal regions. Furthermore, the plastic mold covers the animal’s ears, potentially muffling sound and interfering with auditory testing. Finally, while the authors report that they were able to record neuronal activity and gaze position stably using this head-fixation technique, they note that animals do

234

still move their heads during reward delivery, a potential confound for studies of reward-related processing.

Here we describe a non-invasive means of primate head restraint that mitigates the problems described above. Thermoplastic masks are plastic mesh sheets that become soft upon heating and can be molded to any shape, which they retain when cooled.

Machado & Nelson (2011) adapted thermoplastic masks, which are used in human radiotherapy to stabilize the head, for eye-tracking in macaques. We expand upon their innovation by using these masks for electrophysiological recordings and TMS in macaques. We present a comprehensive comparison of the performance of thermoplastic masks and traditional head-posts on multiple measures relevant to neurobiology, including single-unit recording stability, eye-tracking accuracy, TMS coil placement,

MRI clarity, and financial cost to the lab. We find that thermoplastic masks compare favorably with implanted head-posts on each of these metrics.

A.2 Materials and Methods

All procedures were approved by Duke University Medical Center’s Institutional

Animal Care and Use Committee.

235

A.2.1 Subjects

Subjects were 24 adult rhesus macaque monkeys (Macaca mulatta), 10 female and

14 male, housed individually or in pairs at the Duke University Medical Center. 18 monkeys had head-posts, 5 monkeys had thermoplastic masks, and 1 monkey had a head-post that failed and was subsequently given a thermoplastic mask. Information from all monkeys was used in calculating cost to the lab. Eye-tracking data were collected from one mask monkey (Monkey Fe: female, 19 years old, 7.6kg) and one post monkey (Monkey Br: male, 17 years old, 13.4kg). Neuronal recording data were collected from one mask monkey (Monkey Sc: female, 17 years old, 6.8kg) and one post monkey (Monkey Da: male, 17 years old, 10kg). Implant size data were collected from two mask monkeys (Monkey Fr: female, 7 years old, 5.2kg; Monkey Sc: female, 17 years old, 6.8kg) and two post monkeys (Monkey Go: female, 7 years old, 5.4kg; Monkey Br: male, 17 years old, 13.4kg). Magnetic resonance images were collected from one mask monkey (Monkey Sc: female, 17 years old, 6.8kg) and one post monkey (Monkey Br: male, 17 years old, 13.4kg).

236

A.2.2 Thermoplastic masks

Fitting

The process of fitting a thermoplastic mask and applying it to an awake monkey is shown in Figure 55. We used a commercially-available reinforced thermoplastic mask

(Type-S IMRT Reinforced Style 22 Mask, product number MTAPUID2232, from CIVCO

Medical Solutions, Coralville, Iowa, USA). Since the mask would be used while the monkey was seated in a primate chair, it needed to be molded while the monkey was in a chair, since the angle of the monkey’s head relative to the chair neckplate needed to be replicated. For mask fitting, the animal was sedated with ketamine (3mg/kg, intramuscular (IM)) and dexdomitor (0.075-0.15mg/kg, IM), as well as ondansetron

(0.15mg/kg, IM) in cases where animals were known to become nauseated due to sedation. The monkey's head and face were shaved to achieve a closer fit, and petroleum jelly or vitamin E oil was applied to the animal’s head and any existing implant (i.e. a recording chamber), as well as the top of the primate chair, to make the mask easier to remove. Ophthalmic lubricant was applied to the animal’s eyes. The thermoplastic mask was submerged in hot water (at least 165°F) until soft. It was then removed from the water and allowed to cool until it was comfortable to touch with a bare hand. A hole was cut near the center of the mask where the animal’s nose would be to enable breathing. The monkey was then held in place in a primate chair by an experimenter who ensured that the animal’s head and neck stayed in a proper position for normal 237

breathing. Another experimenter placed the thermoplastic mask on the monkey’s head and molded it to fit the contours of the face and head, with an emphasis on the bridge of the nose and the brow ridge. Adequate space was left at the back of the head and sides of the muzzle and neck such that the mask could be removed once it hardened.

Locations for eye and mouth holes were marked on the mask. Once the mask had hardened, it was removed and the animal was returned to its cage and sedation was reversed with antisedan at the same dose volume as dexdomitor. Holes were cut in the eye and mouth regions of the mask such that the animal would be able to see and drink while wearing it. This entire procedure typically lasted 20-30 minutes. If any further modifications to the mask were needed, such as widening areas that seem too tight on the animal, a heat gun or boiling water was used to soften and reshape parts of the mask while it was off the animal. Bicycle helmet padding (Helmet Foam Kits, USA) was also attached inside some masks to improve comfort and stability by enabling a tighter fit while protecting the animal’s skin. Screws were placed through the top of the primate chair so that the mask could be attached to it using two to four wing nuts or bolts.

238

a b c

d e f

Training

We began training monkeys to accept the thermoplastic masks as early as the day after mask fitting. We used positive reinforcement to gradually condition the monkeys to wear the masks in their primate chairs, providing juice and treats as reinforcement during each of the following phases: 1) Presented the mask to the monkey for increasing amounts of time and increasing proximity to its head; 2) Touched the mask to the monkey’s head and face for increasing amounts of time; 3) Placed mask partway over the monkey’s head for increasing amounts of time; 4) Placed mask fully over the

239

monkey’s head for increasing amounts of time; 5) Attached mask to chair neckplate for increasing amounts of time. For some monkeys it was helpful to offer a syringe of juice through the mask’s mouth hole in order to properly position their heads for the mask to be attached to the chair. While this training sequence has been used successfully in our lab, other strategies may be just as effective for habituating different monkeys.

A.2.3 Head-post implantation

Affixing a head-post to a monkey’s skull required aseptic surgical conditions.

The animal was pre-anesthetized with ketamine (10mg/kg, IM) and atropine sulfate

(0.015mg/kg, IM) or ketamine (3mg/kg, IM) and dexdomitor (0.075-0.15mg/kg, IM); ondansetron (0.15mg/kg, IM) was also administered in cases of unplanned (emergency) surgeries, or in cases where animals were known to become nauseated due to sedation.

When sufficiently sedated, the animal was placed on a mobile treatment table and transported, covered, to the surgical pavilion. Blood samples (3mL) for running blood chemistry panels were taken, and an intravenous catheter was placed in the saphenous vein using aseptic technique. To anesthetize the animal sufficiently for intubation (as determined by corneal blink reflex and pedal reflex), the animal was masked

(isofluorane, 2-5%) and/or given propofol at a dose of 2-5mg/kg intravenously to effect, with repeated doses administered as needed for surgical induction. The animal was then intubated and placed on inhalant anesthesia, typically isofluorane at 0.5-5%. Monitors 240

for cardiac (EKG and pulse-oximeter) and respiratory (thermal respiration sensory) function were then placed and temperature was monitored rectally and maintained with hot water pad heaters. If the surgery was expected to require more than one hour to complete, replacement fluids were provided (typically Normosol-R, or similar) at 5-10 ml/kg/hr. The animal was placed sternally in a stereotaxic instrument (David Kopf

Instruments, Tujunga, California, USA). Surgeons (typically 2) then scrubbed and gowned while a third person, administering anesthesia, monitored the animal. The site of implantation on the animal’s head was shaved and scrubbed at least three times with a chlorhexadine disinfectant alternating with alcohol. The subsequent procedure varied depending on whether the animal was receiving a footed or non-footed head-post, which depended on the animal’s surgical history.

Non-footed head-post

A portion of the dorsal scalp was removed and the skin bordering the incision was retracted. The periosteum of the exposed bone was removed, and, if necessary, a portion of the temporal muscle was retracted or removed. A sterile drill was then used to cut 4-6 small holes. These holes were then tapped using an orthopedic bone tap, and surgical bone screws (typically surgical-grade titanium) were implanted in these prepared holes (equipment and bone screws typically by Synthes, Westchester,

Pennsylvania, USA; Veterinary Orthopedic Implants, St. Augustine, Florida, USA; or 241

Bristol-Myers/Zimmer, Warsaw, Indiana, USA). A sterile titanium head restraint post

(Crist Instruments, Damascus, Maryland, USA, or similar) was lowered stereotaxically into contact with the skull. The post and screws were then bonded together with sterile orthopedic bone cement (Cobalt HV or GHV (Biomet, Warsaw, Indiana, USA) or similar), or dental acrylic (Jet Denture Repair Acrylic (Lang Dental, Wheeling, Illinois,

USA) or similar). The wound was then checked for tight opposition to the cranial implant. Typically, tight opposition between the wound margin and the implant was achieved solely through the elasticity of the skin. If opposition between the wound margin and the implant was compromised, however, the wound margin was secured to the implant using sterile surgical adhesive (VetBond (3M, Maplewood, Minnesota,

USA). If opposition of the wound margin and the implant could not be achieved in this fashion, small incisions with Z-plastic re-anastamoses were performed to reduce any space between the wound margin and the implant. In this procedure, small incisions were made at oblique angles to the wound margin and small lenticular pieces of scalp were removed. The edges of this incision were then stitched together, thereby removing the slack in the skin and bringing the wound margin into tight contact with the cranial implant. Once the wound margin was secured to the implant, the implant margin was treated topically with triple antibiotic ointment (neomycin, polymyxin, bacitracin; Henry

Schein Inc., Melville, New York, USA). This entire procedure typically lasted between

1.5 and 4 hours.

242

Footed head-post

When possible, a "footed" head-post (Adams et al., 2007) was used to reduce the overall size of the implant and wound margin. In this procedure, a sagittal incision was made in the scalp at the desired position of the head-post and the skull was exposed. A sterilized footed titanium head-post (DKW Precision Machining, Manteca, California,

USA or similar) was lowered into position and was visually analyzed to determine the degree to which each foot would need to be bent in order to fit snugly against the skull.

The post was then removed and the feet were bent to match the curvature of the skull by bending them with sterile "bending forks" and/or tapping them against a sterilized metal rod with a sterile hammer. The head-post was again lowered into position to check the fit; subsequent adjustments to the curvature of the feet were sometimes necessary to ensure the closest fit possible to the skull. Once a close fit was achieved, the screw holes on opposite side of the post were marked on the skull using a surgical pen and drilled and tapped using sterile instruments. Titanium or ceramic bone screws (Synthes;

Thomas Recording, Giessen, Germany; or similar) were then driven into each hole. Once all screws were in place and were sufficiently tight, hydroxyapatite bone matrix material

(Mimix (Biomet) or similar) was used to fill any gaps that existed between the skull and the feet to facilitate bone growth under and over the titanium prosthetic. Once complete, the area was irrigated with physiological saline, and the incision was closed, first the 243

galea, with absorbable suture, followed by the skin. This entire procedure typically lasted between 1.5 and 4 hours.

Recovery

Following surgery, a course of analgesic (buprenorphine, 0.01mg/kg, every 6-12 hours for 72 hours), antibiotic (enrofloxacin, 5mg/kg, once daily for 7-10 days), and anti- inflammatory (ketorolac tromethamine, 1mg/kg, once daily for 3 days) began. The animal was also observed for abnormal behavior or physical signs indicative of neurological insult and sometimes received additional drugs (including dexamethasone trisodium phosphate, midazolam, lorazepan, diazepam, and cefazolin) as necessary per veterinary consult. Following a first head-post implantation, animals were allowed at least 28 days for recovery. In the case of subsequent head-post surgeries, animals were allowed at least 7 days for recovery.

Training

We began conditioning monkeys to head fixation after the recovery period. As with mask training (section 2.2.2) we used positive reinforcement, delivering juice or treats when monkeys allowed us to touch their head-post, then to secure their post in the post-holder attached to the chair for increasing amounts of time.

244

A.2.4 Eye-tracking

Gaze position was sampled at 1000 Hz using an Eyelink 1000 infrared eye-tracker

(SR Research, Mississauga, Ontario, Canada). Monkeys performed a visual orienting task that began when a 1° white square appeared at the center of an otherwise black screen. Monkeys had to fixate within 1.5° of the center of the white square (400 ms). A second white 1.5° square then appeared in one of eight evenly distributed locations 10° in the periphery. As soon as the target appeared, the central fixation point disappeared, and the monkey was required to shift gaze to the target (±2.5°) and maintain fixation

(300 milliseconds) to receive a juice reward. Stimulus presentation and data collection were performed using MATLAB (The Mathworks, Natick, Massachusetts, USA) and the

Eyelink and Psychophysics Toolbox (Brainard, 1997; http://psychtoolbox.org/).

We compared data from one session with a mask monkey and one session with a post monkey. Analyses were performed using custom-scripts written in MATLAB. We first considered drift in fixation position over the course of the session by fitting linear models separately to the horizontal and vertical gaze positions of each monkey on each trial for the first 450 trials in the session, where a value for each trial was determined by taking the mean over all eye samples during the fixation period on that trial. We used t- tests to compare the slopes (beta coefficients) of the regression lines for the horizontal gaze position between monkeys, and the vertical gaze position between monkeys. Next 245

we considered reliability of saccadic endpoints by taking the mean of the horizontal and vertical gaze positions over all eye samples during the target fixation period, and computing the variances of the horizontal and vertical positions over 40 trials for each target location. We used F-tests to compare the variances at each target location between monkeys.

A.2.5 Neuronal recording

Recording chamber implantation

Sterile surgeries were performed on several animals in order to implant cylinders permitting electrode/cannula access to the brain. In this procedure, the monkey was anesthetized and prepared for surgery as described above (section A.2.3). The anesthetized animal was placed in a stereotax (David Kopf Instruments). The placement of cylinders was determined by the stereotaxic coordinates of the target brain structure.

Sterile surgeons first removed a circular region of scalp over the pre-selected stereotaxic coordinates. The periosteum was removed and, if necessary, a portion of the temporal muscle was retracted or removed. A circular section of skull was then removed using a sterile drill and bits or a sterile trephine. Two to six additional surgical screws were implanted as above (section A.2.3). A stereotax was used to lower the sterile cylinder

(titanium or polysulfone, Crist Instruments) over the opening in the skull. A sterile bone

246

replacement material (Mimix (Biomet)) was used to fill any gaps that existed between the skull and the base of the recording chamber (Adams et al., 2007). The cylinder was then bonded to the skull and screws with sterile bone cement (Cobalt GHV or HV

(Biomet)) or dental acrylic (Jet Denture Repair Acrylic (Lang Dental)). The interior of the cylinder was then washed with sterile saline and in some cases antibiotic (typically

Gentamicin: 0.62cc, 40mg/ml concentration), dried, and closed with a sterile cap which mates onto the cylinder. This entire procedure typically lasted between 1.5 and 4 hours.

Electrophysiological data collection & analysis

After a post-operative recovery period, neuronal recording studies began. At the beginning of each recording session, the animal was head-restrained and the sterile recording cylinder was opened under aseptic conditions. The cylinder was cleaned with

1% povidone-iodine and then repeatedly lavaged with 0.9% sterile saline. A grid (Crist 247

Instruments) was inserted into the cylinder, and a sterile guide tube was placed through a grid hole and then through the animal’s dura mater. A hydraulic microdive (David

Kopf Instruments) was mounted on the recording cylinder, and a tungsten electrode about 0.005 inches in diameter with an impedance between 1 and 15 megaohms (FHC,

Bowdoin, Maine, USA) was inserted into the guide tube and attached to the microdrive so it could be moved through the brain microns at a time. As the electrode advanced, voltage at its tip was processed through a series of amplifiers and computer software

(Plexon, Dallas, Texas, USA). At the end of the daily session the electrode was withdrawn, the guide tube, grid, and microdrive were removed, and the cylinder was again lavaged with 1% povidone-iodine and sterile saline, then resealed with a sterile cap.

Data from one recording session with a mask monkey and one recording session with a post monkey were analyzed here. Plexon’s Offline Sorter was used to automatically sort waveforms as action potentials from an isolated unit or as background activity. In addition to visually inspecting the voltage traces over time, stability of the recording was quantified using the 2D Pseudo-F metric of sort quality provided by Plexon, including unsorted waveforms as a unit for comparison.

248

A.2.6 Measurement of head movement

In order to directly assess how effective the thermoplastic masks were for head restraint, we took measurements of head movement using a gyroscope. Specifically, we attached an iPhone 6 (Apple Inc., Cupertino, California, USA) via an adhesive strip to the recording cylinders of two mask monkeys and two post monkeys. We sampled rotational velocity in each of three axes at near 100Hz using the Sensor Kinetics Pro application (Innoventions Inc., Houston, Texas, USA; http://www.rotoview.com/sensor_kinetics.htm ) for ten minutes while juice was periodically delivered to the monkey. We obtained power spectra for each animal’s movement in each axis by using a Fast Fourier Transform in MATLAB.

A.2.7 Magnetic resonance imaging

To localize brain regions for electrophysiology and TMS studies, high-resolution structural MRI’s of monkeys’ brains were obtained. Since the animal needed to be completely still during scanning, the monkey was sedated using ketamine, dexdomitor, and sometimes ondansetron (see section A.2.2) and then placed on inhalant anesthesia

(see section A.2.3) for the duration of the scan. Scans were performed with a 3-Tesla

(Trio, Siemens Medical Systems, Munich, Bavaria) MR Imaging and Spectroscopy

System at Duke University Medical Center’s Center for Advanced Magnetic Resonance

249

Development. During scans, the monkey’s recording chamber was filled with sterile saline, and non-ferrous electrode tips were placed in a grid within the chamber, so that the chamber would be visible in the MR images and recording electrode trajectories could be determined. The scan sequence typically included localizers, coronal T1 MP-

RAGE (magnetization-prepared rapid acquisition with gradient echo) 1mm, coronal T2

0.5mm, sagittal T1 MP-RAGE 1mm, and sagittal T2 0.5mm, and could also include a magnetic resonance angiogram, diffusion tensor imaging, and resting-state functional

MRI.

A.2.8 Calculation of cost

Using records of all animals in our lab, we counted how many times each animal had been sedated or undergone surgery for a procedure related to head fixation. We also noted the dates over which the animals had a head-post or thermoplastic mask, so that we could compute the number of sedations and surgeries per year during the relevant periods of their lives. We summed the price of all materials and procedures typically used in post- or mask-related sedations and post-related surgeries, and computed a total cost for each animal based on which type of head-fixation-related procedures s/he had undergone, then divided this by the number of years s/he had a post or mask to produce the cost per year. In averaging these values to compute the cost per animal per year for posts and for masks, each animal was considered either a “post” monkey or a “mask” 250

monkey, except for one animal who was considered a “post” monkey for an initial epoch, and a “mask” monkey for a second epoch after her post failed and she was fitted with a thermoplastic mask.

A.3 Results

First, we note that conditioning monkeys to wear the thermoplastic masks generally took no more than two weeks, equivalent to head-post training (unpublished observations). Monkeys readily performed cognitive tasks for juice rewards requiring various response modalities, including eye movements, joysticks, and touch screens. In the sections that follow, we offer comparisons between the masks and posts on eye- tracking accuracy, neuronal recording stability, head movement, MRI clarity, TMS coil placement, and cost to the lab.

A.3.1 Eye-tracking accuracy

We successfully trained a monkey wearing a thermoplastic mask to perform eye movement tasks with similar precision to that of a monkey with a head-post. Figure 56 shows sample eye traces during the first 120 trials of a visual orienting task (see section

A.2.4) for each monkey. At each of the eight target locations, variance in saccade end point was no higher for the mask then the post monkey. Additionally, there was less

251

drift in central fixation position for a fully-trained mask monkey than a fully-trained post monkey over the course of a session (regression slopes for horizontal gaze position, mask: -0.00032, post: -0.00061; for vertical gaze position, mask: -0.00065, post: 0.00118)

(Figure 57), indicating that eye tracking with a mask is at least as stable as with a post.

252

A.3.2 Neuronal recording stability

Thermoplastic masks also provided sufficient stability to perform intracranial single-unit electrophysiology. Example action potential voltage traces recorded from the parietal cortex of a mask and a post monkey at the beginning and 50 minutes into the session are shown in Figure 58. The similarity in waveform between these two time points illustrates the stability of our recordings. The Pseudo-F value provided by

Plexon’s Offline Sorter, which would be maximal if all waveforms over the course of the session had the same shape and amplitude, was 32020 for the post monkey and 36290 for the mask monkey, indicating that the quality of isolation was at least as high for the mask monkey as the post monkey.

