Deficits in prefrontal cortical and extra-striatal dopamine release in schizophrenia: a PET 1

fMRI study 2

3

1,3Mark Slifstein*, PhD, 1,3Elsmarieke van de Giessen*, MD PhD, 1,3Jared Van 4

Snellenberg PhD, 1,3,8Judy L. Thompson PhD, 7Rajesh Narendran MD, 1,3Roberto Gil 5

MD, 3Elizabeth Hackett RT, 1,3Ragy Girgis MD, 3Najate Ojeil MS, 1,3Holly Moore PhD, 6

6Deepak D’Souza MD, 6Robert T. Malison MD, 4,5Yiyun Huang PhD, 4,5Keun-poong Lim 7

PhD, 4,5Nabeel Nabulsi PhD, 4,5Richard E. Carson PhD, 1,3Jeffery A. Lieberman MD, 8

1,2,3Anissa Abi-Dargham MD 9

1Columbia University, Department of Psychiatry 10

2Columbia University, Department of Radiology 11

3New York State Psychiatric Institute 12

4Yale University School of Medicine PET Center 13

5Yale University School of Medicine Department of Diagnostic Radiology 14

6Yale University School of Medicine Department of Psychiatry 15

7University of Pittsburgh Medical Center Department of Psychiatry 16

8The State University of New Jersey, Rutgers 17

* these authors contributed equally to this study 18

Word Count: Abstract: 348, Main Body: 3310 19

20

Address correspondence to: 21

Mark Slifstein PhD 22 1051 Riverside Drive, Unit 31 23 New York, NY, 10032 24 Email mms218@columbia.edu 25 telephone 646 774 5563 26 27

28

29

30

Abstract 31

32 Importance: Multiple lines of evidence suggest a deficit in dopamine release in 33

prefrontal cortex in schizophrenia. Despite the prevalence of the concept of prefrontal 34

cortical hypodopaminergia in schizophrenia, in vivo imaging of dopamine release in 35

prefrontal cortex has not been possible until recently, when the validity of using the PET 36

D2/3 radiotracer [11C]FLB457 in combination with the amphetamine paradigm was 37

clearly established. 38

39

Objectives: 1) To test amphetamine induced dopamine release in dorsolateral prefrontal 40

cortex (DLPFC) in drug free or drug naïve patients with schizophrenia (SCZ) and healthy 41

controls (HC) matched for age, gender, ethnicity and familial socioeconomic status, 2) to 42

test BOLD fMRI activation during a working memory task in the same subjects and 3) to 43

examine the relationship between PET and fMRI outcome measures. 44

Design, Setting and Participants: PET imaging with [11C]FLB457 before and following 45

0.5 mg/kg P.O. amphetamine. BOLD fMRI during the self-ordered working memory task 46

(SOWT). 20 patients with schizophrenia and 21 healthy controls participated. 47

Main outcome measure: The percent change in binding potential (∆BPND) in DLPFC 48

following amphetamine, BOLD activation during the SOWT compared to the control task, 49

and the correlation between these two outcome measures. 50

Results: We observed: 1) significant differences in the effect of amphetamine on 51

DLPFC BPND (∆BPND in HC: - 7.5 ± 11%, SCZ: +1.8 ± 11%, p = 0.013), 2) a 52

generalized blunting in dopamine release in SCZ involving most extrastriatal regions and 53

the midbrain, 3) a significant relationship between ∆BPND and BOLD activation in 54

DLPFC in the overall sample including patients with SCZ and HC. 55

Conclusions and Relevance: These results provide the first in vivo evidence for a deficit 56

in the capacity for dopamine release in DLPFC in schizophrenia, and suggest a more 57

widespread deficit extending to many cortical and extrastriatal regions, including the 58

midbrain. This contrasts with the well-replicated excess in dopamine release in the 59

associative striatum in schizophrenia, and suggests a differential regulation of striatal 60

dopamine release in associative striatum versus extrastriatal regions. Furthermore, 61

dopamine release in the DLPFC relates to working memory-related activation of this 62

region, suggesting that blunted release may affect frontal cortical function. 63

64

65

Introduction 66

67

The concept of cortical hypodopaminergia in schizophrenia1 has emerged from 68

converging lines of evidence showing that working memory (WM) is deficient in 69

schizophrenia2, that WM depends critically on optimal prefrontal dopamine (DA) 70

transmission in non-human primates3-10

, that it is associated with abnormal prefrontal 71

activation during functional brain imaging studies in schizophrenia11

, and that it can 72

improve with DA agonists12-15

. Furthermore, post-mortem studies reported a decrease 73

in tyrosine hydroxylase immunolabeling in prefrontal cortex in schizophrenia16-18. While 74

Positron Emission Tomography (PET) studies have investigated alterations in cortical D1 75

receptor availability19-21, there have been no in vivo studies examining capacity for DA 76

release in frontal cortex in schizophrenia, a gap that contrasts with the considerable 77

body of evidence from in vivo PET imaging studies showing an increase in stimulant-78

induced DA release in the striatum of patients with schizophrenia22-24. 79

One major impediment to PET studies of cortical DA release has been the lack of a 80

suitable PET radiotracer. For reasons that are not completely understood, D1 81

radiotracers have not proven to be sensitive to stimulant-induced DA release 25 whereas 82

D2/D3 tracers have. While radiotracers such as [11C]raclopride and [11C]-(+)-PHNO are 83

useful for detecting acute fluctuations in DA levels in the striatum, the very low density 84

and limited anatomical distribution of DA D2/D3 receptors in cortex26 precludes their use 85

for quantitative imaging of D2/D3 receptors in the cortex. [11C]FLB457 is a higher-86

affinity PET tracer that has been shown to provide reliable quantification of 87

amphetamine-induced DA release in cortex27,28 (test-retest reproducibility ≤ 15% using 88

conventional compartment analysis methods), although it cannot be quantified in 89

striatum due to its slow washout in this high D2/D3 receptor density region. However, 90

there are challenges in working with this tracer. Most D2/D3 tracers show negligible 91

specific binding in the cerebellum, allowing the use of the cerebellum as a reference 92

region29. This is not the case for [11C]FLB457, as approximately 20% of [11C]FLB457 93

cerebellum distribution volume VT can be displaced by the D2 partial agonist 94

aripiprazole30. 95

In the current study, we measured amphetamine-induced DA release in the dorsolateral 96

prefrontal cortex (DLPFC) in patients with schizophrenia (SCZ) and matched healthy 97

controls (HC) using [11C]FLB457 PET imaging. We implemented a kinetic model with 98

shared parameters across 9 cortical regions, that addressed both the lack of a reference 99

region and the low cortical signal, to quantify receptor availability and DA release. We 100

hypothesized that cortical DA release capacity, especially in the dorsolateral prefrontal 101

cortex (DLPFC) would be reduced in SCZ compared to HC. We also examined a 102

number of brain regions where D2/D3 receptor availability is intermediate between 103

striatal and cortical binding, including midbrain (substantia nigra, ventral tegmental area), 104

thalamus, and medial temporal regions (amygdala, hippocampus). To test the functional 105

significance of cortical DA release capacity, we used functional MRI (fMRI) to measure 106

changes in the BOLD signal in the DLPFC during performance of the Self-Ordered 107

Working Memory Task (SOWT), and examined associations between cortical DA 108

release capacity and WM task-related DLPFC activation. Finally, we examined the 109

relationships between [11C]FLB457 PET and WM-sensitive performance in SCZ and HC 110

subjects, and clinical symptomatology in patients. 111

112

Materials and Methods 113

114

Subjects 115

This study was approved by the Institutional Review Boards of the New York State 116

Psychiatric Institute (NYSPI) of Columbia University Medical Center (CUMC) and the 117

Yale University Human Investigation Committee. All participants provided written 118

informed consent following an independent assessment of capacity by a psychiatrist who 119

was not a member of the research team. Patients were recruited from the inpatient and 120

outpatient research facilities at NYSPI. Healthy controls were recruited through 121

advertisements. Medical screening procedures included a physical examination and 122

history, blood and urine tests, an electrocardiogram and a structural magnetic resonance 123

imaging scan of the brain. 124

Inclusion criteria for patients were: (1) lifetime DSM-IV (Diagnostic and Statistical Manual 125

of Mental Disorders, 4th ed.) diagnosis of schizophrenia, schizoaffective or 126

schizophreniform disorder; (2) no bipolar disorder; (3) no antipsychotics for 3 weeks prior 127

to the PET scan; (5) no history of violent behavior. Inclusion criteria for healthy controls 128

were: (1) absence of any current or past DSM-IV Axis-I diagnosis; and (2) no family 129

history (first-degree) of psychotic illness. 130

Exclusion criteria for both groups: significant medical and neurological illnesses, current 131

misuse of substances other than nicotine, positive urine drug screen, pregnancy and 132

nursing. Groups were matched for age, gender, ethnicity, parental socioeconomic status, 133

and nicotine smoking (Table 1). 134

PET Imaging Study Design 135

Subjects underwent two PET scans on one day with [11C]FLB457 at the Yale University 136