253

A.3.3 Head movement

To further test whether thermoplastic masks truly achieve head fixation comparable to head posts, we directly measured head movement at the location of animals’ recording cylinders using a gyroscope. We collected data from two mask and two post monkeys, including one calm and one restless monkey in each condition. We found that the raw movement traces were similar between mask and post monkeys

(Figure 59), and the power spectra of these movement traces were also similar (Figure

60). These results demonstrate that thermoplastic masks restrain the head just as well as head posts.

254

Figure 59: Head movement traces. Ten minutes of rotational velocity data recorded from the heads of mask and post monkeys are shown. The left column shows a calm mask monkey (red) and a calm post monkey (blue), while the right column shows a restless mask money (red) and a restless post monkey (blue). The top row shows rotation around the “roll” axis, the middle row shows rotation around the “pitch” axis, and the bottom row shows rotation around the “yaw” axis.

255

256

A.3.4 MRI clarity

For neuronal recording and stimulation studies, it is useful to obtain high- resolution structural brain scans to localize target regions. While our monkeys’ head- posts and bone screws are nonferrous and therefore safe to introduce into the MRI scanning environment, they are still metal and cast a “shadow” (that is, create local distortions) on the resulting image, due largely to metal-induced field inhomogeneities which interfere with the imaging gradient fields (Lu et al., 2009; Lüdeke et al., 1985).

Thus it has been our experience that MRIs of monkeys without a head-post – i.e. monkeys that wear thermoplastic masks – are clearer and easier to interpret than those of monkeys with a head-post. Sample MR images from mask and post monkeys are shown in Figure 61 to demonstrate this point.

257

A.3.5 TMS coil placement

To effectively stimulate a particular brain region with TMS, the stimulating coil must lie as close to the scalp as possible over the region of interest. The larger the cranial implant an animal has, the fewer brain regions can be reached by TMS. Animals without head-posts either do not need an implant at all or need an implant only large enough to support a recording cylinder. We measured the implant sizes of our two monkeys who have thermoplastic masks and recording cylinders, as well as two monkeys with head-

258

posts and recording cylinders in similar locations to the mask monkeys. As the irregular shape of these implants makes it difficult to estimate their volume, we report the length

(anterior-posterior), width (medial-lateral), and height (dorsal-ventral) of each implant at its longest point in each direction. Implants were smaller for mask than post monkeys.

For the two monkeys with frontal pole cylinders, sizes were 4.7 cm long, 4.4 cm wide, 1.3 cm high (mask), and 8.5 cm long, 6.5 cm wide, 1.5 cm high (post); and for the two monkeys with parietal cylinders, sizes were 6.0 cm long, 4.5 cm wide, 0.9 cm high

(mask), and 7.5 cm long, 5.0 cm wide, 0.7 cm high (post).

We further note that lack of a head-post allows the stimulating coil more freedom to be moved around the head. There are two reasons for this: first, the head- post itself can obstruct the coil’s placement. Second, stabilizing a monkey’s head via a head-post requires a post-holder above the chair. This is typically attached to the chair via two side panels to the left and right of the monkey’s head. These panels further block the stimulating coil, and they are unnecessary with thermoplastic masks (Figure 62). The only disadvantage of the mask for TMS coil placement is that it is 0.3 cm thick, which could increase the distance between the coil and the brain. However, in our applications, we stimulate around the recording cylinder, where we have cut away part of the mask and where distance between coil and scalp is determined by the thickness of the bone cement.

259

A.3.6 Cost to lab

We first computed the total cost of all the resources required for a typical head- post surgery ($1522.93), a typical non-surgical head-post-related procedure requiring sedation ($63.86), and a typical thermoplastic mask-fitting procedure ($133.87). We then used these values to determine the cost per year of each animal’s head-fixation-related procedures. We found that thermoplastic masks cost substantially less than head-posts

(mean cost per animal per year: masks, $238, standard error of the mean (SEM) = $233; posts, $1166, SEM = $65; two-sample t-test: t(23) = 2.193, p = 0.0387).

Additionally, we considered the number of research days lost due to head- fixation-related procedures as an important non-monetary cost. Head-post surgeries require several days of recovery time before the animal can be used in an experiment,

260

whereas the short-term sedations required for mask-fitting enable the animal to be used the next day. We computed the number of research days lost due to surgeries and procedures requiring short-term sedations related to head fixation for each animal, and divided this number by the number of years the animal spent in the lab with a head fixation device, to determine the number of research days lost per animal per year. We found that thermoplastic masks resulted in substantially fewer research days lost than head-posts (mean number of days lost per animal per year: masks, 1.78, SEM = 0.49; posts, 13.55, SEM = 2.24; two-sample t-test: t(23) = 2.897, p = 0.0081).

A.4 Discussion

Here we show that thermoplastic masks can be used effectively to restrain the head in non-human primates for neurobiological studies. Fitting masks to individual rhesus macaques is a brief, simple procedure that is easy to learn. Once created, these non-invasive masks can be used day after day to restrain monkeys’ heads during experiments. Although monkeys were initially wary of these novel objects, they quickly became conditioned to wearing them through positive reinforcement, and performed cognitive tasks using eye movements, joysticks, and touch screens. Moreover, we were able to record intracranial neuronal activity from animals wearing these masks.

Thermoplastic masks yield a degree of head fixation comparable to traditional head-restraint prostheses. We assessed this directly by recording rotational head

261

movement using a gyroscope affixed to monkeys’ recording cylinders. Similar amounts of movement were observed in monkeys wearing masks and those with head-posts.

Additionally, we examined gaze position data during a saccade task, a common paradigm in neuroscientific studies. Drift in eye-tracker accuracy over the course of a behavioral session, as well as variability in saccade end-point, were at least as low for a monkey wearing a mask as for one with a head-post. Neuronal recording stability was also similar between mask and post subjects.Thus thermoplastic masks may be used instead of head-posts in experiments involving intracranial electrophysiology or high- resolution eye-tracking.

Another neuroscientific method requiring head fixation is TMS. TMS is an increasingly popular method for noninvasive neuromodulation in humans (Hovington et al., 2013; Luber and Lisanby, 2014; Sandrini et al., 2011). Some labs have begun to use

TMS in monkeys in order to explore the contribution of various brain regions to cognitive tasks (Gerits et al., 2011; Gu and Corneil, 2014; Valero-Cabre et al., 2012) and which could be used to study the connectivity between regions. Other labs – including ours (Mueller et al., 2014) – are applying TMS to monkeys in combination with intracranial electrophysiological recordings in order to explore the neural basis of its effects. Subjects’ heads need to be still during TMS application. By using thermoplastic masks rather than head-posts, more of the monkey’s head is accessible, enabling better targeting of TMS to particular brain regions.

262

Thermoplastic masks offer other advantages over head-posts. For example, the reduced amount of metal on monkeys’ heads results in clearer magnetic resonance images, which are useful for targeting brain regions of interest during neuronal recordings or stimulation. Additionally, fitting a thermoplastic mask results in fewer research days lost and lower monetary costs to the lab than implanting a head-post.

Finally, masks are preferable from an animal welfare perspective since they are completely non-invasive and do not require subjects to undergo general anesthesia.

Although we did not directly compare thermoplastic masks to the recently- developed head mold technique (Amemori et al., 2015), we note some ways in which masks might be preferable. First, as previously mentioned, the head mold is quite thick and covers the far posterior and temporal regions of the animal’s head, making TMS coil placement and intracranial recordings in those regions difficult. With thermoplastic masks, a TMS coil can be moved freely around the head, and an opening for a recording chamber could be cut from any part of the mask. The masks also should not muffle sound as the head molds might. Furthermore, thermoplastic masks may provide better head stability: Amemori and colleagues note that animals’ heads did move during reward delivery when restrained with a head mold, which we did not observe in our subjects, and the gaze fixation window in their ocular task was 8° compared with 1.5° for the central fixation point and 2.5° for the peripheral targets in our task. Nonetheless,

Amemori and colleagues were able to perform single-unit electrophysiology using a

263

head mold. One potential advantage of the head mold design over masks is that the head molds might be more comfortable for an animal, since its head can move slightly and nothing touches its face. Also, the masks largely prevent delivery of an air-puff stimulus to the animal’s face, which the head mold design appears to allow. Finally, the masks obscure facial movements, which could preclude visual communication in social tasks; the head mold design leaves most of the animal’s face visible.

A.5 Conclusion

Head fixation of awake, behaving monkeys can be achieved non-invasively using thermoplastic masks. Various types of experiments may be performed with monkeys wearing masks, including those involving cognitive tasks, eye-tracking, intracranial electrophysiology, and transcranial magnetic stimulation. These masks compare favorably to traditional head-posts in terms of experimental outcomes, cost to the lab, and animal welfare. Thus thermoplastic masks may be added to the toolkit of options available that stabilize monkeys’ heads for neuroscientific studies.

264

References

Abe, N. (2009). The neurobiology of deception: evidence from neuroimaging and loss-of- function studies. Current Opinion in Neurology, 22(6), 594–600. http://doi.org/10.1097/WCO.0b013e328332c3cf

Abe, N. (2011). How the brain shapes deception: an integrated review of the literature. The Neuroscientist : a Review Journal Bringing Neurobiology, Neurology and Psychiatry, 17(5), 560–574. http://doi.org/10.1177/1073858410393359

Abe, N., Fujii, T., Hirayama, K., Takeda, A., Hosokai, Y., Ishioka, T., et al. (2009). Do parkinsonian patients have trouble telling lies? The neurobiological basis of deceptive behaviour. Brain : a Journal of Neurology, 132(5), 1386–1395. http://doi.org/10.1093/brain/awp052

Abe, N., Fujii, T., Ito, A., Ueno, A., Koseki, Y., Hashimoto, R., et al. (2014). The neural basis of dishonest decisions that serve to harm or help the target. Brain and Cognition, 90, 41–49.

Abe, N., Suzuki, M., Mori, E., Itoh, M., & Fujii, T. (2007). Deceiving Others: Distinct Neural Responses of the Prefrontal Cortex and Amygdala in Simple Fabrication and Deception with Social Interactions. Journal of Cognitive Neuroscience, 19(2), 287–295. http://doi.org/10.1073/pnas.91.6.2012

Abe, N., Suzuki, M., Tsukiura, T., Mori, E., Yamaguchi, K., Itoh, M., & Fujii, T. (2006). Dissociable roles of prefrontal and anterior cingulate cortices in deception. Cerebral Cortex, 16, 192–199. http://doi.org/10.1093/cercor/bhi097

Adachi, I. (2014). Spontaneous Spatial Mapping of Learned Sequence in Chimpanzees: Evidence for a SNARC-Like Effect. PLoS One, 9(3), e90373. http://doi.org/10.1371/journal.pone.0090373.g002

Adams, D.L., Economides, J.R., Jocson, C.M., and Horton, J.C. (2007). A Biocompatible Titanium Headpost for Stabilizing Behaving Monkeys. J. Neurophysiol. 98, 993–1001.

Adams, G.K., Watson, K.K., Pearson, J., and Platt, M.L. (2012). Neuroethology of decision-making. Curr. Opin. Neurobiol. 22, 982–989.

Agrillo C, Piffer L, Bisazza A (2011) Number versus continuous quantity in numerosity judgments by fish. Cognition 119:281-287. doi: 10.1016/j.cognition.2010.10.022

265

Agrillo, C., Piffer, L., & Bisazza, A. (2010). Large Number Discrimination by Mosquitofish. (G. Chapouthier, Ed.)PLoS One, 5(12), e15232. doi:10.1371/journal.pone.0015232.g006

Aïn, S. A., Giret, N., Grand, M., Kreutzer, M., & Bovet, D. (2008). The discrimination of discrete and continuous amounts in African grey parrots (Psittacus erithacus). Animal Cognition, 12(1), 145–154. doi:10.1007/s10071-008-0178-8

Albantakis, L., & Deco, G. (2011). Changes of Mind in an Attractor Network of Decision- Making. PLoS Computational Biology, 7(6), e1002086. http://doi.org/10.1371/journal.pcbi.1002086

Aleman, A. (2013). Use of Repetitive Transcranial Magnetic Stimulation for Treatment in Psychiatry. Clinical Psychopharmacology and Neuroscience, 11(2), 53. http://doi.org/10.9758/cpn.2013.11.2.53

Allen, E. A., Pasley, B. N., Duong, T., & Freeman, R. D. (2007). Transcranial Magnetic Stimulation Elicits Coupled Neural and Hemodynamic Consequences. Science, 317(5846), 1918–1921. http://doi.org/10.1126/science.1146426

Allendorfer, J. B., Storrs, J. M., & Szaflarski, J. P. (2012). Changes in white matter integrity follow excitatory rTMS treatment of post-stroke aphasia. Restorative Neurology and Neuroscience, 30(2), 103–113. http://doi.org/10.3233/RNN-2011-0627

Amassian, V. E., Cracco, R. Q., Maccabee, P. J., Cracco, J. B., Rudell, A. P., & Eberle, L. (1998). Transcranial magnetic stimulation in study of the visual pathway. Journal of Clinical Neurophysiology : Official Publication of the American Electroencephalographic Society, 15(4), 288–304.

Amemori, S., Amemori, K., Cantor, M.L., and Graybiel, A.M. (2015). A non-invasive head-holding device for chronic neural recordings in awake behaving monkeys. J. Neurosci. Methods 240, 154–160.

Amici, F., Call, J., & Aureli, F. (2009). Variation in withholding of information in three monkey species. Proceedings of the Royal Society B: Biological Sciences. http://doi.org/10.1098/rspb.2009.0759

Andersen, R. A., & Cui, H. (2009). Intention, Action Planning, and Decision Making in Parietal-Frontal Circuits. Neuron, 63(5), 568–583. http://doi.org/10.1016/j.neuron.2009.08.028

Andersen, R. A., Asanuma, C., Essick, G., & Siegel, R. M. (1990). Corticocortical

266

Connections of Anatomically and Physiologically Defined Subdivisions Within the Inferior Parietal Lobule. The Journal of Comparative Neurology, 296, 65–113.

Andersen, R. A., Brotchie, P. R., & Mazzoni, P. (1992). Evidence for the lateral intraparietal area as the parietal eye field. Current Opinion in Neurobiology, 2(6), 840– 846. http://doi.org/10.1016/0959-4388(92)90143-9

Andres, M., Pelgrims, B., Michaux, N., Olivier, E., & Pesenti, M. (2011). Role of distinct parietal areas in arithmetic: An fMRI-guided TMS study. NeuroImage, 54(4), 3048– 3056. http://doi.org/10.1016/j.neuroimage.2010.11.009

Ardila, A., & Rosselli, M. (2002). Acalculia and Dyscalculia. Neuroopsychology Review, 12(4), 179–231.

Arikan, R., Blake, N. M. J., Erinjeri, J. P., Woolsey, T. A., Giraud, L., & Highstein, S. M. (2002). A method to measure the effective spread of focally injected muscimol into the central nervous system with electrophysiology and light microscopy. Journal of Neuroscience Methods, 118(1), 51–57. http://doi.org/10.1016/S0165-0270(02)00143-7

Asch, S. E. (1956). Studies of independence and conformity: I. A minority of one against a unanimous majority. Psychological Monographs: General and Applied.

Ashby, F. G., & Maddox, W. T. (2005). Human Category Learning. Annual Review of Psychology, 56(1), 149–178. http://doi.org/10.1146/annurev.psych.56.091103.070217

Aydin-Abidin, S., Moliadze, V., Eysel, U. T., & Funke, K. (2006). Effects of repetitive TMS on visually evoked potentials and EEG in the anaesthetized cat: dependence on stimulus frequency and train duration. The Journal of Physiology, 574(2), 443–455. http://doi.org/10.1113/jphysiol.2006.108464

Bagherzadeh, Y., Khorrami, A., Zarrindast, M. R., Shariat, S. V., & Pantazis, D. (2016). Repetitive transcranial magnetic stimulation of the dorsolateral prefrontal cortex enhances working memory. Experimental Brain Research, 1–12. http://doi.org/10.1007/s00221-016-4580-1

Bajbouj, M., & Padberg, F. (2014). A perfect match: noninvasive brain stimulation and psychotherapy. European Archives of Psychiatry and Clinical Neuroscience, 264(1), 27– 33. http://doi.org/10.1007/s00406-014-0540-6

Barker, A. T., Jalinous, R., & Freeston, I. L. (1985). Non-invasive magnetic stimulation of human motor cortex. The Lancet, 1106–1107.

267

Baron-Cohen, S. (1992). Out of Sight or Out of Mind? Another Look at Deception in Autism. Journal of Child Psychology and Psychiatry, 33(7), 1141–1155. http://doi.org/10.1111/j.1469-7610.1992.tb00934.x

Barraclough, D. J., Conroy, M. L., & Lee, D. (2004). Prefrontal cortex and decision making in a mixed-strategy game. Nature Neuroscience, 7(4), 404–410. http://doi.org/10.1038/nn1209

Barry, K. L. (2014). Sexual deception in a cannibalistic mating system? Testing the Femme Fatale hypothesis. Proceedings of the Royal Society B: Biological Sciences, 282(1800), 20141428–20141428. http://doi.org/10.1017/S1464793103006158

Bartra, O., McGuire, J. T., & Kable, J. W. (2013). The valuation system: a coordinate- based meta-analysis of BOLD fMRI experiments examining neural correlates of subjective value. NeuroImage.

Batista, A. P., Buneo, C. A., Snyder, L. H., & Andersen, R. A. (1999). Reach Plans in Eye- Centered Coordinates. Science, 285(5425), 257–260. http://doi.org/10.1126/science.285.5425.257

Battaglia-Mayer, A., & Ferraina, S. (2000). Early coding of reaching in the parietooccipital cortex. Journal of Neurophysiology, 83(4), 2374–2391.

Baumgartner, T., Fischbacher, U., Feierabend, A., Lutz, K., & Fehr, E. (2009). The neural circuitry of a broken promise. Neuron, 64, 756–770.

Bechara, A., & Damasio, H. (2002). Decision-making and addiction (part I): impaired activation of somatic states in substance dependent individuals when pondering decisions with negative future consequences. Neuropsychologia, 40(10), 1675–1689. http://doi.org/10.1016/S0028-3932(02)00015-5

Behrens, T. E. J., Hunt, L. T., & Rushworth, M. F. S. (2009). The Computation of Social Behavior. Science, 324(5931), 1160–1164. http://doi.org/10.1126/science.1169694

Beran, M. J. (2008). Monkeys (Macaca mulatta and Cebus apella) track, enumerate, and compare multiple sets of moving items. Journal of Experimental Psychology: Animal Behavior Processes, 34(1), 63–74. doi:10.1037/0097-7403.34.1.63

Berlim, M. T., & Van Den Eynde, F. (2014). Repetitive transcranial magnetic stimulation over the dorsolateral prefrontal cortex for treating posttraumatic stress disorder: an exploratory meta-analysis of randomized, double-blind and sham-controlled trials. Canadian Journal of Psychiatry. Revue Canadienne De Psychiatrie, 59(9), 487–496.

268

Bermudez, M. A., & Schultz, W. (2010). Responses of amygdala neurons to positive reward-predicting stimuli depend on background reward (contingency) rather than stimulus-reward pairing ( …. Journal of Neurophysiology, 103, 1158–1170.

Bestmann, S. (2008). The physiological basis of transcranial magnetic stimulation. Trends in Cognitive Sciences, 12(3), 81–83.

Betelak, K.F., Margiotti, E.A., Wohlford, M.E., and Suzuki, D.A. (2001). The use of titanium implants and prosthodontic techniques in the preparation of non-human primates for long-term neuronal recording studies. J. Neurosci. Methods 112, 9–20.

Bhatt, M. A., Lohrenz, T., Camerer, C. F., & Montague, P. R. (2010). Neural signatures of strategic types in a two-person bargaining game. Proceedings of the National Academy of Sciences of the United States of America, 107(46), 19720–19725. http://doi.org/10.1073/pnas.1009625107

Bhatt, S., Mbwana, J., Adeyemo, A., Sawyer, A., & Hailu, A. (2009). Lying about facial recognition: an fMRI study. Brain and ….

Binder, J. R., Liebenthal, E., Possing, E. T., Medler, D. A., & Ward, B. D. (2004). Neural correlates of sensory and decision processes in auditory object identification. Nature Neuroscience, 7(3), 295–301. http://doi.org/10.1038/nn1198

Blangero, A., Ota, H., Rossetti, Y., Fujii, T., Ohtake, H., Tabuchi, M., et al. (2010). Systematic retinotopic reaching error vectors in unilateral optic ataxia. Cortex, 46, 77–93.