PET Center. A 90 min baseline scan was acquired, followed immediately by oral 137

administration of amphetamine (0.5 mg/kg) and a second 90 min scan 3 hr after 138

amphetamine administration. Arterial plasma data were collected to form metabolite-139

corrected input functions. Data were acquired on an HR+ scanner (Siemens, Knoxville 140

TN) and reconstructed by filtered back projection with correction for attenuation, 141

randoms and scatter. Data were binned into a sequence of frames of increasing 142

duration. 143

PET Data Analysis 144

Preprocessing. A high resolution T1-weighted MRI scan was acquired for each subject. 145

Regions of interest (ROIs) were drawn on each subject’s MRI according to previously 146

described criteria 20,31,32 (See eMethods.1 for operational definitions of amygdala and 147

hippocampus), and included, in addition to the DLPFC, our a priori ROI, the medial 148

frontal cortex (MFC), orbito-frontal cortex (OFC), anterior cingulate (A. CING), occipital 149

cortex (OCC), parietal cortex (PAR), temporal cortex (TEMP), sub-genu of the cingulate 150

(GEN), insula, cerebellum, and 5 subcortical regions: amygdala (AMYG), hippocampus 151

(HIPP), midbrain encompassing substantia nigra and ventral tegmental area (SN/VTA), 152

thalamus (THAL) and uncus. PET data were coregistered to the MRI data using 153

normalized maximization of mutual information (SPM8) and the ROIs were transferred 154

to the coregistered PET using MEDx software (Medical Numerics, Germantown MD). 155

Time activity curves were generated as the average activity in each frame for each ROI. 156

Kinetic Analysis. Data were analyzed with a two tissue compartment model 29,33 (2TC) 157

that additionally incorporated a set of shared parameter estimates across regions, in 158

order to improve reliability of fits by estimating a reduced parameter set compared to 159

conventional 2TC. For each subject, VND, the distribution volume of the nondisplaceable 160

compartment, and k4, the specific binding dissociation constant, were fitted to a single 161

value across cortical regions for both baseline and post-amphetamine scans (See 162

eMethods.2, eTable.1, eFigure.1). The parameters K1, the brain delivery constant and 163

k3, proportional to receptor availability, were fitted in each region and condition. Data 164

were weighted by frame duration; all regions were weighted equally. The same 165

procedure was applied separately to the higher binding subcortical regions, in order to 166

allow for the possibility that the fitting procedure might assign different k4 values in those 167

regions. Distribution volume (VT) was estimated in each region and condition. Binding 168

potential (BPND) was estimated directly from the ratio k3/k4. We report BPND, the 169

relative change following amphetamine (∆BPND), VT and ∆VT. 170

Statistics. In DLPFC, two group t tests were applied to baseline BPND, ∆BPND, baseline 171

VT and ∆VT. Additionally, linear mixed modeling with ROI as repeated measure and 172

group and ROI as fixed variables was applied across all 14 regions. Two-sided t tests 173

were applied to scan parameters including injected activity, injected mass, plasma free 174

fraction (fp) and estimated VND and k4. Parameter estimates are reported as mean ± 175

standard deviation. 176

fMRI Data Analysis 177

A subset of 16 SCZ and 18 HC participated in a BOLD fMRI study in which they 178

performed the SOWT (among the 20 SCZ and 21 HC subjects, 4 SCZ and 2 HC 179

declined to participate in the fMRI procedures, and one HC’s data were unusable due to 180

poor image quality). Complete details of the task as well as acquisition and analysis are 181

in eMethods.3. Briefly, structural and BOLD images were acquired on a Philips 1.5 T 182

Intera scanner. BOLD images during SOWT task performance were acquired at a 3 mm 183

isotropic voxel size with a TR of 2 s, separated into 9 runs of 160 volumes each. BOLD 184

images underwent slice-timing correction, motion realignment, and coregistration to the 185

T1-weighted structural images. A separate set of BOLD images were normalized to the 186

ICBM template for voxelwise statistical analysis.. The SOWT task consists of a 187

presentation of 8 different geometric shapes on a projection screen. Subjects select one 188

of the shapes; on each successive trial, the positions of shapes on the screen are 189

randomly reordered, and subjects are instructed to select a shape they haven’t picked 190

previously. Thus subjects are required to hold up to 7 distinct items in working 191

memory. There was a monetary incentive for correct answers ($0.25 per correct 192

response). Complete analysis of the BOLD response to the SOWT will be presented 193

elsewhere (Van Snellenberg et al, in submission). Here, we report only the relationship 194

between BOLD and PET data. We regressed DLPFC ∆BPND against overall BOLD 195

activation in DLPFC voxels that were significantly activated during the SOWT. A second-196

level TASK - CONTROL contrast was calculated on the ICBM-normalized BOLD data to 197

identify voxels showing significant activation during SOWT performance. This group-198

level map was transformed to the individual subjects’ T1 space and the intersection of 199

the activated region with their DLPFC ROI was used to extract BOLD signal change 200

values within DLPFC for each participant. We used this approach to restrict analysis to 201

DLPFC voxels that showed evidence of involvement in the SOWT. 202

Neurocognitive and Clinical Measures 203

Diagnostic status was determined with the Diagnostic Interview for Genetic Studies34 in 204

patients with SCZ followed by a consensus diagnosis conference and an abbreviated 205

version of the Structured Clinical Interview for DSM-IV Axis I disorders35 in HC. Severity 206

of symptoms was assessed with the Positive and Negative Symptoms Scale (PANSS)36, 207

obtained at the start of the PET day. Participant and parental socio-economic status 208

were calculated according to the Hollingshead scale37. Clinical assessments were 209

administered by trained interviewers. 210

As additional measures of WM, we also assessed performance on the N-back task38 and 211

the Letter Number Span (LNS)39. Both tasks were acquired once on the day preceding 212

the PET scans and a second time after the second PET scan. The n-back task 213

contained three levels of difficulty, including 1, 2 and 3 back. Adjusted hit rate (AHR, the 214

percent of properly identified targets corrected for false positives, see eMethods.4) was 215

assessed as in Abi-Dargham et al20, and ranged from a maximum possible score of 1 for 216

perfect performance to -1 if all true targets were missed and all non-targets were 217

incorrectly identified as targets. 218

219

220

221

Results 222

223

PET Scan Parameters 224

Injected activity, injected mass, plasma free fraction fp for baseline and amphetamine 225

conditions and estimated VND and k4 are shown in Table 3. There were no significant 226

group differences in any of these. 227

PET Results 228

Baseline DLPFC BPND did not differ significantly between groups but ∆BPND did (HC: - 229

7.5 ± 11%, SCZ: +1.8 ± 11%, p = 0.013) (Table 4). Linear mixed modeling of ∆BPND 230

showed a statistically significant effect of group (F(1,39) = 6.95, p = 0.012) but no 231

significant effect of ROI or group by ROI interaction. While the interaction term was not 232

significant, two regions reached trend level group differences including DLPFC (p = 233

0.087) and SN/VTA (p = 0.096). BPND was higher in drug-naïve than drug-free patients 234

but there were no significant differences in any PET outcome measure when age was 235

included as a covariate (see eResults.1). VT results are shown in eTable2. Baseline 236

DLPFC VT did not differ significantly between groups. There was a significant difference 237

in DLPFC ∆VT (HC: - 5 ± 7%, SCZ +1 ± 7%, p = 0.013). Linear mixed modeling of ∆VT 238

showed a statistically significant effect of group (F(1,39) = 4.11, p = 0.049) but no 239

significant effect of ROI or group by ROI interaction. 240

241

242

Associations with fMRI Activation 243

A significant relationship between DLPFC ∆BPND and BOLD activation in DLPFC was 244

observed in SCZ and HC. Regression of ∆BPND onto BOLD activation had a significant 245

effect of group ( = -10.2, t31 = -2.707; P = 0.0109) and a significant effect of BOLD ( = 246

52.9; t31 = 2.211; P = 0.0345), but no group by BOLD interaction, thus the most 247

parsimonious model contained the same slope for both groups but different intercepts: 248

BPND% decrease (HC) = 53 * BOLD (HC) % increase +6%, BPND% decrease (SCZ) = 249

53 * BOLD (SCZ) % increase – 4% (Figure 1). 250

Associations with Working Memory Performance and Symptoms 251

Patients performed significantly worse on the following WM measures: baseline 1-back 252

and post-amphetamine 2-back task and on baseline and post-amphetamine LNS (Table 253