Bogale, B. A., Kamata, N., Mioko, K., & Sugita, S. (2011). Quantity discrimination in jungle crows, Corvus macrorhynchos. Animal Behaviour, 82(4), 635–641. doi:10.1016/j.anbehav.2011.05.025

Bond, C. F., Jr, & Robinson, M. (1988). The evolution of deception. Journal of Nonverbal Behavior, 12(4), 295–307. http://doi.org/10.1007/BF00987597

Bond, R., & Smith, P. B. (1996). Culture and conformity: A meta-analysis of studies using Asch's (1952b, 1956) line judgment task. Psychological Bulletin, 119(1), 111–137.

Boroojerdi, B., Meister, I. G., Foltys, H., Sparing, R., Cohen, L. G., & Topper, R. (2002). Visual and motor cortex excitability: a transcranial magnetic stimulation study. Clinical Neurophysiology, 113, 1501–1504.

Boureau, Y.-L., Sokol-Hessner, P., & Daw, N. D. (2016, February). Deciding how to

269

decide: Self-control and meta-decision making. Retrieved February 23, 2016, from http://www.princeton.edu/~ndaw/bsd15.pdf

Boyer, T. W., Levine, S. C., & Huttenlocher, J. (2008). Development of proportional reasoning: Where young children go wrong. Developmental Psychology, 44(5), 1478– 1490. http://doi.org/10.1037/a0013110

Boyle, P. A., Yu, L., Wilson, R. S., Gamble, K., Buchman, A. S., & Bennett, D. A. (2012). Poor Decision Making Is a Consequence of Cognitive Decline among Older Persons without Alzheimer’s Disease or Mild Cognitive Impairment. Journals.Plos.org, 7(8), e43647. http://doi.org/10.1371/journal.pone.0043647

Brainard, D. H. (1997). The Psychophysics Toolbox. Spatial Vision, 10(4), 433–436. http://doi.org/10.1163/156856897X00357

Brannon, E., & Terrace, H. (1998). Ordering of the numerosities 1 to 9 by monkeys. Science, 282(5389), 746.

Brinke, ten, L., Porter, S., & Baker, A. (2012). Darwin the detective: Observable facial muscle contractions reveal emotional high-stakes lies. Evolution and Human Behavior, 33(4), 411–416. http://doi.org/10.1016/j.evolhumbehav.2011.12.003

Britten, K. H., Newsome, W. T., Shadlen, M. N., Celebrini, S., & Movshon, J. A. (1996). A relationship between behavioral choice and the visual responses of neurons in macaque MT. Visual Neuroscience, 13, 87–100.

Britten, K. H., Shadlen, M. N., & Newsome, W. T. (1992). The analysis of visual motion: a comparison of neuronal and psychophysical performance. The Journal of Neuroscience, 12(12), 4745–4765.

Bro Jørgensen, J., & Pangle, W. M. (2010). Male Topi Antelopes Alarm Snort Deceptively to Retain Females for Mating. The American Naturalist, 176(1), E33–E39. http://doi.org/10.1086/653078

Brody, C. D., & Hanks, T. D. (2016). Neural underpinnings of the evidence accumulator. Current Opinion in Neurobiology. http://doi.org/10.1016/j.conb.2016.01.003

Brody, C. D., Hernandez, A., Zainos, A., & Romo, R. (2003). Timing and neural encoding of somatosensory parametric working memory in macaque prefrontal cortex. Cerebral Cortex.

Bromberg-Martin, E. S., Matsumoto, M., Hong, S., & Hikosaka, O. (2010). A pallidus-

270

habenula-dopamine pathway signals inferred stimulus values. Journal of Neurophysiology, 104, 1068–1076.

Brown, C., Garwood, M. P., & Williamson, J. E. (2012). It pays to cheat: tactical deception in a cephalopod social signalling system. Biology Letters, rsbl20120435. http://doi.org/10.1098/rsbl.2012.0435

Browndyke, J. N., Paskavitz, J., Sweet, L. H., & Cohen, R. A. (2008). Neuroanatomical correlates of malingered memory impairment: event-related fMRI of deception on a recognition memory task. Brain Injury.

Bruce, C. J., & Goldberg, M. E. (1985). Primate frontal eye fields. I. Single neurons discharging before saccades. Journal of Neurophysiology, 53(3), 603–635.

Brunel, N., & Wang, X.-J. (2001). Effects of Neuromodulation in a Cortical Network Model of Object Working Memory Dominated by Recurrent Inhibition. Journal of Computational Neuroscience, 11(1), 63–85. http://doi.org/10.1023/A:1011204814320

Brunton, B. W., Botvinick, M. M., & Brody, C. D. (2013). Rats and Humans Can Optimally Accumulate Evidence for Decision-Making. Science, 340(6128), 95–98. http://doi.org/10.1126/science.1233912

Bueti, D., & Walsh, V. (2009). The parietal cortex and the representation of time, space, number and other magnitudes. Philosophical Transactions of the Royal Society B: Biological Sciences, 364(1525), 1831–1840. http://doi.org/10.1098/rstb.2009.0028

Bugnyar, T., & Heinrich, B. (2006). Pilfering ravens, Corvus corax, adjust their behaviour to social context and identity of competitors. Animal Cognition, 9(4), 369-376.

Bugnyar, T., & Kotrschal, K. (2002). Observational learning and the raiding of food caches in ravens, Corvus corax: is it “tactical”deception? Animal Behaviour, 64(2), 185-196.

Bugnyar, T., & Kotrschal, K. (2004). Leading a conspecific away from food in ravens (Corvus corax)? Animal Cognition, 7(2), 69-76.

Butterworth, M. D. B. (1997). A Dissociation of Number Meanings. Cognitive Neuropsychology, 14(4), 613–636. doi:10.1080/026432997381501

Byrne, R. W., & Corp, N. (2004). Neocortex size predicts deception rate in primates. Proceedings of the Royal Society B: Biological Sciences, 271(1549), 1693–1699. http://doi.org/10.1098/rspb.2004.2780

271

Byrne, R. W., & Whiten, A. (1985). Tactical deception of familiar individuals in baboons (Papio ursinus). Animal Behaviour, 33(2), 669–673. http://doi.org/10.1016/S0003- 3472(85)80093-2

Byrne, R. W., & Whiten, A. (1990). Tactical deception in primates: the 1990 database. German Primate Center.

Cabeza, R., Ciaramelli, E., & Moscovitch, M. (2012). Cognitive contributions of the ventral parietal cortex: an integrative theoretical account. Trends in Cognitive Sciences, 16(6), 338–352. http://doi.org/10.1016/j.tics.2012.04.008

Call, J., & Tomasello, M. (2008). Does the chimpanzee have a theory of mind? 30 years later. Trends in Cognitive Sciences, 12(5), 187–192.

Cant, J. S., & Xu, Y. (2014). The Impact of Density and Ratio on Object-Ensemble Representation in Human Anterior-Medial Ventral Visual Cortex. Cerebral Cortex. http://doi.org/10.1093/cercor/bhu145

Cantlon, J. F., & Brannon, E. M. (2005). Semantic congruity affects numerical judgments similarly in monkeys and humans. Proceedings of the National Academy of Sciences of the United States of America, 102(45), 16507–16511. http://doi.org/10.1073/pnas.0506463102

Cantlon, J. F., & Brannon, E. M. (2006a). Shared system for ordering small and large numbers in monkeys and humans. Psychological Science, 17(5), 401.

Cantlon, J. F., & Brannon, E. M. (2006b). The effect of heterogeneity on numerical ordering in rhesus monkeys. Infancy, 9(2), 173–189.

Cantlon, J. F., & Brannon, E. M. (2007a). Basic math in monkeys and college students. PLoS Biology, 5(12), e328. http://doi.org/10.1371/journal.pbio.0050328

Cantlon, J. F., & Brannon, E. M. (2007b). How much does number matter to a monkey (Macaca mulatta)? Journal of Experimental Psychology: Animal Behavior Processes, 33(1), 32–41. http://doi.org/10.1037/0097-7403.33.1.32

Cantlon, J. F., Merritt, D. J., & Brannon, E. M. (2016). Monkeys display classic signatures of human symbolic arithmetic. Animal Cognition, 19(2), 405–415. http://doi.org/10.1007/s10071-015-0942-5

Cantlon, J. F., Platt, M. L., & Brannon, E. M. (2009). Beyond the number domain. Trends in Cognitive Sciences, 13(2), 83–91. http://doi.org/10.1016/j.tics.2008.11.007

272

Caraco, T., Martindale, S., & Whittam, T. S. (1980). An empirical demonstration of risk- sensitive foraging preferences. Animal Behaviour, 28(3), 820–830. http://doi.org/10.1016/S0003-3472(80)80142-4

Caro, T. (2014). Antipredator deception in terrestrial vertebrates. Current Zoology.

Castronovo, J., & Seron, X. (2007). Semantic numerical representation in blind subjects: The role of vision in the spatial format of the mental number line. The Quarterly Journal of Experimental Psychology, 60(1), 101–119. doi:10.1080/17470210600598635

Celebrini, S., & Newsome, W. T. (1994). Neuronal and psychophysical sensitivity to motion signals in extrastriate area MST of the macaque monkey. The Journal of Neuroscience, 14(7), 4109–4124.

Celebrini, S., & Newsome, W. T. (1995). Microstimulation of extrastriate area MST influences performance on a direction discrimination task. Journal of Neurophysiology, 73(2), 437–448.

Chang, S. W. C., Gariépy, J.-F., & Platt, M. L. (2013). Neuronal reference frames for social decisions in primate frontal cortex. Nature Publishing Group, 16(2), 243–250. http://doi.org/doi:10.1038/nn.3287

Chang, S. W., Winecoff, A. A., & Platt, M. L. (2011). Vicarious reinforcement in rhesus macaques (Macaca mulatta). Frontiers in Neuroscience.

Chang, S.W.C., Barter, J.W., Ebitz, R.B., Watson, K.K., and Platt, M.L. (2012). Inhaled oxytocin amplifies both vicarious reinforcement and self reinforcement in rhesus macaques (Macaca mulatta). Proc. Natl. Acad. Sci. 109, 959–964.

Chang, S.W.C., Brent, L.J.N., Adams, G.K., Klein, J.T., Pearson, J.M., Watson, K.K., and Platt, M.L. (2013). Neuroethology of primate social behavior. Proc. Natl. Acad. Sci. 110, 10387–10394.

Chen, S., Swartz, K. B., & Terrace, H. S. (1997). Knowledge of the ordinal position of list items in rhesus monkeys. Psychological Science, 8(2), 80–86.

Chen, Z. X., Xue, L., Liang, C. Y., Wang, L. L., Mei, W., Zhang, Q., & Zhao, H. (2015). Specific marker of feigned memory impairment: The activation of left superior frontal gyrus. Journal of Forensic and Legal Medicine, 36, 164–171.

Chervyakov, A. V., Chernyavsky, A. Y., Sinitsyn, D. O., & Piradov, M. A. (2015). Possible Mechanisms Underlying the Therapeutic Effects of Transcranial Magnetic

273

Stimulation. Frontiers in Human Neuroscience, 9(Pt 7), 615. http://doi.org/10.3389/fnhum.2015.00303

Cho, S. S., & Strafella, A. P. (2009). rTMS of the Left Dorsolateral Prefrontal Cortex Modulates Dopamine Release in the Ipsilateral Anterior Cingulate Cortex and Orbitofrontal Cortex. PLoS One, 4(8), e6725. http://doi.org/10.1371/journal.pone.0006725.t003

Christ, S. E., Van Essen, D. C., Watson, J. M., Brubaker, L. E., & McDermott, K. B. (2009). The Contributions of Prefrontal Cortex and Executive Control to Deception: Evidence from Activation Likelihood Estimate Meta-analyses. Cerebral Cortex, 19(7), 1557–1566. http://doi.org/10.1093/cercor/bhn189

Churchland, A. K., & Ditterich, J. (2012). New advances in understanding decisions among multiple alternatives. Current Opinion in Neurobiology, 22(6), 920–926. http://doi.org/10.1016/j.conb.2012.04.009

Churchland, A. K., Kiani, R., & Shadlen, M. N. (2008). Decision-making with multiple alternatives. Nature Neuroscience, 11(6), 693–702.

Churchland, A. K., Kiani, R., Chaudhuri, R., Wang, X.-J., Pouget, A., & Shadlen, M. N. (2011). Variance as a Signature of Neural Computations during Decision Making. Neuron, 69(4), 818–831. http://doi.org/10.1016/j.neuron.2010.12.037

Clithero, J. A., & Rangel, A. (2014). Informatic parcellation of the network involved in the computation of subjective value. Social Cognitive and Affective Neuroscience, 9(9), 1289–1302. http://doi.org/10.1093/scan/nst106

Connolly, K. R., Helmer, A., Cristancho, M. A., Cristancho, P., & O'Reardon, J. P. (2012). Effectiveness of Transcranial Magnetic Stimulation in Clinical Practice Post-FDA Approval in the United States. The Journal of Clinical Psychiatry, 73(04), e567–e573. http://doi.org/10.4088/JCP.11m07413

Constantinidis, C., & Goldman-Rakic, P. S. (2002). Correlated discharges among putative pyramidal neurons and interneurons in the primate prefrontal cortex. Journal of Neurophysiology, 88(6), 3487–3497. http://doi.org/10.1152/jn.00188.2002

Constantinidis, C., Bucci, D. J., & Rugg, M. D. (2013). Cognitive functions of the posterior parietal cortex. Frontiers in Integrative Neuroscience, 7. http://doi.org/10.3389/fnint.2013.00035

Conway, B. R. (2009). Color Vision, Cones, and Color-Coding in the Cortex. The

274

Neuroscientist, 15(3), 274–290. http://doi.org/10.1177/1073858408331369

Cook, R. G., & Wasserman, E. A. (2007). Learning and transfer of relational matching-to- sample by pigeons. Psychonomic Bulletin & Review, 14(6), 1107–1114.

Coricelli, G., & Nagel, R. (2009). Neural correlates of depth of strategic reasoning in medial prefrontal cortex. Proceedings of the National Academy of Sciences of the United States of America, 106(23), 9163–9168. http://doi.org/10.1073/pnas.0807721106

Coussi-Korbel, S. (1994). Learning to outwit a competitor in mangabeys (Cercocebus torquatus torquatus). Journal of Comparative Psychology, 108(2), 164-171.

Cui, Q., Vanman, E. J., Wei, D., Yang, W., Jia, L., & Zhang, Q. (2014). Detection of deception based on fMRI activation patterns underlying the production of a deceptive response and receiving feedback about the success of the deception after a mock murder crime. Social Cognitive and …, 9(10), 1472–1480.

Dacke, M. (2008). Evidence for counting in insects. Animal Cognition, 11(4), 683-689.

Dally, J. M., Clayton, N. S., & Emery, N. J. (2006a). The behaviour and evolution of cache protection and pilferage. Animal Behaviour.

Dally, J. M., Emery, N. J., & Clayton, N. S. (2005). Cache protection strategies by western scrub-jays, Aphelocoma californica: implications for social cognition. Animal Behaviour, 70(6), 1251–1263. http://doi.org/10.1016/j.anbehav.2005.02.009

Dally, J. M., Emery, N. J., & Clayton, N. S. (2006b). Food-Caching Western Scrub-Jays Keep Track of Who Was Watching When. Science, 312(5780), 1662–1665. http://doi.org/10.1126/science.1126539

Davatzikos, C., Ruparel, K., Fan, Y., Shen, D. G., Acharyya, M., Loughead, J. W., et al. (2005). Classifying spatial patterns of brain activity with machine learning methods: Application to lie detection. NeuroImage, 28(3), 663–668. http://doi.org/10.1016/j.neuroimage.2005.08.009

Davis, H., & Bradford, S. A. (1986). Counting behavior by rats in a simulated natural environment. Ethology, 73(4), 265–280.

Daw, N. D., Gershman, S. J., Seymour, B., Dayan, P., & Dolan, R. J. (2011). Model-Based Influences on Humans' Choices and Striatal Prediction Errors. Neuron, 69(6), 1204– 1215. http://doi.org/10.1016/j.neuron.2011.02.027

275

Daw, N. D., Niv, Y., & Dayan, P. (2005). Uncertainty-based competition between prefrontal and dorsolateral striatal systems for behavioral control. Nature Neuroscience, 8(12), 1704–1711. http://doi.org/10.1038/nn1560 de Hevia, M. D., & Spelke, E. S. (2010). Number-Space Mapping in Human Infants. Psychological Science, 21(5), 653–660. doi:10.1177/0956797610366091 de Lafuente, V., Jazayeri, M., & Shadlen, M. N. (2015). Representation of Accumulating Evidence for a Decision in Two Parietal Areas. Journal of Neuroscience, 35(10), 4306– 4318. http://doi.org/10.1523/JNEUROSCI.2451-14.2015

Dehaene, S. (1999). The Number Sense: How the Mind Creates Mathematics (1st ed.). Oxford University Press.

Dehaene, S. (2011). The Number Sense: How the Mind Creates Mathematics, Revised and Updated Edition (2nd ed.). Oxford University Press.

Dehaene, S., & Cohen, L. (1997). Cerebral Pathways for Calculation: Double Dissociation between Rote Verbal and Quantitative Knowledge of Arithmetic. Cortex, 33(2), 219– 250. http://doi.org/10.1016/S0010-9452(08)70002-9

Dehaene, S., Bossini, S., & Giraux, P. (1993). The mental representation of parity and number magnitude. Journal of Experimental Psychology General, 122(3), 371.

Dehaene, S., Dehaene-Lambertz, G., & Cohen, L. (1998). Abstract representations of numbers in the animal and human brain. Trends in Neurosciences, 21(8), 355–361.

Demirtas-Tatlidede, A., Vahabzadeh-Hagh, A. M., & Pascual-Leone, A. (2013). Can noninvasive brain stimulation enhance cognition in neuropsychiatric disorders? Neuropharmacology, 64, 566–578. http://doi.org/10.1016/j.neuropharm.2012.06.020

Deng, Z.-D., Lisanby, S. H., & Peterchev, A. V. (2013). Electric field depth–focality tradeoff in transcranial magnetic stimulation: Simulation comparison of 50 coil designs. Brain Stimulation, 6(1), 1–13. http://doi.org/10.1016/j.brs.2012.02.005

Denison, S., & Xu, F. (2010). Twelve- to 14-month-old infants can predict single-event probability with large set sizes. Developmental Science, 13(5), 798–803. http://doi.org/10.1111/j.1467-7687.2009.00943.x

Denison, S., & Xu, F. (2014). The origins of probabilistic inference in human infants. Cognition, 130(3), 335–347.

276

Desimone, R., & Ungerleider, L. G. (1986). Multiple visual areas in the caudal superior temporal sulcus of the macaque. The Journal of Comparative Neurology, 248(2), 164– 189. http://doi.org/10.1002/cne.902480203

Desimone, R., Albright, T. D., Gross, C. G., & Bruce, C. (1984). Stimulus-selective properties of inferior temporal neurons in the macaque. The Journal of Neuroscience, 4(8), 2051–2062.

DeWind, N. K., & Brannon, E. M. (2012). Malleability of the approximate number system: effects of feedback and training. Frontiers in Human Neuroscience, 6, 68. http://doi.org/10.3389/fnhum.2012.00068

DeWind, N. K., Adams, G. K., Platt, M. L., & Brannon, E. M. (2015). Modeling the approximate number system to quantify the contribution of visual stimulus features. Cognition, 142, 247–265. http://doi.org/10.1016/j.cognition.2015.05.016

Di Lazzaro, V., Dileone, M., Pilato, F., Capone, F., Musumeci, G., Ranieri, F., et al. (2011). Modulation of motor cortex neuronal networks by rTMS: comparison of local and remote effects of six different protocols of stimulation. Journal of Neurophysiology, 105(5), 2150–2156. http://doi.org/10.1152/jn.00781.2010

Dias, E. C., Kiesau, M., & Segraves, M. A. (1995). Acute activation and inactivation of macaque frontal eye field with GABA-related drugs. Journal of Neurophysiology, 74(6), 2744–2748. http://doi.org/10.1152/jn.00865.2015

Dike, C. C., Baranoski, M., & Griffith, E. E. H. (2005). Pathological lying revisited. Journal of the American ….