2). 254

In SCZ, there were no correlations between WM performance and DLPFC BPND, 255

ΔBPND, VT, or ΔVT. In HC, WM performance correlated with DLPFC BPND (baseline 2-256

back, = .50, p = .031, baseline 3-back, = .79, p <.001, SOWT, = .50, p = .031) and 257

VT (baseline 3-back, = .68, p = 0.001) in the DLPFC. Exploratory analyses of 258

correlations at each level of the SOWT revealed significant correlations between DLPFC 259

BPND and WM performance when WM load was greatest (step 7: r = 0.48, p = 0.042, 260

step 8: r = 0.66, p = 0.003). Exploratory analyses including all fourteen ROIs, using a 261

linear mixed model with ROIs as repeated measures resulted in an overall positive 262

association between BPND in the analyzed regions and baseline 2-back (F(1,16.9) = 263

4.99, p = 0.039), 3-back (F(1,16.9) = 14.18, p = 0.002) and SOWT (F(103.4) = 8.08, p = 264

0.005) in HC. The same design applied to VT showed an overall positive association of 265

VT with baseline 3-back (F(1,16.6) = 14.18, p = 0.002) in HC. There were no significant 266

correlations between WM performance and ∆BPND or ∆VT in HC. There were no 267

significant correlations between PANSS scores (Table 2) and BPND, ΔBPND,, VT, or ΔVT.. 268

269

Discussion 270

271

In this study we observed that patients with schizophrenia show blunted amphetamine-272

induced DA release in the DLPFC in vivo. This deficit in DA release extended to other 273

extrastriatal regions including midbrain. We also observed a correlation between this 274

index of DA release capacity and WM-related activation of the DLPFC, as measured 275

with BOLD fMRI. 276

277

Despite the prevalence of the concept of hypodopaminergia in schizophrenia, there had 278

been no empirical evidence for decreased cortical DA release prior to this study. This 279

was related to the difficulty in measuring DA release in the cortex due to the low level of 280

cortical D2 receptors26 and the small range of displacement of D2 radiotracers by DA. 281

Here, we adopted and optimized an [11C]FLB457 displacement paradigm shown to be a 282

valid and reliable proxy for changes in extracellular DA following an amphetamine 283

challenge27,40,41 . 284

Precise quantification of [11C]FLB457 displacement is challenging both because the 285

signal is quite small despite the high affinity of [11C]FLB457, and because the 286

cerebellum cannot be used as a suitable reference tissue. Considering these factors, we 287

developed a kinetic approach that was sensitive enough to detect a small change within 288

a small signal. The shared parameter method we’ve applied here took advantage of the 289

fact that [11C]FLB457 kinetics are similar in many cortical regions, and greater 290

parsimony can be achieved through the dramatic reduction of the number of estimated 291

parameters. In simulations (see eMethods.2), we extensively tested cases in which the 292

underlying assumptions of the shared parameter method - uniform k4 across cortical 293

regions and between scans, uniform VND between scans - were intentionally violated, 294

and found that the method performed more precisely than conventional 2TC. We also 295

note that the average estimated VND is 70% of cerebellum VT, in agreement with 296

Narendran et al30. 297

In the primate PFC, [11C]FLB457 displacement correlates with changes in extracellular 298

DA across doses of amphetamine41. Amphetamine increases synaptic and extracellular 299

DA by reversing the DA transporter42. Microdialysis measures the summed effects of a 300

given drug on synaptic release, extrasynaptic release, and uptake. D2/D3 tracer 301

displacement on the other hand is considered an index of synaptic DA release. This 302

interpretation comes from the fact that while the PET measure reliably correlates with 303

microdialysis measurements of extracellular DA across doses of a given DA-releasing 304

drug, the slopes of these correlations differ across drugs and across brain regions. This 305

presumably reflects differences in the relative contributions of DA release and uptake. 306

This may be important for our interpretation of regional differences in amphetamine-307

induced D2/D3 tracer displacement in SCZ. Regulation of DA release and reuptake in 308

the cortex and other extra-striatal regions differs from regulation in the striatum43-46. For 309

example, in the cortex and other regions with noradrenergic inputs, but not in the 310

striatum, the norepinephrine transporter (NET) is also a major regulator of extracellular 311

DA levels43,44,47,48. A decrease in amphetamine-induced release in the DLPFC observed 312

in SCZ could reflect decreased synthesis and vesicular storage, altered metabolism, or 313

altered regulation of synaptic DA by the DAT or NET. 314

Surprisingly, our PET study uncovered widespread lower DA release in SCZ compared 315

to HC encompassing most cortical and extrastriatal regions, including the ventral 316

midbrain. Notably, this observation appears to be inconsistent with [18F]FDOPA PET 317

and post-mortem findings supporting an increase in DA synthesis and/or storage in the 318

midbrain in SCZ49. Further experiments are needed to confirm whether the increase in 319

[18F]FDOPA uptake and decreased DA release capacity in the midbrain co-exist within 320

the same patients or whether they reflect an uncoupling of DA synthesis and release in 321

the midbrain. 322

The contrast between the generalized DA deficit in cortical and extrastriatal regions and 323

the increase in DA release in the striatum in SCZ24,50-52 is particularly intriguing. This 324

dissociation may reflect abnormal local presynaptic regulation of DA specific to the 325

striatum existing on a background of DA release deficits. Alternatively, a discrete DA 326

neuron subpopulation within the midbrain, innervating the associative striatum (AST), 327

may be over-active. Taken together, the apparently discordant abnormalities across the 328

midbrain, striatum and cortex raises the possibility that SCZ involves a widespread DA 329

release deficit, co-existing with abnormal local dysregulation of DA release or uptake in 330

the striatum (particularly AST), and possible uncoupling of DA synthesis and storage 331

from dendritic DA release in the midbrain. More basic research is required to clarify the 332

mechanisms regulating DA synthesis, vesicular storage, release and reuptake - and their 333

coupling - in midbrain, cortex and striatum45,46. 334

Since DA is important for frontal cortex-dependent cognition including WM, we examined 335

the relationship between DLPFC DA release capacity and fMRI BOLD activation within 336

the DLPFC during WM performance. DA release correlated with BOLD activation, and 337

did not differ between groups. The relationship between BOLD and DA release suggests 338

that fluctuations in DA release in the DLPFC may modulate the strength of the 339

hemodynamic (and presumably neuronal) response to cognitive “processing demands” 340

placed on DLPFC circuitry. Interestingly, while release capacity correlated with the 341

cortical response to the WM challenge, it did not predict performance, which was 342

impaired in patients with SCZ. One potential explanation is that WM performance is 343

tightly coupled to DA release dynamics during cognitive challenges53, a measure not 344

captured by amphetamine-induced release. Notably, we found a positive association 345

between D2 BPND and WM performance in HC but not in SCZ. Consistent with reports 346

that D2 stimulation can effectively gate synaptic plasticity in cortical projection neurons54, 347

this finding suggests that under normal conditions, D2 availability may be a rate-limiting 348

factor for WM whereas in SCZ, WM capacity is limited by mechanisms upstream or 349

independent of DLPFC D2 receptors. 350

In summary, our study established that in SCZ, amphetamine-induced DA release is 351

deficient. This contrasts with the well-replicated increased DA storage and release in 352

the striatum in SCZ. Moreover, the relationships between DA indices and prefrontal 353

cortical function during WM are complex and may be modulated in part by the availability 354

of DA receptors. These findings highlight the need to fully determine the molecular 355

mechanisms regulating DA synthesis, storage, release and reuptake, and examine how 356

these mechanisms operate in different DA projection fields. Such studies will lead to an 357

understanding of the complex dopaminergic phenotype in schizophrenia, and advance 358

the development of a coordinated treatment strategy for symptoms and cognitive 359

disturbances in this disorder. 360

361

362

Acknowledgements 363 The authors wish to acknowledge the staff of the Division of Translational Imaging at the 364

New York State Psychiatric Institute and the staff of Yale University School of Medicine 365

PET Center, whose hard work and expertise made this study possible. 366

Funding support for this study was provided by NIMH 1 P50 MH066171-01A1, The 367

Sylvio O. Conte Center for the study of Dopamine Dysfunction in Schizophrenia 368

369

References 370

371

1. Weinberger DR, Berman KF, Daniel DG. Mesoprefrontal cortical dopaminergic 372 activity and prefrontal hypofunction in schizophrenia. Clin Neuropharmacol. 373 1992;15 Suppl 1 Pt A:568A-569A. 374

2. Green MF. What are the functional consequences of neurocognitive deficits in 375 schizophrenia? Am J Psychiatry. Mar 1996;153(3):321-330. 376

3. Arnsten AF, Goldman-Rakic PS. Noise stress impairs prefrontal cortical cognitive 377 function in monkeys: evidence for a hyperdopaminergic mechanism. Arch Gen 378 Psychiatry. 1998;55(4):362-368. 379