Ding, L., & Gold, J. I. (2012). Neural Correlates of Perceptual Decision Making before, during, and after Decision Commitment in Monkey Frontal Eye Field. Cerebral Cortex, 22(5), 1052–1067. http://doi.org/10.1093/cercor/bhr178

Ding, X. P., Sai, L., Fu, G., Liu, J., & Lee, K. (2014). Neural correlates of second-order verbal deception: A functional near-infrared spectroscopy (fNIRS) study. NeuroImage.

Ditterich, J., Mazurek, M. E., & Shadlen, M. N. (2003). Microstimulation of visual cortex affects the speed of perceptual decisions. Nature Neuroscience, 6(8), 891–898. http://doi.org/10.1038/nn1094

Dixon, M. L., & Christoff, K. (2014). The lateral prefrontal cortex and complex value- based learning and decision making. Neuroscience & Biobehavioral Reviews, 45, 9–18.

277

http://doi.org/10.1016/j.neubiorev.2014.04.011

Doll, B. B., Bath, K. G., Daw, N. D., & Frank, M. J. (2016). Variability in Dopamine Genes Dissociates Model-Based and Model-Free Reinforcement Learning. Journal of Neuroscience, 36(4), 1211–1222. http://doi.org/10.1523/JNEUROSCI.1901-15.2016

Dormal, V., Andres, M., & Pesenti, M. (2008). Dissociation of numerosity and duration processing in the left intraparietal sulcus: A transcranial magnetic stimulation study. Cortex, 44(4), 462–469. http://doi.org/10.1016/j.cortex.2007.08.011

Dormal, V., Andres, M., & Pesenti, M. (2012). Contribution of the right intraparietal sulcus to numerosity and length processing: An fMRI-guided TMS study. Cortex, 48(5), 623–629. http://doi.org/10.1016/j.cortex.2011.05.019

Drayton, L. A., & Santos, L. R. (2014). A decade of theory of mind research on cayo santiago: Insights into rhesus macaque social cognition. American Journal of Primatology, 78(1), 106–116. http://doi.org/10.1002/ajp.22362

Drucker, C. B., & Brannon, E. M. (2014). Rhesus monkeys (Macaca mulatta) map number onto space. Cognition, 132(1), 57–67. http://doi.org/10.1016/j.cognition.2014.03.011

Drucker, C. B., & Brannon, E. M. (2015). Commentary on: "Number-space mapping in the newborn chick resembles humans' mental number line". Frontiers in Psychology, 6, 352. http://doi.org/10.3389/fpsyg.2015.00352

Drucker, C. B., Carlson, M. L., Toda, K., DeWind, N. K., & Platt, M. L. (2015a). Non- invasive primate head restraint using thermoplastic masks. Journal of Neuroscience Methods, 253, 90–100. http://doi.org/10.1016/j.jneumeth.2015.06.013

Drucker, C. B., Rossa, M. A., & Brannon, E. M. (2015b). Comparison of discrete ratios by rhesus macaques (Macaca mulatta). Animal Cognition, 19(1), 75–89. http://doi.org/10.1007/s10071-015-0914-9

Drugowitsch, J., & Pouget, A. (2012). Probabilistic vs. non-probabilistic approaches to the neurobiology of perceptual decision-making. Current Opinion in Neurobiology, 22(6), 963–969. http://doi.org/10.1016/j.conb.2012.07.007

Drugowitsch, J., Moreno-Bote, R., Churchland, A. K., Shadlen, M. N., & Pouget, A. (2012). The cost of accumulating evidence in perceptual decision making. Journal of Neuroscience, 32(11), 3612–3628.

Du, E., & Chang, S. W. C. (2015). Neural components of altruistic punishment. Frontiers

278

in Neuroscience, 9. http://doi.org/10.3389/fnins.2015.00026

Ducoing, A. M., & Thierry, B. (2003). Withholding information in semifree-ranging Tonkean macaques (Macaca tonkeana). Journal of Comparative Psychology.

Edeline, J.-M., Hars, B., Hennevin, E., & Cotillon, N. (2002). Muscimol Diffusion after Intracerebral Microinjections: A Reevaluation Based on Electrophysiological and Autoradiographic Quantifications. Neurobiology of Learning and Memory, 78(1), 100– 124. http://doi.org/10.1006/nlme.2001.4035

Eldaief, M. C., Press, D. Z., & Pascual-Leone, A. (2013). Transcranial magnetic stimulation in neurology: A review of established and prospective applications. Neurology Clinical Practice, 3(6), 519–526. http://doi.org/10.1212/01.CPJ.0000436213.11132.8e

Emery, N. J., Dally, J. M., & Clayton, N. S. (2004). Western scrub-jays ( Aphelocoma californica ) use cognitive strategies to protect their caches from thieving conspecifics. Animal Cognition, 7(1), 37–43. http://doi.org/10.1007/s10071-003-0178-7

Emmerton, J. (2001). Pigeons' discrimination of color proportion in computer-generated visual displays. Animal Learning & Behavior, 29(1), 21–35.

Epstein, R., & Kanwisher, N. (1998). A cortical representation of the local visual environment. Nature, 392(6676), 598–601. http://doi.org/10.1038/33402

Erlich, J. C., Brunton, B. W., Duan, C. A., Hanks, T. D., Brody, C. D., & Carandini, M. (2015). Distinct effects of prefrontal and parietal cortex inactivations on an accumulation of evidence task in the rat. eLife, 4, e05457. http://doi.org/10.7554/eLife.05457

Etiévant, A., Manta, S., Latapy, C., Magno, L. A. V., Fecteau, S., & Beaulieu, J.-M. (2015). Repetitive transcranial magnetic stimulation induces long-lasting changes in protein expression and histone acetylation. Scientific Reports, 5, 16873. http://doi.org/10.1038/srep16873

Evarts, E.V. (1968). Relation of pyramidal tract activity to force exerted during voluntary movement. J. Neurophysiol. 31, 14–27.

Fabre-Thorpe, M., Richard, G., & Thorpe, S. J. (1998). Rapid categorization of natural images by rhesus monkeys. Neuroreport, 9(2), 303–308.

Fagot, J., & Maugard, A. (2013). Analogical reasoning in baboons (Papio papio): Flexible

279

reencoding of the source relation depending on the target relation. Learning & Behavior, 41(3), 229–237. http://doi.org/10.3758/s13420-012-0101-7

Fazio, L. K., Bailey, D. H., Thompson, C. A., & Siegler, R. S. (2014). Relations of different types of numerical magnitude representations to each other and to mathematics achievement. Journal of Experimental Child Psychology, 123, 53–72. http://doi.org/10.1016/j.jecp.2014.01.013

Fecteau, S., Boggio, P., Fregni, F., & Pascual-Leone, A. (2013). Modulation of untruthful responses with non-invasive brain stimulation. Frontiers in Psychiatry, 3, 97. http://doi.org/10.3389/fpsyt.2012.00097

Feigenson, L., Dehaene, S., & Spelke, E. (2004). Core systems of number. Trends in Cognitive Sciences, 8(7), 307–314.

Felleman, D. J., & Van Essen, D. C. (1991). Distributed Hierarchical Processing in the Primate Cerebral Cortex. Cerebral Cortex, 1(1), 1–47. http://doi.org/10.1093/cercor/1.1.1

Fetsch, C. R., Kiani, R., Newsome, W. T., & Shadlen, M. N. (2014). Effects of cortical microstimulation on confidence in a perceptual decision. Neuron, 83, 797–804.

Fias, W., & Fischer, M. H. (2005). Spatial representation of number. In J. Campbell (Ed.), Handbook of mathematical cognition (pp. 43–54). Pschology Press.

Filimon, F., Philiastides, M. G., Nelson, J. D., Kloosterman, N. A., & Heekeren, H. R. (2013). How Embodied Is Perceptual Decision Making? Evidence for Separate Processing of Perceptual and Motor Decisions. Journal of Neuroscience, 33(5), 2121– 2136. http://doi.org/10.1523/JNEUROSCI.2334-12.2013

Fischer, M.H., & Brugger, P. (2010). When digits help digits: Spatial-numerical associations point to finger counting as prime example of embodied cognition. Frontiers in Psychology, 2, article 260, 1-7.

Fitch, W.T., & Braccini, S.N. (2013). Primate laterality and the biology and evolution of human handedness: a review and synthesis. Annals of the New York Academy of Sciences, 1288, 70-85.

Fitzgerald, J. K., Swaminathan, S. K., & Freedman, D. J. (2012). Visual categorization and the parietal cortex. Frontiers in Integrative Neuroscience, 1–6. http://doi.org/10.3389/fnint.2012.00018/abstract

280

Fitzgerald, P. B., Fountain, S., & Daskalakis, Z. J. (2006). A comprehensive review of the effects of rTMS on motor cortical excitability and inhibition. Clinical Neurophysiology, 117(12), 2584–2596. http://doi.org/10.1016/j.clinph.2006.06.712

Flemming, T. M., Beran, M. J., Thompson, R. K. R., Kleider, H. M., & Washburn, D. A. (2008). What meaning means for same and different: Analogical reasoning in humans (Homo sapiens), chimpanzees (Pan troglodytes), and rhesus monkeys (Macaca mulatta). Journal of Comparative Psychology, 122(2), 176–185. http://doi.org/10.1037/0735-7036.122.2.176

Flemming, T. M., Thompson, R. K. R., Beran, M. J., & Washburn, D. A. (2011). Analogical reasoning and the differential outcome effect: Transitory bridging of the conceptual gap for rhesus monkeys (Macaca mulatta). Journal of Experimental Psychology: Animal Behavior Processes, 37(3), 353–360. http://doi.org/10.1037/a0022142

Foeller, P., and Tychsen, L. (2002). Eye movement training and recording in alert macaque monkeys: 1. Operant visual conditioning; 2. Magnetic search coil and head restraint surgical implantation; 3. Calibration and recording. Strabismus 10, 5–22.

Fontanari, L., Gonzalez, M., Vallortigara, G., & Girotto, V. (2014). Probabilistic cognition in two indigenous Mayan groups. Proceedings of the National Academy of Sciences, 111(48), 17075–17080. http://doi.org/10.1073/pnas.1410583111

Franklin, M. S., Jonides, J., & Smith, E. E. (2009). Processing of order information for numbers and months. Memory & Cognition, 37(5), 644–654. doi:10.3758/MC.37.5.644

Freedman, D. J., & Assad, J. A. (2006). Experience-dependent representation of visual categories in parietal cortex. Nature, 443(7107), 85–88. http://doi.org/10.1038/nature05078

Freedman, D. J., & Assad, J. A. (2009). Distinct Encoding of Spatial and Nonspatial Visual Information in Parietal Cortex. Journal of Neuroscience, 29(17), 5671–5680. http://doi.org/10.1523/JNEUROSCI.2878-08.2009

Freedman, D. J., & Assad, J. A. (2011). A proposed common neural mechanism for categorization and perceptual decisions. Nature Neuroscience, 14(2), 143–146.

Freedman, D. J., & Miller, E. K. (2008). Neural mechanisms of visual categorization: Insights from neurophysiology. Neuroscience & Biobehavioral Reviews, 32(2), 311–329. http://doi.org/10.1016/j.neubiorev.2007.07.011

Freedman, D. J., & Riesenhuber, M. (2002). Visual categorization and the primate

281

prefrontal cortex: neurophysiology and behavior. Journal of Neurophysiology, 88, 929– 941.

Freedman, D. J., Riesenhuber, M., Poggio, T., & Miller, E. K. (2001). Categorical representation of visual stimuli in the primate prefrontal cortex. Science, 291(5502), 312.

Freedman, D. J., Riesenhuber, M., Poggio, T., & Miller, E. K. (2003). A comparison of primate prefrontal and inferior temporal cortices during visual categorization. The Journal of Neuroscience, 23(12), 5235–5246.

Friendlich, A.R. (1973). Primate head restrainer using a nonsurgical technique. J Appl Physiol 35, 934–935.

Frith, C. D., & Frith, U. (2006). The Neural Basis of Mentalizing. Neuron, 50(4), 531–534. http://doi.org/10.1016/j.neuron.2006.05.001

Fullam, R. S., McKie, S., & Dolan, M. C. (2009). Psychopathic traits and deception: functional magnetic resonance imaging study. The British Journal of Psychiatry, 194(3), 229–235. http://doi.org/10.1192/bjp.bp.108.053199

Galletti, C., Fattori, P., Kutz, D. F., & Battaglini, P. P. (1997). Arm Movement-related Neurons in the Visual Area V6A of the Macaque Superior Parietal Lobule. European Journal of Neuroscience, 9(2), 410–413. http://doi.org/10.1111/j.1460- 9568.1997.tb01410.x

Gallistel, C., & Gelman, R. (2000). Non-verbal numerical cognition: From reals to integers. Trends in Cognitive Sciences.

Ganis, G. (2003). Neural Correlates of Different Types of Deception: An fMRI Investigation. Cerebral Cortex, 13(8), 830–836. http://doi.org/10.1093/cercor/13.8.830

Gardner, M., & Steinberg, L. (2005). Peer influence on risk taking, risk preference, and risky decision making in adolescence and adulthood: an experimental study. Developmental Psychology.

Gee, J., & Button, M. (2015, January). The Financial Cost of Fraud 2015. Retrieved February 23, 2016, from http://www.pkf.com/media/31640/PKF-The-financial-cost- of-fraud-2015.pdf

Gerits, A., Ruff, C.C., Guipponi, O., Wenderoth, N., Driver, J., and Vanduffel, W. (2011). Transcranial magnetic stimulation of macaque frontal eye fields decreases saccadic

282

reaction time. Exp. Brain Res. 212, 143–152.

Gersner, R., Kravetz, E., Feil, J., Pell, G., & Zangen, A. (2011). Long-term effects of repetitive transcranial magnetic stimulation on markers for neuroplasticity: differential outcomes in anesthetized and awake animals. The Journal of Neuroscience : the Official Journal of the Society for Neuroscience, 31(20), 7521–7526. http://doi.org/10.1523/JNEUROSCI.6751-10.2011

Gevers, W., Santens, S., Dhooge, E., Chen, Q., Van den Bossche, L., Fias, W., & Verguts, T. (2010). Verbal-spatial and visuospatial coding of number-space interactions. Journal of Experimental Psychology: General, 139(1), 180-190.

Gilmore, C. K., McCarthy, S. E., & Spelke, E. S. (2010). Non-symbolic arithmetic abilities and mathematics achievement in the first year of formal schooling. Cognition, 115(3), 394–406. http://doi.org/10.1016/j.cognition.2010.02.002

Gleichgerrcht, E., Ibáñez, A., Roca, M., Torralva, T., & Manes, F. (2010). Decision-making cognition in neurodegenerative diseases. - PubMed - NCBI. Nature Reviews Neurology, 6(11), 611–623. http://doi.org/10.1038/nrneurol.2010.148

Gobel, S. M., Shaki, S., & Fischer, M. H. (2011). The Cultural Number Line: A Review of Cultural and Linguistic Influences on the Development of Number Processing. Journal of Cross-Cultural Psychology, 42(4), 543–565. doi:10.1177/0022022111406251

Goetz, S. M., Luber, B., Lisanby, S. H., Murphy, D. L. K., Kozyrkov, I. C., Grill, W. M., & Peterchev, A. V. (2016). Enhancement of Neuromodulation with Novel Pulse Shapes Generated by Controllable Pulse Parameter Transcranial Magnetic Stimulation. Brain Stimulation, 9(1), 39–47. http://doi.org/10.1016/j.brs.2015.08.013

Gold, J. I., & Shadlen, M. N. (2000). Representation of a perceptual decision in developing oculomotor commands. Nature, 404(6776), 390–394. http://doi.org/10.1038/35006062

Gold, J. I., & Shadlen, M. N. (2007). The Neural Basis of Decision Making. Annual Review of Neuroscience, 30(1), 535–574. http://doi.org/10.1146/annurev.neuro.29.051605.113038

Gómez-Laplaza, L. M., & Gerlai, R. (2012). Quantification abilities in angelfish (Pterophyllum scalare): the influence of continuous variables. Animal Cognition. doi:10.1007/s10071-012-0578-7

Goulas, A., Bastiani, M., Bezgin, G., Uylings, H.B.M., Roebroeck, A., and Stiers, P. (2014).

283

Comparative Analysis of the Macroscale Structural Connectivity in the Macaque and Human Brain. PLoS Comput. Biol. 10, e1003529.

Gould, S. J., & Vrba, E. S. (1982, February). Exaptation - a missing term in the science of form. Retrieved February 23, 2016, from http://www.jstor.org/stable/2400563?seq=1#page_scan_tab_contents

Grant, S. (2000). Drug abusers show impaired performance in a laboratory test of decision making. Neuropsychologia, 38(8), 1180–1187. http://doi.org/10.1016/S0028- 3932(99)00158-X

Greene, J. D., & Paxton, J. M. (2009). Patterns of neural activity associated with honest and dishonest moral decisions. Proceedings of the National Academy of Sciences, 106(30), 12506–12511. http://doi.org/10.1073/pnas.0900152106

Grefkes, C., and Fink, G.R. (2005). REVIEW: The functional organization of the intraparietal sulcus in humans and monkeys. J. Anat. 207, 3–17.

Grehl, S., Viola, H. M., Fuller-Carter, P. I., Carter, K. W., Dunlop, S. A., Hool, L. C., et al. (2015). Cellular and Molecular Changes to Cortical Neurons Following Low Intensity Repetitive Magnetic Stimulation at Different Frequencies. Brain Stimulation, 8(1), 114–123. http://doi.org/10.1016/j.brs.2014.09.012

Gu, C., and Corneil, B.D. (2014). Transcranial Magnetic Stimulation of the Prefrontal Cortex in Awake Nonhuman Primates Evokes a Polysynaptic Neck Muscle Response That Reflects Oculomotor Activity at the Time of Stimulation. J. Neurosci. 34, 14803–14815.

Hadar, A. A., Makris, S., & Yarrow, K. (2012). The truth-telling motor cortex: Response competition in M1 discloses deceptive behaviour. Biological Psychology, 89, 495–502.

Haegens, S., Nacher, V., Hernández, A., Luna, R., Jensen, O., & Romo, R. (2011a). Beta oscillations in the monkey sensorimotor network reflect somatosensory decision making. Proceedings of the National Academy of Sciences of the United States of America, 108(26), 10708–10713. http://doi.org/10.2307/27978678

Haegens, S., Nacher, V., Luna, R., Romo, R., & Jensen, O. (2011b). α-Oscillations in the monkey sensorimotor network influence discrimination performance by rhythmical inhibition of neuronal spiking. Proceedings of the National Academy of Sciences of the United States of America, 108(48), 19377–19382. http://doi.org/10.2307/23066771

Haegens, S., Vazquez, Y., Zainos, A., Alvarez, M., Jensen, O., & Romo, R. (2014).

284

Thalamocortical rhythms during a vibrotactile detection task. Proceedings of the National Academy of Sciences, 111(17), E1797–E1805. http://doi.org/10.1073/pnas.1405516111

Hakun, J. G., Seelig, D., Ruparel, K., Loughead, J. W., Busch, E., Gur, R. C., & Langleben, D. D. (2008). fMRI investigation of the cognitive structure of the Concealed Information Test. Neurocase, 14(1), 59–67. http://doi.org/10.1080/13554790801992792

Halberda, J., Mazzocco, M., & Feigenson, L. (2008). Individual differences in non-verbal number acuity correlate with maths achievement. Nature, 455(7213), 665–668.

Hanks, T. D., Ditterich, J., & Shadlen, M. N. (2006). Microstimulation of macaque area LIP affects decision-making in a motion discrimination task. Nature Neuroscience, 9(5), 682–689. http://doi.org/10.1038/nn1683

Hanks, T. D., Kopec, C. D., Brunton, B. W., Duan, C. A., Erlich, J. C., & Brody, C. D. (2015). Distinct relationships of parietal and prefrontal cortices to evidence accumulation. Nature, 520(7546), 220–223. http://doi.org/10.1038/nature14066

Hanks, T., Kiani, R., & Shadlen, M. N. (2014). A neural mechanism of speed-accuracy tradeoff in macaque area LIP. eLife. http://doi.org/10.7554/eLife.02260

Hanlon, C. A., Dowdle, L. T., Austelle, C. W., DeVries, W., Mithoefer, O., Badran, B. W., & George, M. S. (2015). What goes up, can come down: Novel brain stimulation paradigms may attenuate craving and craving-related neural circuitry in substance dependent individuals. Brain Research, 1628, 199–209. http://doi.org/10.1016/j.brainres.2015.02.053

Hanus, D., & Call, J. (2007). Discrete quantity judgments in the great apes (Pan paniscus, Pan troglodytes, Gorilla gorilla, Pongo pygmaeus): The effect of presenting whole sets versus item-by-item. Journal of Comparative Psychology, 121(3), 241–249. doi:10.1037/0735-7036.121.3.241

Haroush, K., & Williams, Z. M. (2015). Neuronal prediction of opponent's behavior during cooperative social interchange in primates. Cell, 160, 1–13.

Harper, D. G. C. (1982). Competitive foraging in mallards: ``Ideal free' ducks. Animal Behaviour, 30(2), 575–584. http://doi.org/10.1016/S0003-3472(82)80071-7

Hartnett, P., & Gelman, R. (1998). Early understandings of numbers: paths or barriers to the construction of new understandings? Learning and Instruction, 8(4), 341–374. http://doi.org/10.1016/S0959-4752(97)00026-1

285

Harvey, B. M., Klein, B. P., Petridou, N., & Dumoulin, S. O. (2013). Topographic Representation of Numerosity in the Human Parietal Cortex. Science, 341(6150), 1123–1126. http://doi.org/10.1126/science.1239052

Hayden, B. Y., Heilbronner, S. R., & Platt, M. L. (2010). Ambiguity aversion in rhesus macaques. Frontiers in Neuroscience, 4. http://doi.org/10.3389/fnins.2010.00166

Heekeren, H. R., Marrett, S., Bandettini, P. A., & Ungerleider, L. G. (2004). A general mechanism for perceptual decision-making in the human brain. Nature, 431(7010), 859–862. http://doi.org/10.1038/nature02966

Heekeren, H. R., Marrett, S., Ruff, D. A., Bandettini, P. A., & Ungerleider, L. G. (2006). Involvement of human left dorsolateral prefrontal cortex in perceptual decision making is independent of response modality. Proceedings of the National Academy of Sciences, 103(26), 10023–10028. http://doi.org/10.1073/pnas.0603949103

Hernandez, A., Salinas, E., Garcı́a, R., & Romo, R. (1997). Discrimination in the sense of flutter: new psychophysical measurements in monkeys. The Journal of Neuroscience, 17(16), 6391–6400.

Hernandez, A., Zainos, A., & Romo, R. (2000). Neuronal correlates of sensory discrimination in the somatosensory cortex. Proceedings of the National Academy of Sciences, 97(11), 6191–6196. http://doi.org/10.1073/pnas.120018597

Hernandez, A., Zainos, A., & Romo, R. (2002). Temporal evolution of a decision-making process in medial premotor cortex. Neuron, 33, 959–972.

Hirata, S., & Matsuzawa, T. (2001). Tactics to obtain a hidden food item in chimpanzee pairs (Pan troglodytes). Animal Cognition, 4(3-4), 285–295. http://doi.org/10.1007/s100710100096

Ho, T. C., Brown, S., & Serences, J. T. (2009). Domain General Mechanisms of Perceptual Decision Making in Human Cortex. Journal of Neuroscience, 29(27), 8675–8687. http://doi.org/10.1523/JNEUROSCI.5984-08.2009

Holmes, K. J., & Lourenco, S. F. (2011). Common spatial organization of number and emotional expression: A mental magnitude line. Brain and Cognition, 77(2), 315–323. http://doi.org/10.1016/j.bandc.2011.07.002

Honig, W. K., & Stewart, K. E. (1989). Discrimination of relative numerosity by pigeons. Animal Learning & Behavior, 17(2), 134–146.

286

Honig, W. K., & Stewart, K. E. (1993). Relative numerosity as a dimension of stimulus control: The peak shift. Animal Learning & Behavior, 21(4), 346–354.

Hoogendam, J. M., Ramakers, G. M. J., & Di Lazzaro, V. (2010). Physiology of repetitive transcranial magnetic stimulation of the human brain. Brain Stimulation, 3(2), 95–118. http://doi.org/10.1016/j.brs.2009.10.005

Hopkins, W.D. (2013). Comparing human and nonhuman primate handedness: Challenges and a modest proposal for consensus. Developmental Psychobiology, 55(6), 621-636.

Hopkins, W.D., Stoinski, T.S., Lukas, K.E., Ross, S.R., & Wesley, M.J. (2003). Comparative assessment of handedness for a coordinated bimanual task in chimpanzees (Pan troglodytes), gorillas (Gorilla gorilla), and orangutans (Pongo pygmaeus). Journal of Comparative Psychology, 117(3), 302-308.

Horschig, J. M., Zumer, J., & Bahramisharif, A. (2014). Hypothesis-driven methods to augment human cognition by optimizing cortical oscillations. 1662-5137, 8, 268. http://doi.org/10.3389/fnsys.2014.00119

Horwitz, G. D., & Newsome, W. T. (1999). Separate Signals for Target Selection and Movement Specification in the Superior Colliculus. Science, 284(5417), 1158–1161. http://doi.org/10.1126/science.284.5417.1158

Horwitz, G. D., & Newsome, W. T. (2001). Target Selection for Saccadic Eye Movements: Prelude Activity in the Superior Colliculus During a Direction-Discrimination Task. Journal of Neurophysiology, 86(5), 2543–2558.

Hosokawa, T., & Watanabe, M. (2012). Prefrontal neurons represent winning and losing during competitive video shooting games between monkeys. The Journal of Neuroscience : the Official Journal of the Society for Neuroscience, 32(22), 7662–7671. http://doi.org/10.1523/JNEUROSCI.6479-11.2012

Hot, P., Ramdeen, K. T., Borg, C., Bollon, T., & Couturier, P. (2013). Impaired Decision Making in Alzheimer’s Disease A Deficit of Cognitive Strategy Selection? Clinical Psychological Science, 2(3). http://doi.org/10.1177/2167702613504094

Hovington, C. L., McGirr, A., Lepage, M., & Berlim, M. T. (2013). Repetitive transcranial magnetic stimulation (rTMS) for treating major depression and schizophrenia: a systematic review of recent meta-analyses. Annals of Medicine, 45(4), 308–321. http://doi.org/10.3109/07853890.2013.783993

287

Hoy, K. E., Bailey, N., Michael, M., Fitzgibbon, B., Rogasch, N. C., Saeki, T., & Fitzgerald, P. B. (2015). Enhancement of Working Memory and Task-Related Oscillatory Activity Following Intermittent Theta Burst Stimulation in Healthy Controls. Cerebral Cortex (New York, N.Y. : 1991), bhv193. http://doi.org/10.1093/cercor/bhv193

Huang, Y.-Z., Chen, R.-S., Rothwell, J. C., & Wen, H.-Y. (2007). The after-effect of human theta burst stimulation is NMDA receptor dependent. Clinical Neurophysiology, 118(5), 1028–1032. http://doi.org/10.1016/j.clinph.2007.01.021

Hubbard, E. M., Piazza, M., Pinel, P., & Dehaene, S. (2005). Interactions between number and space in parietal cortex. Nature Reviews Neuroscience, 6(6), 435–448.

Huk, A. C., & Shadlen, M. N. (2005). Neural activity in macaque parietal cortex reflects temporal integration of visual motion signals during perceptual decision making. The Journal of Neuroscience, 25(45), 10420–10436.

Hung, C. P., Kreiman, G., Poggio, T., & DiCarlo, J. J. (2005). Fast Readout of Object Identity from Macaque Inferior Temporal Cortex. Science, 310(5749), 863–866. http://doi.org/10.2307/3842768

Hutchison, R.M., Gallivan, J.P., Culham, J.C., Gati, J.S., Menon, R.S., and Everling, S. (2012). Functional connectivity of the frontal eye fields in humans and macaque monkeys investigated with resting-state fMRI. J. Neurophysiol. 107, 2463–2474.

Hwang, E. J., Hauschild, M., Wilke, M., & Andersen, R. A. (2012). Inactivation of the parietal reach region causes optic ataxia, impairing reaches but not saccades. Neuron, 76, 1021–1029.

Hyvarinen, J., Sakata, H., Talbot, W. H., & Mountcastle, V. B. (1968). Neuronal Coding by Cortical Cells of the Frequency of Oscillating Peripheral Stimuli. Science, 162(3858), 1130–1132. http://doi.org/10.1126/science.162.3858.1130

Iba, M., & Sawaguchi, T. (2003). Involvement of the Dorsolateral Prefrontal Cortex of Monkeys in Visuospatial Target Selection. Journal of Neurophysiology, 89(1), 587–599. http://doi.org/10.1152/jn.00148.2002

Ischebeck, A., Schocke, M., & Delazer, M. (2009). The processing and representation of fractions within the brain. NeuroImage, 47(1), 403–413.

Ishai, A., Ungerleider, L. G., Martin, A., Schouten, H. L., & Haxby, J. V. (1999). Distributed representation of objects in the human ventral visual pathway. Proceedings of the National Academy of Sciences of the United States of America, 96(16),

288

9379–9384.

Isoda, M., Tsutsui, K.-I., Katsuyama, N., Naganuma, T., Saito, N., Furusawa, Y., Mushiake, H., Taira, M., and Tanji, J. (2005). Design of a head fixation device for experiments in behaving monkeys. J. Neurosci. Methods 141, 277–282.

Ito, M., & Doya, K. (2011). Multiple representations and algorithms for reinforcement learning in the cortico-basal ganglia circuit. Current Opinion in Neurobiology.

Jaakkola, K., Fellner, W., Erb, L., Rodriguez, M., & Guarino, E. (2005). Understanding of the concept of numerically “less” by bottlenose dolphins (Tursiops truncatus). Journal of Comparative Psychology, 119(3), 296–303. doi:10.1037/0735-7036.119.3.296

Jacob, S. N., & Nieder, A. (2009a). Notation-Independent Representation of Fractions in the Human Parietal Cortex. Journal of Neuroscience, 29(14), 4652–4657. http://doi.org/10.1523/JNEUROSCI.0651-09.2009

Jacob, S. N., & Nieder, A. (2009b). Tuning to non-symbolic proportions in the human frontoparietal cortex. European Journal of Neuroscience, 30(7), 1432–1442. http://doi.org/10.1111/j.1460-9568.2009.06932.x

Jacob, S. N., Vallentin, D., & Nieder, A. (2012). Relating magnitudes: the brain's code for proportions. Trends in Cognitive Sciences, 16(3), 157–166. http://doi.org/10.1016/j.tics.2012.02.002

Jaramillo, S., & Zador, A. M. (2014). Mice and rats achieve similar levels of performance in an adaptive decision-making task. Frontiers in Systems Neuroscience.

Jeong, Y., Levine, S. C., & Huttenlocher, J. (2007). The Development of Proportional Reasoning: Effect of Continuous Versus Discrete Quantities. Journal of Cognition and Development, 8(2), 237–256. http://doi.org/10.1080/15248370701202471

Johnson, R., Barnhardt, J., & Zhu, J. (2004). The contribution of executive processes to deceptive responding. Neuropsychologia.

Jones, S. M., Cantlon, J. F., Merritt, D. J., & Brannon, E. M. (2010). Context affects the numerical semantic congruity effect in rhesus monkeys (Macaca mulatta). Behavioural Processes, 83(2), 191–196. http://doi.org/10.1016/j.beproc.2009.12.009

Jones, S.M., Pearson, J., DeWind, N.K., Paulsen, D., Tenekedjieva, A.M., & Brannon, E.M. (2014). Lemurs and macaques share similar numerical sensitivity. Animal Cognition, 17, 503-515. doi: 10.1007/s10071-013-0682-3.

289

Jordan, K. E., & Brannon, E. M. (2006). Weber's Law influences numerical representations in rhesus macaques (Macaca mulatta). Animal Cognition, 9(3), 159– 172. http://doi.org/10.1007/s10071-006-0017-8

Kadosh, R. C., Bien, N., & Sack, A. T. (2012). Automatic and Intentional Number Processing Both Rely on Intact Right Parietal Cortex: A Combined fMRI and Neuronavigated TMS Study. Frontiers in Human Neuroscience, 6. http://doi.org/10.3389/fnhum.2012.00002/abstract

Kadosh, R. C., Soskic, S., Iuculano, T., & Kanai, R. (2010). Modulating Neuronal Activity Produces Specific and Long-Lasting Changes in Numerical Competence. Current Biology.

Kamble, N., Netravathi, M., & Pal, P. K. (2014). Therapeutic applications of repetitive transcranial magnetic stimulation (rTMS) in movement disorders: A review. Parkinsonism & Related Disorders, 20(7), 695–707. http://doi.org/10.1016/j.parkreldis.2014.03.018

Kanwisher, N., McDermott, J., & Chun, M. M. (1997). The fusiform face area: A module in human extrastriate cortex specialized for face perception. The Journal of Neuroscience, 17(11), 4302–4311.

Kao, M. H., Doupe, A. J., & Brainard, M. S. (2005). Contributions of an avian basal ganglia--forebrain circuit to real-time modulation of song. Nature, 433(7026), 638– 643. http://doi.org/10.1038/nature03127

Kapogiannis, D., & Wassermann, E. M. (2008). Transcranial magnetic stimulation in clinical pharmacology. Central Nervous System Agents in Medicinal Chemistry, 8(4), 234–240.

Karim, A. A., Schneider, M., Lotze, M., Veit, R., Sauseng, P., Braun, C., & Birbaumer, N. (2010). The truth about lying: inhibition of the anterior prefrontal cortex improves deceptive behavior. Cerebral Cortex (New York, N.Y. : 1991), 20(1), 205–213. http://doi.org/10.1093/cercor/bhp090

Karton, I., & Bachmann, T. (2011). Effect of prefrontal transcranial magnetic stimulation on spontaneous truth-telling. Behavioural Brain Research, 225(1), 209–214. http://doi.org/10.1016/j.bbr.2011.07.028

Karton, I., Palu, A., Jõks, K., & Bachmann, T. (2014a). Deception rate in a “lying game”: Different effects of excitatory repetitive transcranial magnetic stimulation of right and left dorsolateral prefrontal cortex not found with …. Neuroscience Letters, 583, 290

21–25. http://doi.org/10.1016/j.neulet.2014.09.020

Karton, I., Rinne, J. M., & Bachmann, T. (2014b). Facilitating the right but not left DLPFC by TMS decreases truthfulness of object-naming responses. Behavioural Brain Research, 271, 89–93.

Kaylor-Hughes, C. J., Lankappa, S. T., Fung, R., Hope-Urwin, A. E., Wilkinson, I. D., & Spence, S. A. (2011). The functional anatomical distinction between truth telling and deception is preserved among people with schizophrenia. Criminal Behaviour and Mental Health, 21(1), 8–20. http://doi.org/10.1002/cbm.785

Kayser, A. S., Buchsbaum, B. R., Erickson, D. T., & D'Esposito, M. (2010). The Functional Anatomy of a Perceptual Decision in the Human Brain. J Neurophysiol, 103(3), 1179– 1194. http://doi.org/10.1152/jn.00364.2009

Kelly, S. P., & O'Connell, R. G. (2013). Internal and External Influences on the Rate of Sensory Evidence Accumulation in the Human Brain. The Journal of Neuroscience, 33(50), 19434–19441. http://doi.org/10.1523/JNEUROSCI.3355-13.2013

Kelly, S. P., & O'Connell, R. G. (2015). The neural processes underlying perceptual decision making in humans: Recent progress and future directions. Journal of Physiology-Paris, 109(1-3), 27–37. http://doi.org/10.1016/j.jphysparis.2014.08.003

Kiani, R., & Shadlen, M. N. (2009). Representation of confidence associated with a decision by neurons in the parietal cortex. Science, 324, 759–764.

Kiani, R., Cueva, C. J., Reppas, J. B., & Newsome, W. T. (2014). Dynamics of Neural Population Responses in Prefrontal Cortex Indicate Changes of Mind on Single Trials. Current Biology, 24(13), 1542–1547. http://doi.org/10.1016/j.cub.2014.05.049

Killgore, W. D. S., Balkin, T. J., & Wesensten, N. J. (2006). Impaired decision making following 49 h of sleep deprivation. Journal of Sleep Research, 15(1), 7–13. http://doi.org/10.1111/j.1365-2869.2006.00487.x

Kim, J.-N., & Shadlen, M. N. (1999). Neural correlates of a decision in the dorsolateral prefrontal cortex of the macaque - Nature Neuroscience. Nature Neuroscience, 2(2), 176–185. http://doi.org/10.1038/5739

Kimmel, D.L., Mammo, D., and Newsome, W.T. (2012). Tracking the eye non-invasively: simultaneous comparison of the scleral search coil and optical tracking techniques in the macaque monkey. Front. Behav. Neurosci. 6:49, 1-17.

291

Kireev, M. V., Korotkov, A. D., & Medvedev, S. V. (2012). Functional magnetic resonance study of deliberate deception. Human Physiology, 38(1), 32–39. http://doi.org/10.1134/S0362119711060090

Kireev, M., Korotkov, A., Medvedeva, N., & Medvedev, S. (2013). Possible role of an error detection mechanism in brain processing of deception: PET-fMRI study. International Journal of Psychophysiology, 90(3), 291–299. http://doi.org/10.1016/j.ijpsycho.2013.09.005

Kirjakovski, A., & Utsuki, N. (2012). From SNARC to SQUARC: Universal Mental Quantity Line? International Journal of Psychological Studies, 4(2). http://doi.org/10.5539/ijps.v4n2p217

Koob, G. F. (1992). Dopamine, addiction and reward. Seminars in Neuroscience, 4(2), 139– 148. http://doi.org/10.1016/1044-5765(92)90012-Q

Kozel, F. A., Padgett, T. M., & George, M. S. (2004a). A replication study of the neural correlates of deception. Behavioral Neuroscience.

Kozel, F. A., Revell, L. J., & Lorberbaum, J. P. (2004b). A pilot study of functional magnetic resonance imaging brain correlates of deception in healthy young men. The Journal of ….

Kozyrev, V., Eysel, U. T., & Jancke, D. (2014). Voltage-sensitive dye imaging of transcranial magnetic stimulation-induced intracortical dynamics. Proceedings of the National Academy of Sciences of the United States of America, 111(37), 13553–13558. http://doi.org/10.1073/pnas.1405508111

Krebs, J. R., Kacelnik, A., & Taylor, P. (1978). Test of optimal sampling by foraging great tits. Nature, 275, 27–31.

Krnjevic, K. (1992). Cellular and synaptic actions of general anaesthetics. General Pharmacology, 23(6), 965–975.

Krusche, P., Uller, C., & Dicke, U. (2010). Quantity discrimination in salamanders. Journal of Experimental Biology, 213(11), 1822–1828. doi:10.1242/jeb.039297

Kurata, K., and Hoffman, D.S. (1994). Differential effects of muscimol microinjection into dorsal and ventral aspects of the premotor cortex of monkeys. J. Neurophysiol. 71, 1151–1164.

Kurzban, R., Burton-Chellew, M. N., & West, S. A. (2015). The Evolution of Altruism in

292

Humans. Annual Review of Psychology, 66, 575–599.

LaMotte, R. H., & Mountcastle, V. B. (1975). Capacities of humans and monkeys to discriminate vibratory stimuli of different frequency and amplitude: a correlation between neural events and psychological measurements. Journal of Neurophysiology, 38(3), 539–559.

LaMotte, R. H., & Mountcastle, V. B. (1979). Disorders in somesthesis following lesions of parietal lobe. Journal of Neurophysiology.

Langleben, D. D., Loughead, J. W., Bilker, W. B., Ruparel, K., Childress, A. R., Busch, S. I., & Gur, R. C. (2005). Telling truth from lie in individual subjects with fast event- related fMRI. Human Brain Mapping, 26(4), 262–272. http://doi.org/10.1002/hbm.20191

Langleben, D. D., Schroeder, L., Maldjian, J. A., Gur, R. C., McDonald, S., Ragland, J. D., et al. (2002). Brain Activity during Simulated Deception: An Event-Related Functional Magnetic Resonance Study. NeuroImage, 15(3), 727–732. http://doi.org/10.1006/nimg.2001.1003

Larsson, M. (2013). The optic chiasm: a turning point in the evolution of eye/hand coordination. Frontiers in Zoology, 10(41).