4. Arnsten AF, Cai JX, Murphy BL, Goldman-Rakic PS. Dopamine D1 receptor 380 mechanisms in the cognitive performance of young adult and aged monkeys. 381 Psychopharmacology (Berl). Oct 1994;116(2):143-151. 382

5. Sawaguchi T, Goldman-Rakic PS. D1 dopamine receptors in prefrontal cortex: 383 involvement in working memory. Science. 1991;251(4996):947-950. 384

6. Sawaguchi T, Goldman-Rakic PS. The role of D1-dopamine receptor in working 385 memory: local injections of dopamine antagonists into the prefrontal cortex of 386 rhesus monkeys performing an oculomotor delayed-response task. J 387 Neurophysiol. Feb 1994;71(2):515-528. 388

7. Castner SA, Goldman-Rakic PS. Enhancement of working memory in aged 389 monkeys by a sensitizing regimen of dopamine D1 receptor stimulation. J. 390 Neurosci. Feb 11 2004;24(6):1446-1450. 391

8. Castner SA, Williams GV, Goldman-Rakic PS. Reversal of antipsychotic-induced 392 working memory deficits by short-term dopamine D1 receptor stimulation. 393 Science. 2000;287(5460):2020-2022. 394

9. Cai JX, Arnsten AF. Dose-dependent effects of the dopamine D1 receptor 395 agonists A77636 or SKF81297 on spatial working memory in aged monkeys. J 396 Pharmacol Exp Ther. Oct 1997;283(1):183-189. 397

10. Arnsten AF, Jin LE. Molecular influences on working memory circuits in 398 dorsolateral prefrontal cortex. Prog Mol Biol Transl Sci. 2014;122:211-231. 399

11. Minzenberg MJ, Laird AR, Thelen S, Carter CS, Glahn DC. Meta-analysis of 41 400 functional neuroimaging studies of executive function in schizophrenia. Arch Gen 401 Psychiatry. Aug 2009;66(8):811-822. 402

12. Barch DM, Carter CS. Amphetamine improves cognitive function in medicated 403 individuals with schizophrenia and in healthy volunteers. Schizophr Res. Sep 1 404 2005;77(1):43-58. 405

13. Barch DM, Ceaser A. Cognition in schizophrenia: core psychological and neural 406 mechanisms. Trends Cogn Sci. Jan 2012;16(1):27-34. 407

14. Daniel DG, Weinberger DR, Jones DW, et al. The effect of amphetamine on 408 regional cerebral blood flow during cognitive activation in schizophrenia. J 409 Neurosci. 1991;11(7):1907-1917. 410

15. Glahn DC, Ragland JD, Abramoff A, et al. Beyond hypofrontality: a quantitative 411 meta-analysis of functional neuroimaging studies of working memory in 412 schizophrenia. Hum Brain Mapp. May 2005;25(1):60-69. 413

16. Akil M, Edgar CL, Pierri JN, Casali S, Lewis DA. Decreased density of tyrosine 414 hydroxylase-immunoreactive axons in the entorhinal cortex of schizophrenic 415 subjects. Biol. Psychiatry. 2000;47(5):361-370. 416

17. Akil M, Pierri JN, Whitehead RE, et al. Lamina-specific alterations in the 417 dopamine innervation of the prefrontal cortex in schizophrenic subjects. Am J 418 Psychiatry. 1999;156(10):1580-1589. 419

18. Akil M, Lewis DA. The cathecolaminergic innervation of the human entorhinal 420 cortex: comparisons of schizophrenics and controls. Schizophrenia Res. 421 1995;15:S25. 422

19. Abi-Dargham A, Mawlawi O, Lombardo I, et al. Prefrontal dopamine D1 receptors 423 and working memory in schizophrenia. J. Neurosci. May 1 2002;22(9):3708-424 3719. 425

20. Abi-Dargham A, Xu X, Thompson JL, et al. Increased prefrontal cortical D(1) 426 receptors in drug naive patients with schizophrenia: a PET study with 427 [(1)(1)C]NNC112. Journal of psychopharmacology. Jun 2012;26(6):794-805. 428

21. Okubo Y, Suhara T, Suzuki K, et al. Decreased prefrontal dopamine D1 429 receptors in schizophrenia revealed by PET. Nature. 1997;385(6617):634-636. 430

22. Laruelle M, ABi-Dargham A. Dopamine as the wind of the psychotic fire: new 431 evidence from brain imaging studies. J Psychopharmacology. 1999;in press. 432