Latimer, K. W., Yates, J. L., Meister, M. L. R., Huk, A. C., & Pillow, J. W. (2015). Single- trial spike trains in parietal cortex reveal discrete steps during decision-making. Science, 349(6244), 184–187. http://doi.org/10.1126/science.aaa4056

Le Moal, M., & Simon, H. (1991). Mesocorticolimbic dopaminergic network: functional and regulatory roles. Physiological Reviews, 71(1), 155–234. le Roux, A., Snyder-Mackler, N., Roberts, E. K., Beehner, J. C., & Bergman, T. J. (2013). Evidence for tactical concealment in a wild primate. Nature Communications, 4, 1462– 1462. http://doi.org/10.1038/ncomms2468

Lee, D., & Seo, H. (2016). Neural Basis of Strategic Decision Making. Trends in Neurosciences, 39(1), 40–48. http://doi.org/10.1016/j.tins.2015.11.002

Lee, D., Seo, H., & Jung, M. W. (2012). Neural Basis of Reinforcement Learning and Decision Making. Annual Review of Neuroscience, 35(1), 287–308. http://doi.org/10.1146/annurev-neuro-062111-150512

Lee, J., Choi, B. H., Oh, E., Sohn, E. H., & Lee, A. Y. (2016). Treatment of Alzheimer's

293

Disease with Repetitive Transcranial Magnetic Stimulation Combined with Cognitive Training: A Prospective, Randomized, Double-Blind, Placebo-Controlled Study. Journal of Clinical Neurology (Seoul, Korea), 12(1), 57–64. http://doi.org/10.3988/jcn.2016.12.1.57

Lee, T. M. C., Lee, T. M. Y., Raine, A., & Chan, C. C. H. (2010). Lying about the Valence of Affective Pictures: An fMRI Study. PLoS One, 5(8), e12291. http://doi.org/10.1371/journal.pone.0012291

Lee, T. M. C., Liu, H.-L., Tan, L.-H., Chan, C. C. H., Mahankali, S., Feng, C.-M., et al. (2002). Lie detection by functional magnetic resonance imaging. Human Brain Mapping, 15(3), 157–164. http://doi.org/10.1002/hbm.10020

Lee, T., Au, R., Liu, H. L., Ting, K. H., & Huang, C. M. (2009). Are errors differentiable from deceptive responses when feigning memory impairment? An fMRI study. Brain and ….

Lefaucheur, J.-P., André-Obadia, N., Antal, A., Ayache, S. S., Baeken, C., Benninger, D. H., et al. (2014). Evidence-based guidelines on the therapeutic use of repetitive transcranial magnetic stimulation (rTMS). Clinical Neurophysiology : Official Journal of the International Federation of Clinical Neurophysiology, 125(11), 2150–2206. http://doi.org/10.1016/j.clinph.2014.05.021

Lemus, L., Hernandez, A., Luna, R., Zainos, A., Nacher, V., & Romo, R. (2007). Neural correlates of a postponed decision report (pp. 17174–17179). Presented at the Proceedings of the ….

Lenz, M., Galanis, C., Müller-Dahlhaus, F., Opitz, A., Wierenga, C. J., Szabó, G., et al. (2016). Repetitive magnetic stimulation induces plasticity of inhibitory synapses. Nature Communications, 7, 10020. http://doi.org/10.1038/ncomms10020

Lenz, M., Platschek, S., Priesemann, V., Becker, D., Willems, L. M., Ziemann, U., et al. (2014). Repetitive magnetic stimulation induces plasticity of excitatory postsynapses on proximal dendrites of cultured mouse CA1 pyramidal neurons. Brain Structure and Function, 220(6), 3323–3337. http://doi.org/10.1007/s00429-014-0859-9

Liang, C. Y., Xu, Z. Y., Mei, W., Wang, L. L., Xue, L., & Lu, D. J. (2012). Neural correlates of feigned memory impairment are distinguishable from answering randomly and answering incorrectly: an fMRI and behavioral study. Brain and ….

Liberman, A. M., Cooper, F. S., Shankweiler, D. P., & Studdert-Kennedy, M. (1967). Perception of the Speech Code. 294

Liberman, A. M., Harris, K. S., Kinney, J. A., & Lane, H. (1961). The discrimination of relative onset-time of the components of certain speech and nonspeech patterns. Journal of Experimental Psychology, 61(6), 379–388.

Libertus, M. E., Feigenson, L., & Halberda, J. (2011). Preschool acuity of the approximate number system correlates with school math ability. Developmental Science, 14(6), 1292–1300. http://doi.org/10.1111/j.1467-7687.2011.01080.x

Liu, T., & Pleskac, T. J. (2011). Neural correlates of evidence accumulation in a perceptual decision task. J Neurophysiol, 106(5), 2383–2398. http://doi.org/10.1152/jn.00413.2011

Lo, Y. L., Fook-Chong, S., & Tan, E. K. (2003). Increased cortical excitability in human deception. Neuroreport, 14(7), 1021–1024. http://doi.org/10.1097/01.wnr.0000070829.57864.e4

Lu, W., Pauly, K.B., Gold, G.E., Pauly, J.M., and Hargreaves, B.A. (2009). SEMAC: Slice encoding for metal artifact correction in MRI. Magn. Reson. Med. 62, 66–76.

Luber, B. (2014). Neuroenhancement by noninvasive brain stimulation is not a net zero- sum proposition. Frontiers in Systems Neuroscience, 8(12), 1058. http://doi.org/10.3389/fnsys.2014.00127

Luber, B., & Lisanby, S. H. (2013). Enhancement of human cognitive performance using transcranial magnetic stimulation (TMS). NeuroImage, 85(P3), 961–970. http://doi.org/10.1016/j.neuroimage.2013.06.007

Luber, B., and Lisanby, S.H. (2014). Enhancement of human cognitive performance using transcranial magnetic stimulation (TMS). NeuroImage 85, Part 3, 961–970.

Luber, B., Fisher, C., Appelbaum, P. S., Ploesser, M., & Lisanby, S. H. (2009). Non- invasive brain stimulation in the detection of deception: Scientific challenges and ethical consequences. Behavioral Sciences & the Law, 27(2), 191–208. http://doi.org/10.1002/bsl.860

Lüdeke, K.M., Röschmann, P., and Tischler, R. (1985). Susceptibility artefacts in NMR imaging. Magn. Reson. Imaging 3, 329–343.

Lyon, B. E. (2003). Egg recognition and counting reduce costs of avian conspecific brood parasitism. Nature, 422(6931), 495–499.

Lyons, I. M., & Beilock, S. L. (2011). Numerical ordering ability mediates the relation

295

between number-sense and arithmetic competence. Cognition, 1–6. http://doi.org/10.1016/j.cognition.2011.07.009

Machado, C.J., and Nelson, E.E. (2011). Eye-tracking with nonhuman primates is now more accessible than ever before. Am. J. Primatol. 73, 562–569.

Machens, C. K., Romo, R., & Brody, C. D. (2005). Flexible control of mutual inhibition: a neural model of two-interval discrimination. Science, 307(5712), 1121–1124. http://doi.org/10.1126/science.1105783

Mante, V., Sussillo, D., Shenoy, K. V., & Newsome, W. T. (2013). Context-dependent computation by recurrent dynamics in prefrontal cortex. Nature, 503(7474), 78–84. http://doi.org/10.1038/nature12742

Marchewka, A., Jednorog, K., Falkiewicz, M., Szeszkowski, W., Grabowska, A., & Szatkowska, I. (2012). Sex, Lies and fMRI---Gender Differences in Neural Basis of Deception. PLoS One, 7(8), e43076. http://doi.org/10.1371/journal.pone.0043076

Mars, R. B., Sallet, J., Neubert, F.-X., & Rushworth, M. F. S. (2013). Connectivity profiles reveal the relationship between brain areas for social cognition in human and monkey temporoparietal cortex. Proceedings of the National Academy of Sciences of the United States of America, 110(26), 10806–10811. http://doi.org/10.1073/pnas.1302956110

Martin, J. H., & Ghez, C. (1999). Pharmacological inactivation in the analysis of the central control of movement. Journal of Neuroscience Methods, 86(2), 145–159. http://doi.org/10.1016/S0165-0270(98)00163-0

Maunsell, J. H., & Van Essen, D. C. (1983). Functional properties of neurons in middle temporal visual area of the macaque monkey. I. Selectivity for stimulus direction, speed, and orientation. Journal of Neurophysiology.

Mays, L. E., & Sparks, D. L. (1980). Dissociation of visual and saccade-related responses in superior colliculus neurons. Journal of Neurophysiology, 43(1), 207–232.

Mazzocco, M. M. M., Feigenson, L., & Halberda, J. (2011a). Impaired Acuity of the Approximate Number System Underlies Mathematical Learning Disability (Dyscalculia). Child Development, 82(4), 1224–1237. http://doi.org/10.1111/j.1467- 8624.2011.01608.x

Mazzocco, M. M. M., Feigenson, L., & Halberda, J. (2011b). Preschoolers' Precision of the Approximate Number System Predicts Later School Mathematics Performance.

296

Journals.Plos.org, 6(9), e23749. http://doi.org/10.1371/journal.pone.0023749

McCarthy, G., Puce, A., Gore, J. C., & Allison, T. (1997). Face-Specific Processing in the Human Fusiform Gyrus. Journal of Cognitive Neuroscience, 9(5), 605–610.

McComb, K., Packer, C., & Pusey, A. (1994). Roaring and numerical assessment in contests between groups of female lions, Panthera leo. Animal Behaviour, 47(2), 379– 387. http://doi.org/10.1006/anbe.1994.1052

McCrink, K., & Wynn, K. (2007). Ratio Abstraction by 6-Month-Old Infants. Psychological Science, 18(8), 740–745. http://doi.org/10.1111/j.1467-9280.2007.01969.x

McGrew, W.C., & Marchant, L.F. (1997). On the other hand: Current issues in and meta- analysis of the behavioral laterality of hand function in nonhuman primates. Yearbook of Physical Anthropology, 40, 201-232.

McPherson, B., McMahon, K., Wilson, W., & Copland, D. (2011). ``I know you can hear me“”: Neural correlates of feigned hearing loss. Human Brain Mapping, 33(8), 1964– 1972. http://doi.org/10.1002/hbm.21337

McWhirter, G. (2015). Behavioural Deception and Formal Models of Communication. The British Journal for the Philosophy of Science, axv001. http://doi.org/10.1093/bjps/axv001

Meck, W. H., & Church, R. H. (1983). A mode control model of counting and timing processes. … Psychology: Animal Behavior Processes.

Menzel, E. W. (1974). A group of young chimpanzees in a one-acre field: Leadership and communication. In A. M. Shrier & F. Stollnitz (Eds.), Behavior of nonhuman primates (pp. 83–153). Academic Press.

Merritt, D. J., Rugani, R., & Brannon, E. M. (2009). Empty sets as part of the numerical continuum: conceptual precursors to the zero concept in rhesus monkeys. Journal of Experimental Psychology General, 138(2), 258–269. http://doi.org/10.1037/a0015231

Miller, A. K., Maritz, B., McKay, S., Glaudas, X., & Alexander, G. J. (2015). An ambusher's arsenal: chemical crypsis in the puff adder ( Bitis arietans). Proceedings of the Royal Society B: Biological Sciences, 282(1821), 20152182–8. http://doi.org/10.1098/rspb.2015.2182

Miller, P., & Wang, X.-J. (2006). Inhibitory control by an integral feedback signal in prefrontal cortex: a model of discrimination between sequential stimuli. Proceedings

297

of the National Academy of Sciences of the United States of America, 103(1), 201–206.

Milner, A. D., & Goodale, M. A. (2008). Two visual systems re-viewed. Neuropsychologia, 46(3), 774–785. http://doi.org/10.1016/j.neuropsychologia.2007.10.005

Mimura, M., Oeda, R., & Kawamura, M. (2006). Impaired decision-making in Parkinson's disease. Parkinsonism & Related Disorders, 12(3), 169–175.

Miniussi, C., Harris, J. A., & Ruzzoli, M. (2013). Modelling non-invasive brain stimulation in cognitive neuroscience. Neuroscience & Biobehavioral Reviews, 37(8), 1702–1712. http://doi.org/10.1016/j.neubiorev.2013.06.014

Miranda-Dominguez, O., Mills, B.D., Grayson, D., Woodall, A., Grant, K.A., Kroenke, C.D., and Fair, D.A. (2014). Bridging the Gap between the Human and Macaque Connectome: A Quantitative Comparison of Global Interspecies Structure-Function Relationships and Network Topology. J. Neurosci. 34, 5552–5563.

Mishkin, M., Ungerleider, L. G., & Macko, K. A. (1983). Object vision and spatial vision: two cortical pathways. Trends in Neurosciences, 6, 414–417. http://doi.org/10.1016/0166-2236(83)90190-X

Moliadze, V. (2003). Effect of transcranial magnetic stimulation on single-unit activity in the cat primary visual cortex. The Journal of Physiology, 553(2), 665–679. http://doi.org/10.1113/jphysiol.2003.050153

Montague, P. R., Dayan, P., Person, C., & Sejnowski, T. J. (1995). Bee foraging in uncertain environments using predictive hebbian learning. Nature.

Mountcastle, V. B., Steinmetz, M. A., & Romo, R. (1990). Frequency discrimination in the sense of flutter: psychophysical measurements correlated with postcentral events in behaving monkeys. Journal of Neuroscience, 10(9), 3032–3044.

Mountcastle, V. B., Talbot, W. H., Darian-Smith, I., & Kornhuber, H. H. (1967). Neural Basis of the Sense of Flutter-Vibration. Science, 155(3762), 597–600. http://doi.org/10.1126/science.155.3762.597

Mueller, J.K., Grigsby, E.M., Prevosto, V., Petraglia, F.W., Rao, H., Deng, Z.-D., Peterchev, A.V., Sommer, M.A., Egner, T., Platt, M.L., et al. (2014). Simultaneous transcranial magnetic stimulation and single-neuron recording in alert non-human primates. Nat. Neurosci. 17, 1130-1136. http://doi.org/10.1038/nn.3751

Mulder, M. J., Van Maanen, L., & Forstmann, B. U. (2014). Perceptual decision

298

neurosciences--a model-based review. Neuroscience, 277, 872–884.

Muller-Dahlhaus, F., & Vlachos, A. (2013). Unraveling the cellular and molecular mechanisms of repetitive magnetic stimulation, 1–7. http://doi.org/10.3389/fnmol.2013.00050/abstract

Nacher, V., Ledberg, A., Deco, G., & Romo, R. (2013). Coherent delta-band oscillations between cortical areas correlate with decision making. Proceedings of the National Academy of Sciences, 110(37), 15085–15090. http://doi.org/10.1073/pnas.1314681110

Nakamura, K., Sakai, K., & Hikosaka, O. (1999). Effects of Local Inactivation of Monkey Medial Frontal Cortex in Learning of Sequential Procedures. Journal of Neurophysiology, 82(2), 1063–1068. http://doi.org/10.1007/BF00275005

Nevler, N., & Ash, E. L. (2015). TMS as a Tool for Examining Cognitive Processing. Current Neurology and Neuroscience Reports, 15(8), 1–11. http://doi.org/10.1007/s11910- 015-0575-8

Newsome, W. T., & Pare, E. B. (1988). A selective impairment of motion perception following lesions of the middle temporal visual area (MT). The Journal of Neuroscience, 8(6), 2201–2211.

Newsome, W. T., Britten, K. H., & Movshon, J. A. (1989). Neuronal correlates of a perceptual decision. Nature, 341(6237), 52–54. http://doi.org/10.1038/341052a0

Nieder, A. (2012a). Coding of abstract quantity by `number neurons' of the primate brain. Journal of Comparative Physiology A, 199(1), 1–16. http://doi.org/10.1007/s00359- 012-0763-9

Nieder, A. (2012b). Coding of abstract quantity by “number neurons” of the primate brain. Journal of Comparative Physiology A, 199(1), 1–16. http://doi.org/10.1007/s00359- 012-0763-9

Nieder, A., & Dehaene, S. (2009). Representation of Number in the Brain. Annual Review of Neuroscience, 32(1), 185–208. http://doi.org/10.1146/annurev.neuro.051508.135550

Nieder, A., & Miller, E. K. (2004). A parieto-frontal network for visual numerical information in the monkey. Proceedings of the National Academy of Sciences of the United States of America, 101(19), 7457–7462. http://doi.org/10.1073/pnas.0402239101

Nose, I., Murai, J., & Taira, M. (2009). Disclosing concealed information on the basis of cortical activations. NeuroImage.

299

Nunez, J. M., Casey, B. J., Egner, T., Hare, T., & Hirsch, J. (2005). Intentional false responding shares neural substrates with response conflict and cognitive control. NeuroImage.

O'Connell, R. G., Dockree, P. M., & Kelly, S. P. (2012). A supramodal accumulation-to- bound signal that determines perceptual decisions in humans. Nature Neuroscience, 15(12), 1729–1735. http://doi.org/10.1038/nn.3248

Oberman, L. M., Rotenberg, A., & Pascual-Leone, A. (2013). Use of Transcranial Magnetic Stimulation in Autism Spectrum Disorders. Journal of Autism and Developmental Disorders, 45(2), 524–536. http://doi.org/10.1007/s10803-013-1960-2

Opfer, J.E., & Furlong, E.E. (2011). How numbers bias preschoolers' spatial search. Journal of Cross-Cultural Psychology, 42(4), 682-695.

Opfer, J.E., Thompson, C.A., & Furlong, E.E. (2010). Early development of spatial- numerica associations: evidence from spatial and quantitative performance of preschoolers. Developmental Science, 13(5), 761-771.

Overduin-de Vries, A. M., Spruijt, B. M., De Vries, H., & Sterck, E. H. M. (2015). Tactical deception to hide sexual behaviour: macaques use distance, not visibility. Behavioral Ecology and Sociobiology, 69(8), 1333–1342. http://doi.org/10.1007/s00265-015-1946-5

Padoa-Schioppa, C., & Assad, J. A. (2006). Neurons in the orbitofrontal cortex encode economic value. Nature, 441(7090), 223–226. http://doi.org/10.1038/nature04676

Pahl, M., Si, A., & Zhang, S. (2013). Numerical Cognition in Bees and Other Insects. Frontiers in Psychology, 4. http://doi.org/10.3389/fpsyg.2013.00162

Papademetriou, E., Sheu, C.F., & Michel, G.F. (2005). A meta-analysis of primate hand preferences, particularly for reaching. Journal of Comparative Psychology, 119(1), 33-48.

Park, J., & Brannon, E. M. (2013). Training the Approximate Number System Improves Math Proficiency. Psychological Science, 24(10), 0956797613482944–2019. http://doi.org/10.1177/0956797613482944

Parkin, B. L., Ekhtiari, H., & Walsh, V. F. (2015). Non-invasive Human Brain Stimulation in Cognitive Neuroscience: A Primer. Neuron, 87(5), 932–945. http://doi.org/10.1016/j.neuron.2015.07.032

Pell, G. S., Roth, Y., & Zangen, A. (2011). Modulation of cortical excitability induced by repetitive transcranial magnetic stimulation: Influence of timing and geometrical

300

parameters and underlying mechanisms. Progress in Neurobiology, 93(1), 59–98. http://doi.org/10.1016/j.pneurobio.2010.10.003

Pelli, D. G. (1997). The VideoToolbox software for visual psychophysics: transforming numbers into movies. Spatial Vision, 10(4), 437–442. http://doi.org/10.1163/156856897X00366

Penn, D. C., Holyoak, K. J., & Povinelli, D. J. (2008). Darwin's mistake: explaining the discontinuity between human and nonhuman minds. Behavioral and Brain Sciences, 31(2), 109–30; discussion 130–178. http://doi.org/10.1017/S0140525X08003543

Phan, K. L., Magalhaes, A., Ziemlewicz, T. J., Fitzgerald, D. A., Green, C., & Smith, W. (2005). Neural correlates of telling lies: A functional magnetic resonance imaging study at 4 Tesla. Academic Radiology.

Phillips, A. G., Brooke, S. M., & Fibiger, H. C. (1975). Effects of amphetamine isomers and neuroleptics on self-stimulation from the nucleus accumbens and dorsal nor- adrenergenic bundle. Brain Research, 85(1), 13–22. http://doi.org/10.1016/0006- 8993(75)90998-1

Piazza, M. (2010). Neurocognitive start-up tools for symbolic number representations. Trends in Cognitive Sciences, 14(12), 542–551. http://doi.org/10.1016/j.tics.2010.09.008

Pica, P., Lemer, C., Izard, V., & Dehaene, S. (2004). Exact and approximate arithmetic in an Amazonian indigene group. Science, 306(5695), 499–503.

Picht, T. (2014). Current and potential utility of transcranial magnetic stimulation in the diagnostics before brain tumor surgery. CNS Oncology, 3(4), 299–310. http://doi.org/10.2217/cns.14.25

Piffer, L., Agrillo, C., & Hyde, D. C. (2011). Small and large number discrimination in guppies. Animal Cognition, 15(2), 215–221. doi:10.1007/s10071-011-0447-9

Pigarev, I.N., Nothdurft, H.-C., and Kastner, S. (1997). A reversible system for chronic recordings in macaque monkeys. J. Neurosci. Methods 77, 157–162.

Pigarev, I.N., Saalmann, Y.B., and Vidyasagar, T.R. (2009). A minimally invasive and reversible system for chronic recordings from multiple brain sites in macaque monkeys. J. Neurosci. Methods 181, 151–158.

Platt, M. L., & Glimcher, P. W. (1999). Neural correlates of decision variables in parietal cortex. Nature, 400(6741), 233–238. http://doi.org/10.1038/22268

301

Platt, M. L., & Huettel, S. A. (2008). Risky business: the neuroeconomics of decision making under uncertainty. Nature Neuroscience, 11(4), 398–403. http://doi.org/10.1038/nn2062

Pleger, B., Ruff, C. C., Blankenburg, F., Bestmann, S., Wiech, K., Stephan, K. E., et al. (2006). Neural Coding of Tactile Decisions in the Human Prefrontal Cortex. Journal of Neuroscience, 26(48), 12596–12601. http://doi.org/10.1523/JNEUROSCI.4275-06.2006

Porter, R., Lewis, M.M., and Linklater, G.F. (1971). A headpiece for recording discharges of neurons in unrestrained monkeys. Electroencephalogr. Clin. Neurophysiol. 30, 91–93.

Premack, D. (1983). The codes of man and beasts. Behavioral and Brain Sciences, 6(01), 125–136. http://doi.org/10.1017/S0140525X00015077

Premack, D., & Woodruff, G. (1978). Does the chimpanzee have a theory of mind? Behavioral and Brain Sciences, 4, 515–526.

Priori, A., Mameli, F., Cogiamanian, F., Marceglia, S., Tiriticco, M., Mrakic-Sposta, S., et al. (2008). Lie-specific involvement of dorsolateral prefrontal cortex in deception. Cerebral Cortex (New York, N.Y. : 1991), 18(2), 451–455. http://doi.org/10.1093/cercor/bhm088

Radman, T., Ramos, R. L., Brumberg, J. C., & Bikson, M. (2009). Role of cortical cell type and morphology in subthreshold and suprathreshold uniform electric field stimulation in vitro. Brain Stimulation, 2(4), 215–228.e3. http://doi.org/10.1016/j.brs.2009.03.007

Radomski, A. D., Power, C., Purdon, S. E., Emery, D. J., Blevins, G., Warren, K. G., & Fujiwara, E. (2015). Decision-making under explicit risk is impaired in multiple sclerosis: relationships with ventricular width and disease disability. BMC Neurology, 15(1), 1. http://doi.org/10.1186/s12883-015-0318-0

Rakoczy, H., Clüver, A., Saucke, L., Stoffregen, N., Gräbener, A., Migura, J., & Call, J. (2014). Apes are intuitive statisticians. Cognition, 131(1), 60–68. http://doi.org/10.1016/j.cognition.2013.12.011

Rangel, A., Camerer, C., & Montague, P. R. (2008). A framework for studying the neurobiology of value-based decision making. Nature Reviews Neuroscience, 9(7), 545–556. http://doi.org/10.1038/nrn2357

Raposo, D., Kaufman, M. T., & Churchland, A. K. (2014). A category-free neural population supports evolving demands during decision-making. Nature Publishing

302

Group, 17(12), 1784–1792. http://doi.org/doi:10.1038/nn.3865

Raposo, D., Sheppard, J. P., Schrater, P. R., & Churchland, A. K. (2012). Multisensory Decision-Making in Rats and Humans. Journal of Neuroscience, 32(11), 3726–3735. http://doi.org/10.1523/JNEUROSCI.4998-11.2012

Recanzone, G. H., Merzenich, M. M., & Schreiner, C. E. (1992). Changes in the distributed temporal response properties of SI cortical neurons reflect improvements in performance on a temporally based tactile discrimination task. Journal of Neurophysiology, 67(5), 1071–1091.

Reithler, J., Peters, J. C., & Sack, A. T. (2011). Multimodal transcranial magnetic stimulation: Using concurrent neuroimaging to reveal the neural network dynamics of noninvasive brain stimulation. Progress in Neurobiology, 94(2), 149–165. http://doi.org/10.1016/j.pneurobio.2011.04.004

Resulaj, A., Kiani, R., Wolpert, D. M., & Shadlen, M. N. (2009). Changes of mind in decision-making. Nature, 461, 263–268.

Richter, L., Neumann, G., Oung, S., Schweikard, A., & Trillenberg, P. (2013). Optimal coil orientation for transcranial magnetic stimulation. Journals.Plos.org, 8(4), e60358. http://doi.org/10.1371/journal.pone.0060358

Roberts, W. A. (2010). Distance and magnitude effects in sequential number discrimination by pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 36(2), 206–216. doi:10.1037/a0017226

Roberts, W. A., & Mazmanian, D. S. (1988). Concept Learning at Different Levels of Abstraction by Pigeons, Monkeys, and People. Journal of Experimental Psychology: Animal Behavior Processes, 14(3), 247–260.

Rogers, L.J., Vallortigara, G., & Andrews, R.J. (2013). Divided Brains: The Biology and Behaviour of Brain Asymmetries. Cambridge University Press.

Roitman, J. D., & Shadlen, M. N. (2002). Response of neurons in the lateral intraparietal area during a combined visual discrimination reaction time task. The Journal of Neuroscience, 22(21), 9475–9489.

Roitman, J. D., Brannon, E. M., & Platt, M. L. (2007). Monotonic coding of numerosity in macaque lateral intraparietal area. PLoS Biology, 5(8), e208. http://doi.org/10.1371/journal.pbio.0050208

303

Romo, R., Brody, C. D., Hernández, A., & Lemus, L. (1999). Neuronal correlates of parametric working memory in the prefrontal cortex. Nature, 399(6735), 470–473. http://doi.org/10.1038/20939

Romo, R., Hernández, A., & Zainos, A. (2004). Neuronal correlates of a perceptual decision in ventral premotor cortex. Neuron, 41(1), 165–173.

Romo, R., Hernández, A., Zainos, A., & Salinas, E. (1998). Somatosensory discrimination based on cortical microstimulation. Nature, 392(6674), 387–390. http://doi.org/10.1038/32891

Romo, R., Hernández, A., Zainos, A., Lemus, L., & Brody, C. D. (2002). Neuronal correlates of decision-making in secondary somatosensory cortex. Nature Neuroscience, 5(11), 1217–1225. http://doi.org/10.1038/nn950

Rorie, A. E., Gao, J., McClelland, J. L., & Newsome, W. T. (2010). Integration of Sensory and Reward Information During Perceptual Decision-Making in Lateral Intraparietal Cortex (LIP). PLoS One, 5(2), e9308. http://doi.org/10.1371/journal.pone.0009308

Rosset, A., Spadola, L., & Ratib, O. (2004). OsiriX: An Open-Source Software for Navigating in Multidimensional DICOM Images. Journal of Digital Imaging, 17(3), 205–216. http://doi.org/10.1007/s10278-004-1014-6

Roy, A., Shepherd, S.V., and Platt, M.L. (2014). Reversible inactivation of pSTS suppresses social gaze following in the macaque (Macaca mulatta). Soc. Cogn. Affect. Neurosci. 9, 209–217.

Roy, S., and Wang, X. (2012). Wireless multi-channel single unit recording in freely moving and vocalizing primates. J. Neurosci. Methods 203, 28–40.

Rubinsten, O., & Sury, D. (2011). Processing Ordinality and Quantity: The Case of Developmental Dyscalculia. PLoS One, 6(9), e24079.

Ruby, L. M. (1984). An Investigation of Number-concept Appreciation in a Rhesus Monkey. Primates, 25(2), 236–242.

Rugani, R., Kelly, D., Szelest, I., Regolin, L., & Vallortigara, G. (2010). Is it only humans that count from left to right? Biology Letters, 6, 290–292.

Rugani, R., Regolin, L., & Vallortigara, G. (2007). Rudimental numerical competence in 5-day-old domestic chicks (Gallus gallus): Identification of ordinal position. Journal

304

of Experimental Psychology: Animal Behavior Processes, 33(1), 21–31. http://doi.org/10.1037/0097-7403.33.1.21

Rugani, R., Vallortigara, G., Priftis, K., & Regolin, L. (2015). Number-space mapping in the newborn chick resembles humans' mental number line. Science, 347(6221), 534– 536.

Rugani, R., Vallortigara, G., Vallini, B., & Regolin, L. (2011). Asymmetrical number-space mapping in the avian brain. Neurobiology of Learning and Memory, 95(3), 231–238. http://doi.org/10.1016/j.nlm.2010.11.012

Russ, B. E., Orr, L. E., & Cohen, Y. E. (2008). Prefrontal Neurons Predict Choices during an Auditory Same-Different Task. Current Biology, 18(19), 1483–1488. http://doi.org/10.1016/j.cub.2008.08.054

Sai, L., Zhou, X., Ding, X. P., Fu, G., & Sang, B. (2014). Detecting Concealed Information Using Functional Near-Infrared Spectroscopy. Brain Topography, 27(5), 652–662. http://doi.org/10.1007/s10548-014-0352-z

Salillas, E., Basso, D., Baldi, M., Semenza, C., & Vecchi, T. (2009). Motion on numbers: transcranial magnetic stimulation on the ventral intraparietal sulcus alters both numerical and motion processes. Journal of Cognitive Neuroscience, 21(11), 2129–2138.

Salinas, E., Hernandez, A., Zainos, A., & Romo, R. (2000). Periodicity and firing rate as candidate neural codes for the frequency of vibrotactile stimuli. The Journal of Neuroscience, 20(14), 5503–5515.

Salzman, C. D., Murasugi, C. M., & Britten, K. H. (1992). Microstimulation in visual area MT: effects on direction discrimination performance. The Journal of Neuroscience.

Salzman, C., & Newsome, W. (1994). Neural mechanisms for forming a perceptual decision. Science, 264(5156), 231–237. http://doi.org/10.1126/science.8146653

Samejima, K., & Doya, K. (2007). Multiple Representations of Belief States and Action Values in Corticobasal Ganglia Loops. Annals of the New York Academy of Sciences, 1104(1), 213–228. http://doi.org/10.1196/annals.1390.024

Samejima, K., Ueda, Y., Doya, K., & Kimura, M. (2005). Representation of Action-Specific Reward Values in the Striatum. Science, 310(5752), 1337–1340. http://doi.org/10.1126/science.1115270

Sandrini, M., & Rusconi, E. (2009). A brain for numbers. Cortex, 45(7), 796–803.

305

http://doi.org/10.1016/j.cortex.2008.09.002

Sandrini, M., Umiltà, C., & Rusconi, E. (2011). The use of transcranial magnetic stimulation in cognitive neuroscience: A new synthesis of methodological issues. Neuroscience & Biobehavioral Reviews, 35(3), 516–536. http://doi.org/10.1016/j.neubiorev.2010.06.005

Sawaguchi, T., & Iba, M. (2001). Prefrontal Cortical Representation of Visuospatial Working Memory in Monkeys Examined by Local Inactivation With Muscimol. Journal of Neurophysiology, 86(4), 2041–2053. http://doi.org/10.1016/0304- 3940(80)90187-1

Sawamura, H., Shima, K., & Tanji, J. (2002). Numerical representation for action in the parietal cortex of the monkey. Nature, 415(6874), 918–922.

Sawamura, H., Shima, K., & Tanji, J. (2010). Deficits in Action Selection Based on Numerical Information After Inactivation of the Posterior Parietal Cortex in Monkeys. Journal of Neurophysiology, 104(2), 902–910. http://doi.org/10.1152/jn.01014.2009

Scarf, D., & Colombo, M. (2010). Representation of serial order in pigeons (Columba livia). Journal of Experimental Psychology: Animal Behavior Processes, 36(4), 423–429. doi:10.1037/a0020926

Schoener, T. W. (1971). Theory of Feeding Strategies. Annual Review of Ecology and Systematics, 2, 369–404. http://doi.org/10.2307/2096934?ref=search- gateway:345aafdb3b4d3b503e1e1c781e10997c

Schultz, W. (2015, June). Neuronal reward and decision signals: From theories to data. Retrieved February 23, 2016, from http://physrev.physiology.org/content/physrev/95/3/853.full.pdf

Schultz, W., Dayan, P., & Montague, P. R. (1997). A Neural Substrate of Prediction and Reward. Science, 275(5306), 1593–1599. http://doi.org/10.1126/science.275.5306.1593

Schulz, R., Gerloff, C., & Hummel, F. C. (2013). Non-invasive brain stimulation in neurological diseases. Neuropharmacology, 64, 579–587. http://doi.org/10.1016/j.neuropharm.2012.05.016

Schutter, D. J. L. G., & van Honk, J. (2005). A framework for targeting alternative brain regions with repetitive transcranial magnetic stimulation in the treatment of depression. Journal of Psychiatry & Neuroscience, 30(2), 91–97.

306

Schwarz, D.A., Lebedev, M.A., Hanson, T.L., Dimitrov, D.F., Lehew, G., Meloy, J., Rajangam, S., Subramanian, V., Ifft, P.J., Li, Z., et al. (2014). Chronic, wireless recordings of large-scale brain activity in freely moving rhesus monkeys. Nat. Methods 11, 670–676.

Searcy, W. A., & Nowicki, S. (2005). The evolution of communication: reliability and deception in animal signaling systems. Princeton University.

Seo, H., Cai, X., Donahue, C. H., & Lee, D. (2014). Neural correlates of strategic reasoning during competitive games. Science, 346(6207), 340–343. http://doi.org/10.1126/science.1256254

Serences, J. T., & Boynton, G. M. (2007). The representation of behavioral choice for motion in human visual cortex. The Journal of Neuroscience, 27(47), 12893–12899. http://doi.org/10.1523/JNEUROSCI.4021-07.2007

Sereno, A. B., & Amador, S. C. (2006). Attention and memory-related responses of neurons in the lateral intraparietal area during spatial and shape-delayed match-to- sample tasks. Journal of Neurophysiology, 95, 1078–1098.

Shadlen, M. N., & Newsome, W. T. (1996). Motion perception: seeing and deciding (pp. 628–633). Presented at the PNAS.

Shadlen, M. N., & Newsome, W. T. (2001). Neural Basis of a Perceptual Decision in the Parietal Cortex (Area LIP) of the Rhesus Monkey. Journal of Neurophysiology, 86(4), 1916–1936.

Shadlen, M. N., Britten, K. H., Newsome, W. T., & Movshon, J. A. (1996). A computational analysis of the relationship between neuronal and behavioral responses to visual motion. The Journal of Neuroscience, 16(4), 1486–1510.

Shaki, S., & Fischer, M. (2008). Reading space into numbers-a cross-linguistic comparison of the SNARC effect. Cognition, 108(2), 590–599.

Shaki, S., & Fischer, M. H. (2008). Reading space into numbers-a cross-linguistic comparison of the SNARC effect. Cognition, 108(2), 590–599.

Shaki, S., Fischer, M. H., & Petrusic, W. M. (2009). Reading habits for both words and numbers contribute to the SNARC effect. Psychonomic Bulletin & Review, 16(2), 328– 331.

Shaki, S., Fischer, M. H., & Petrusic, W. M. (2009). Reading habits for both words and

307

numbers contribute to the SNARC effect. Psychonomic Bulletin & Review, 16(2), 328– 331. doi:10.3758/PBR.16.2.328

Sheppard, J. P., Raposo, D., & Churchland, A. K. (2013). Dynamic weighting of multisensory stimuli shapes decision-making in rats and humans. Journal of Vision, 13(6), 4–4. http://doi.org/10.1167/13.6.4

Shohamy, D., & Daw, N. D. (2015). Integrating memories to guide decisions. Current Opinion in Behavioral Sciences.

Siegler, R. S., Fazio, L. K., Bailey, D. H., & Zhou, X. (2013). Fractions: the new frontier for theoriesof numerical development. Trends in Cognitive Sciences, 17(1), 13–19. http://doi.org/10.1016/j.tics.2012.11.004

Siemann, J. K., Muller, C. L., Bamberger, G., Allison, J. D., Veenstra-VanderWeele, J., & Wallace, M. T. (2014). A novel behavioral paradigm to assess multisensory processing in mice. Frontiers in Behavioral Neuroscience, 8. http://doi.org/10.3389/fnbeh.2014.00456

Silva, K., Shulman, E. P., Chein, J., & Steinberg, L. (2015). Peers Increase Late Adolescents' Exploratory Behavior and Sensitivity to Positive and Negative Feedback. Journal of Research on Adolescence, n/a–n/a. http://doi.org/10.1111/jora.12219

Sip, K. E., Lynge, M., Wallentin, M., McGregor, W. B., Frith, C. D., & Roepstorff, A. (2010). The production and detection of deception in an interactive game. Neuropsychologia, 48, 3619–3626.

Slotema, C. W., Blom, J. D., van Lutterveld, R., Hoek, H. W., & Sommer, I. E. C. (2014). Review of the Efficacy of Transcranial Magnetic Stimulation for Auditory Verbal Hallucinations. Biological Psychiatry, 76(2), 101–110. http://doi.org/10.1016/j.biopsych.2013.09.038

Smirnova, A., Zorina, Z., Obozova, T., & Wasserman, E. (2015). Crows Spontaneously Exhibit Analogical Reasoning. Current Biology, 25(2), 256–260.

Smith, P. L., & Ratcliff, R. (2004). Psychology and neurobiology of simple decisions. Trends in Neurosciences, 27(3), 161–168. http://doi.org/10.1016/j.tins.2004.01.006

Snyder, L. H., Batista, A. P., & Andersen, R. A. (1997). Coding of intention in the posterior parietal cortex. Nature.

Sodian, B., & Frith, U. (1992). Deception and sabotage in autistic, retarded and normal

308

children. Journal of Child Psychology and Psychiatry.

Sommer, M.A., and Wurtz, R.H. (2006). Influence of the thalamus on spatial visual processing in frontal cortex. Nature 444, 374–377.

Spence, S. A., Farrow, T., Herford, A. E., & Wilkinson, I. D. (2001). Behavioural and functional anatomical correlates of deception in humans. NeuroReport, 12(13), 2849- 2853.

Spence, S. A., Kaylor-Hughes, C., Farrow, T. F. D., & Wilkinson, I. D. (2008). Speaking of secrets and lies: The contribution of ventrolateral prefrontal cortex to vocal deception. NeuroImage, 40(3), 1411–1418. http://doi.org/10.1016/j.neuroimage.2008.01.035

Spinillo, A. G., & Bryant, P. E. (1999). Proportional reasoning in young children: Part- part comparisons about continuous and discontinuous quantity. Mathematical Cognition, 5(2), 181–197.

Stancher, G., Sovrano, V.A., Potrich, D., & Vallortigara, G. (2013). Discrimination of small quantities by fish (redtail splitfin, Xenotoca eiseni). Animal Cognition, 16(2), 307- 312.

Starr, A., Libertus, M. E., & Brannon, E. M. (2013). Number sense in infancy predicts mathematical abilities in childhood. Proceedings of the National Academy of Sciences of the United States of America, 110(45), 18116–18120. http://doi.org/10.1073/pnas.1302751110

Steele, M. A., Halkin, S. L., Smallwood, P. D., & McKenna, T. J. (2008). Cache protection strategies of a scatter-hoarding rodent: do tree squirrels engage in behavioural deception? Animal Behaviour.

Steinbeis, N., Bernhardt, B. C., & Singer, T. (2012). Impulse Control and Underlying Functions of the Left DLPFC Mediate Age-Related and Age-Independent Individual Differences in Strategic Social Behavior. Neuron, 73(5), 1040–1051. http://doi.org/10.1016/j.neuron.2011.12.027

Stepanek, L., & Doupe, A. J. (2010). Activity in a cortical-basal ganglia circuit for song is required for social context-dependent vocal variability. Journal of Neurophysiology, 104(5), 2474–2486. http://doi.org/10.1152/jn.00977.2009

Stevens, M. (2007). Predator perception and the interrelation between different forms of protective coloration. Proceedings of the Royal Society B: Biological Sciences, 274(1617),

309

1457–1464. http://doi.org/10.1098/rspb.2007.0220

Stewart, L. M., Walsh, V., & Rothwell, J. C. (2001). Motor and phosphene thresholds: a transcranial magnetic stimulation correlation study. Neuropsychologia, 39, 415–419.

Suanda, S. H., Tompson, W., & Brannon, E. M. (2008). Changes in the Ability to Detect Ordinal Numerical Relationships Between 9 and 11 Months of Age. Infancy, 13(4), 308–337. http://doi.org/10.1080/15250000802188800

Sugrue, L. P., Corrado, G. S., & Newsome, W. T. (2004). Matching Behavior and the Representation of Value in the Parietal Cortex. Science, 304(5678), 1782–1787. http://doi.org/10.1126/science.1094765

Summers, K. (2014). Sexual conflict and deception in poison frogs. Current Zoology.

Suzuki, K., & Kobayashi, T. (2000). Numerical competence in rats (Rattus norvegicus): Davis and Bradford (1986) extended. Journal of Comparative Psychology, 114(1), 73–85. doi:10.1037//0735-7036.114.1.73

Suzuki, Mototaka, & Gottlieb, J. (2013). Distinct neural mechanisms of distractor suppression in the frontal and parietal lobe. Nature Neuroscience, 16(1), 98–104. http://doi.org/10.1038/nn.3282

Swaminathan, S. K., & Freedman, D. J. (2012). Preferential encoding of visual categories in parietal cortex compared with prefrontal cortex. Nature Neuroscience, 15, 315–320.

Swaminathan, S. K., Masse, N. Y., & Freedman, D. J. (2013). A Comparison of Lateral and Medial Intraparietal Areas during a Visual Categorization Task. Journal of Neuroscience, 33(32), 13157–13170. http://doi.org/10.1523/JNEUROSCI.5723-12.2013

Takayama, Y., Sugishita, M., Akiguchi, I., & Kimura, J. (1994). Isolated Acalculia due to Left Parietal Lesion. Archives of Neurology, 51(3), 286–291. http://doi.org/10.1001/archneur.1994.00540150084021

Tang, A., Thickbroom, G., & Rodger, J. (2015). Repetitive Transcranial Magnetic Stimulation of the Brain: Mechanisms from Animal and Experimental Models. The Neuroscientist : a Review Journal Bringing Neurobiology, Neurology and Psychiatry. http://doi.org/10.1177/1073858415618897

Tang, J., Ward, J., & Butterworth, B. (2008). Number forms in the brain. Journal of Cognitive Neuroscience, 20(9), 1547–1556. doi:10.1162/jocn.2008.20120

310

Tehovnik, E.J., Tolias, A.S., Sultan, F., Slocum, W.M., and Logothetis, N.K. (2006). Direct and Indirect Activation of Cortical Neurons by Electrical Microstimulation. J. Neurophysiol. 96, 512–521.

Teo, J. T. H., Swayne, O. B., & Rothwell, J. C. (2007). Further evidence for NMDA- dependence of the after-effects of human theta burst stimulation. Clinical Neurophysiology, 118(7), 1649–1651. http://doi.org/10.1016/j.clinph.2007.04.010

Terrace, H. (2005). The simultaneous chain: A new approach to serial learning. Trends in Cognitive Sciences, 9(4), 202–210.

Terrace, H. S. (1993). The phylogeny and ontogeny of serial memory: List learning by pigeons and monkeys. Psychological Science, 4(3), 162–169.

Terrace, H. S., Chen, S., & Jaswal, V. (1996). Recall of three-item sequences by pigeons. Learning & Behavior, 24(2), 193–205.

Terrace, H., Son, L., & Brannon, E. (2003). Serial expertise of rhesus macaques. Psychological Science, 14(1), 66.

Théry, M., & Casas, J. (2002). Visual systems: Predator and prey views of spider camouflage. Nature, 415(6868), 133–133. http://doi.org/10.1038/415133a

Thompson, C. A., & Opfer, J. E. (2008). Costs and benefits of representational change: Effects of context on age and sex differences in symbolic magnitude estimation. Journal of Experimental Child Psychology, 101(1), 20–51. http://doi.org/10.1016/j.jecp.2008.02.003

Thompson, R. K. R., & Oden, D. L. (2000). Categorical perception and conceptual judgments by nonhuman primates: the paleological monkey and the analogical ape. Cognitive Science, 24(3), 363–396. http://doi.org/10.1016/S0364-0213(00)00029-X

Thomson, R. H., Cleve, T. J., Bailey, N. W., Rogasch, N. C., Maller, J. J., Daskalakis, Z. J., & Fitzgerald, P. B. (2013). Blood oxygenation changes modulated by coil orientation during prefrontal transcranial magnetic stimulation. Brain Stimulation, 6(4), 576–581. http://doi.org/10.1016/j.brs.2012.12.001

Thurmon, J.C., Tranquilli, W.J., and Benson, G.J. (1996). Lumb & Jones’ Veterinary Anesthesia. Baltimore, MD: Williams & Wilkins.

Tischler, H., Moran, A., Belelovsky, K., Bronfeld, M., Korngreen, A., & Bar-Gad, I. (2012). Changes in basal ganglia processing of cortical input following magnetic stimulation

311

in Parkinsonism. Neurobiology of Disease, 48(3), 464–473. http://doi.org/10.1016/j.nbd.2012.07.021

Tischler, H., Wolfus, S., Friedman, A., Perel, E., Pashut, T., Lavidor, M., et al. (2011). Mini-coil for magnetic stimulation in the behaving primate. Journal of Neuroscience Methods, 194(2), 242–251. http://doi.org/10.1016/j.jneumeth.2010.10.015

Tosoni, A., Galati, G., Romani, G. L., & Corbetta, M. (2008). Sensory-motor mechanisms in human parietal cortex underlie arbitrary visual decisions. Nature Neuroscience, 11(12), 1446–1453. http://doi.org/10.1038/nn.2221

Truppa, V., Piano Mortari, E., Garofoli, D., Privitera, S., & Visalberghi, E. (2011). Same/Different Concept Learning by Capuchin Monkeys in Matching-to-Sample Tasks. PLoS One, 6(8), e23809. http://doi.org/10.1371/journal.pone.0023809.s003

Tsao, D.Y., Moeller, S., and Freiwald, W.A. (2008). Comparing face patch systems in macaques and humans. Proc. Natl. Acad. Sci. 105, 19514–19519.

Tsunada, J., Liu, A. S. K., Gold, J. I., & Cohen, Y. E. (2015). Causal contribution of primate auditory cortex to auditory perceptual decision-making. Nature Publishing Group, 19(1), 135–142. http://doi.org/10.1038/nn.4195

Turconi, E., & Seron, X. (2002). Dissociation Between Order and Quantity Meanings in a Patient with Gerstmann Syndrome. Cortex, 38(5), 911–914. doi:10.1016/S0010- 9452(08)70069-8

Turconi, E., Campbell, J., & Seron, X. (2006). Numerical order and quantity processing in number comparison. Cognition, 98(3), 273–285.

Turconi, E., Jemel, B., Rossion, B., & Seron, X. (2004). Electrophysiological evidence for differential processing of numerical quantity and order in humans. Cognitive Brain Research, 21(1), 22–38.

Uller, C. (2008). Developmental and evolutionary considerations on numerical cognition: a review. Journal of Evolutionary Psychology, 6(4), 237–254. http://doi.org/10.1556/jep.6.2008.4.7

Uller, C., & Lewis, J. (2009). Horses (Equus caballus) select the greater of two quantities in small numerical contrasts. Animal Cognition, 12(5), 733–738. doi:10.1007/s10071- 009-0225-0

Uller, C., Jaeger, R., & Guidry, G. (2003). Salamanders (Plethodon cinereus) go for more:

312

rudiments of number in an amphibian. Animal Cognition.

Utevsky, A. V., & Huettel, S. A. (2015). Social Decision Making. In A. W. Toga (Ed.), Brain Mapping: An Encyclopedic Reference (pp. 231–234). Waltham, MA: Academic Press.

Valero-Cabre, A., Wattiez, N., Monfort, M., François, C., Rivaud-Péchoux, S., Gaymard, B., and Pouget, P. (2012). Frontal Non-Invasive Neurostimulation Modulates Antisaccade Preparation in Non-Human Primates. PLoS ONE 7, e38674.

Vallentin, D., & Nieder, A. (2008). Behavioral and prefrontal representation of spatial proportions in the monkey. Current Biology, 18(18), 1420–1425.

Vallentin, D., & Nieder, A. (2010). Representations of visual proportions in the primate posterior parietal and prefrontal cortices. European Journal of Neuroscience, 32(8), 1380–1387. http://doi.org/10.1111/j.1460-9568.2010.07427.x

Vallesi, A., Binns, M. A., & Shallice, T. (2008). An effect of spatial--temporal association of response codes: Understanding the cognitive representations of time. Cognition, 107(2), 501–527. http://doi.org/10.1016/j.cognition.2007.10.011

Vallortigara, G. (2012). Core knowledge of object, number, and geometry: A comparative and neural approach. Cognitive Neuropsychology, 29(1-2), 213–236. http://doi.org/10.1080/02643294.2012.654772 van Dijck, J.-P., & Fias, W. (2011). A working memory account for spatial-numerical associations. Cognition, 119(1), 114-119. van Dijk, K. D., Møst, E. I. S., Van Someren, E. J. W., Berendse, H. W., & van der Werf, Y. D. (2009). Beneficial effect of transcranial magnetic stimulation on sleep in Parkinson's disease. Movement Disorders, 24(6), 878–884. http://doi.org/10.1002/mds.22462 van Honk, J., Schutter, D. J. L. G., Putman, P., de Haan, E. H. F., & d'Alfonso, A. A. L. (2003). Reductions in phenomenological, physiological and attentional indices of depressive mood after 2 Hz rTMS over the right parietal cortex in healthy human subjects. Psychiatry Research, (120), 95–101. http://doi.org/10.1016/S0165- 1781Ž03.00114-8

Vartanian, O., Kwantes, P. J., Mandel, D. R., Bouak, F., Nakashima, A., Smith, I., & Lam, Q. (2013). Right inferior frontal gyrus activation as a neural marker of successful lying. Frontiers in Human Neuroscience, 7, 1–6.

313

http://doi.org/10.3389/fnhum.2013.00616

Veniero, D., Vossen, A., Gross, J., & Thut, G. (2015). Lasting EEG/MEG Aftereffects of Rhythmic Transcranial Brain Stimulation: Level of Control Over Oscillatory Network Activity. Frontiers in Cellular Neuroscience, 9(13), 11262. http://doi.org/10.3389/fncel.2015.00477

Vighneshvel, T., & Arun, S. P. (2015). Coding of relative size in monkey inferotemporal cortex. Journal of Neurophysiology, 113(7), 2173–2179. http://doi.org/10.1152/jn.00907.2014

Vlachos, A., Muller-Dahlhaus, F., Rosskopp, J., Lenz, M., Ziemann, U., & Deller, T. (2012). Repetitive Magnetic Stimulation Induces Functional and Structural Plasticity of Excitatory Postsynapses in Mouse Organotypic Hippocampal Slice Cultures. Journal of Neuroscience, 32(48), 17514–17523. http://doi.org/10.1523/JNEUROSCI.0409- 12.2012

Volz, K. G., Vogeley, K., Tittgemeyer, M., Cramon, von, D. Y., & Sutter, M. (2015). The neural basis of deception in strategic interactions. Frontiers in Behavioral Neuroscience, 9, 1–12. http://doi.org/10.3389/fnbeh.2015.00027

Vonk, J. (2003). Gorilla (Gorilla gorilla gorilla) and orangutan (Pongo abelii) understanding of first- and second-order relations. Animal Cognition, 6(2), 77–86. http://doi.org/10.1007/s10071-003-0159-x

Vonk, J., & Beran, M. J. (2012). Bears `count' too: quantity estimation and comparison in black bears, Ursus americanus. Animal Behaviour, 1–8. doi:10.1016/j.anbehav.2012.05.001

Wang, X.-J. (2012). Neural dynamics and circuit mechanisms of decision-making. Current Opinion in Neurobiology, 22(6), 1039–1046. http://doi.org/10.1016/j.conb.2012.08.006

Ward, C., & Smuts, B. B. (2007). Quantity-based judgments in the domestic dog (Canis lupus familiaris). Animal Cognition, 10(1), 71–80. doi:10.1007/s10071-006-0042-7

Wassermann, E. M., & Zimmermann, T. (2012). Transcranial magnetic brain stimulation: Therapeutic promises and scientific gaps. Pharmacology and Therapeutics, 133(1), 98– 107. http://doi.org/10.1016/j.pharmthera.2011.09.003

Werner, E. E., & MittelBach, G. G. (1981). Optimal Foraging: Field Tests of Diet Choice and Habitat Switching. American Zoologist, 21(4), 813–829.

314

http://doi.org/10.1093/icb/21.4.813

West, E. A., DesJardin, J. T., Gale, K., & Malkova, L. (2011). Transient inactivation of orbitofrontal cortex blocks reinforcer devaluation in macaques. The Journal of Neuroscience : the Official Journal of the Society for Neuroscience, 31(42), 15128–15135. http://doi.org/10.1523/JNEUROSCI.3295-11.2011

Whiten, A., & Byrne, R. W. (1988). Tactical deception in primates. Behavioral and Brain Sciences, 11, 233–273. http://doi.org/10.1017/S0140525X00049682

Wilson, M. L., Britton, N. F., & Franks, N. R. (2002). Chimpanzees and the mathematics of battle. Proceedings of the Royal Society B: Biological Sciences, 269(1496), 1107–1112. http://doi.org/10.1098/rspb.2001.1926

Wischnewski, M., & Schutter, D. J. L. G. (2015). Efficacy and time course of theta burst stimulation in healthy humans. Brain Stimulation: Basic, Translational, and Clinical Research in Neuromodulation, 0(0), –. http://doi.org/10.1016/j.brs.2015.03.004

Wise, R. A., & Rompré, P. P. (1989). Brain dopamine and reward. Annual Review of Psychology. http://doi.org/10.2307/25464368?ref=search- gateway:d9b9e4039b3fa0abe19294d679534667

Wong, K. F., & Wang, X.-J. (2006). A Recurrent Network Mechanism of Time Integration in Perceptual Decisions. Journal of Neuroscience, 26(4), 1314–1328. http://doi.org/10.1523/JNEUROSCI.3733-05.2006

Wong, K. F., Huk, A. C., & Shadlen, M. N. (2007). Neural circuit dynamics underlying accumulation of time-varying evidence during perceptual decision making. Frontiers in Computational Neuroscience.

Woodruff, G., & Premack, D. (1979). Intentional communication in the chimpanzee: The development of deception. Cognition.

Woodruff, G., & Premack, D. (1981). Primative mathematical concepts in the chimpanzee: proportionality and numerosity. Nature, 293, 568–570.

Woolley, S. C., Rajan, R., Joshua, M., & Doupe, A. J. (2014). Emergence of Context- Dependent Variability across a Basal Ganglia Network. Neuron, 82(1), 208–223. http://doi.org/10.1016/j.neuron.2014.01.039

Wurtz, R. H., & Goldberg, M. E. (1971). Superior Colliculus Cell Responses Related to Eye Movements in Awake Monkeys. Science, 171(3966), 82–84.

315

http://doi.org/10.1126/science.171.3966.82

Wurtz, R.H., and Sommer, M.A. (2006). "Single Neurons and Primate Behavior," in Methods in Mind, ed. C. Senior, T. Russell, & M. S. Gazzaniga (Cambridge: MIT Press), pp. 123–139.

Wyttenbach, R. A., May, M. L., & Hoy, R. R. (1996). Categorical perception of sound frequency by crickets. Science, 273(5281), 1542–1544. http://doi.org/10.1126/science.273.5281.1542

Yin, L., Reuter, M., & Weber, B. (2016). Let the man choose what to do: Neural correlates of spontaneous lying and truth-telling. Brain and Cognition.

Yirmiya, N., Solomonica-Levi, D., & Shulman, C. (1996). The ability to manipulate behavior and to understand manipulation of beliefs: A comparison of individuals with autism, mental retardation, and normal development. Developmental ….

Yoshida, W., Dolan, R. J., & Friston, K. J. (2008). Game Theory of Mind. PLoS Computational Biology, 4(12), e1000254. http://doi.org/10.1371/journal.pcbi.1000254

Zangen, A., Roth, Y., Voller, B., & Hallett, M. (2005). Transcranial magnetic stimulation of deep brain regions: evidence for efficacy of the H-Coil. Clinical Neurophysiology, 116(4), 775–779. http://doi.org/10.1016/j.clinph.2004.11.008

Zebian, S. (2005). Linkages between number concepts, spatial thinking, and directionality of writing: The SNARC effect and the reverse SNARC effect in English and Arabic monoliterates, biliterates, and illiterate Arabic speakers. Journal of Cognition and Culture, 5, 1(2), 165–190.

Zeki, S. M. (1974). Functional organization of a visual area in the posterior bank of the superior temporal sulcus of the rhesus monkey. The Journal of Physiology, 236(3), 549.

Zentall, T. R., Wasserman, E. A., Lazareva, O. F., Thompson, R. K. R., & Rattermann, M. J. (2008). Concept learning in animals. Comparative Cognition & Behavior Reviews, 3, 13–45. http://doi.org/10.3819/ccbr.2008.30002

Zhu, L., Jenkins, A. C., Set, E., Scabini, D., Knight, R. T., Chiu, P. H., et al. (2014). Damage

316

to dorsolateral prefrontal cortex affects tradeoffs between honesty and self-interest. Nature Neuroscience, 17(10), 1319–1321. http://doi.org/10.1038/nn.3798

Zorzi, M., Di Bono, M. G., & Fias, W. (2011). Distinct representations of numerical and non-numerical order in the human intraparietal sulcus revealed by multivariate pattern recognition. NeuroImage, 56(2), 674–680. doi:10.1016/j.neuroimage.2010.06.035

317

Biography

Caroline Beth Drucker was born on January 2, 1989, in Ridgewood, New Jersey, to Sharon Stern Drucker and David Drucker. She and her older sister Eve grew up in

Fair Lawn, New Jersey, and attended high school at the Bergen County Academies.

Caroline received a BS with Distinction in Psychology (Neuroscience Track) under the mentorship of Dr. Laurie Santos and a BA in Mathematics from Yale College in New

Haven, Connecticut, in May 2011. She entered graduate school through the Cognitive

Neuroscience Admitting Program at Duke University in 2011. Caroline is the recipient of a National Science Foundation Graduate Research Fellowship and the Holland-Trice

Graduate Fellowship in Brain Science and Disease, as well as a member of Phi Beta

Kappa. Her published journal articles are listed below; she is also preparing the study presented in section 3 of this dissertation for publication.

1) Drucker, C.B., Baghdoyan, T., & Brannon, E.M. (2016) Implicit sequence learning by ring-tailed lemurs (Lemur catta). Journal of the Experimental Analysis of Behavior, 105(1), 123-132. 2) Drucker, C.B., Rossa, M.A., & Brannon, E.M. (2016) Comparison of discrete ratios by rhesus macaques (Macaca mulatta). Animal Cognition, 19(1), 75-89. 3) Drucker, C.B., Carlson, M.L., Toda, K., DeWind, N.K., & Platt, M.L. (2015) Non-invasive primate head restraint using thermoplastic masks. Journal of Neuroscience Methods, 250, 90-100. 4) Drucker, C.B., & Brannon, E.M. (2015) Commentary on: “Number-space mapping in the newborn chick resembles humans’ mental number line”. Frontiers in Psychology, 6:352. 5) Drucker, C.B., & Brannon, E.M. (2014) Rhesus monkeys (Macaca mulatta) map number onto space. Cognition, 132, 57-67. 6) Westerman, E.L., Drucker, C.B., & Monteiro, A. (2014) Male and female mating behavior is dependent on social context in the butterfly Bicyclus anynana. Journal of Insect Behavior, 27:4, 478-495.

318