23. Laruelle M. Imaging dopamine transmission in schizophrenia. A review and 433 meta- analysis. Q J Nucl Med. 1998;42(3):211-221. 434

24. Howes OD, Kambeitz J, Kim E, et al. The nature of dopamine dysfunction in 435 schizophrenia and what this means for treatment. Arch Gen Psychiatry. Aug 436 2012;69(8):776-786. 437

25. Abi-Dargham A, Simpson N, Kegeles L, et al. PET studies of binding competition 438 between endogenous dopamine and the D-1 radiotracer [C-11]NNC 756. 439 Synapse. May 1999;32(2):93-109. 440

26. Kessler RM, Whetsell WO, Ansari MS, et al. Identification of extrastriatal 441 dopamine D2 receptors in postmortem human brain with [125I]epidipride. Brain 442 Res. 1993;609:237-343. 443

27. Narendran R, Frankel WG, Mason NS, et al. Positron emission tomography 444 imaging of amphetamine-induced dopamine release in the human cortex: a 445 comparative evaluation of the high affinity dopamine D2/3 radiotracers [11C]FLB 446 457 and [11C]fallypride. Synapse. 2009;63(6):447-461. 447

28. Narendran R, Mason NS, May MA, et al. Positron emission tomography imaging 448 of dopamine D(2)/(3) receptors in the human cortex with [(1)(1)C]FLB 457: 449 reproducibility studies. Synapse. Jan 2011;65(1):35-40. 450

29. Innis RB, Cunningham VJ, Delforge J, et al. Consensus nomenclature for in vivo 451 imaging of reversibly binding radioligands. J. Cereb. Blood Flow Metab. Sep 452 2007;27(9):1533-1539. 453

30. Narendran R, Mason NS, Chen CM, et al. Evaluation of dopamine D(2)/(3) 454 specific binding in the cerebellum for the positron emission tomography 455 radiotracer [(1)(1)C]FLB 457: implications for measuring cortical dopamine 456 release. Synapse. Oct 2011;65(10):991-997. 457

31. Abi-Dargham A, Martinez D, Mawlawi O, et al. Measurement of striatal and 458 extrastriatal dopamine D-1 receptor binding potential with [C-11]NNC 112 in 459 humans: Validation and reproducibility. J Cerebr Blood F Met. FEB 460 2000;20(2):225-243. 461

32. Kegeles LS, Slifstein M, Xu XY, et al. Striatal and Extrastriatal Dopamine D-2/D-3 462 Receptors in Schizophrenia Evaluated With [F-18]fallypride Positron Emission 463 Tomography. Biological Psychiatry. Oct 2010;68(7):634-641. 464

33. Slifstein M, Laruelle M. Models and methods for derivation of in vivo 465 neuroreceptor parameters with PET and SPECT reversible radiotracers. Nucl. 466 Med. Bio. 2001;28:595-608. 467

34. Nurnberger JI, Jr., Blehar MC, Kaufmann CA, et al. Diagnostic interview for 468 genetic studies. Rationale, unique features, and training. NIMH Genetics 469 Initiative. Arch Gen Psychiatry. 1994;51(11):849-859; discussion 863-844. 470

35. First M, Spitzer R, Gibbon M, Williams J. Structured Clinical Interview for DSM-IV 471 Axis I Disorders (SCID-I/P, Version 2.0). New York: Biometrics Research Dept., 472 New York State Psychiatric Institute; 1995. 473

36. Kay SR, Fiszbein A, Opler LA. The Positive and Negative Syndrome scale 474 (PANSS) for schizophrenia. Schiz. Bull. 1987;13:261-276. 475

37. Hollingshead AB. Four factor index of social status. New Haven, Connecticut: 476 Working paper published by the author; 1975. 477

38. Cohen JD, Forman SD, Braver TS, Casey BJ, Servan-Schreiber D, Noll DC. 478 Activation of the prefrontal cortex in a nonspatial working memory task with 479 functional MRI. Human Brain Mapping. 1994;1:293-304. 480

39. Wechsler D. Wechsler Adult Intelligence Scale - Third Edition Manual. 481 40. Narendran R, Mason NS, May MA, et al. Positron Emission Tomography Imaging 482

of Dopamine D2/3 Receptors in the Human Cortex With C-11 FLB 457: 483 Reproducibility Studies. Synapse. Jan 2011;65(1):35-40. 484

41. Narendran R, Jedema HP, Lopresti BJ, et al. Imaging dopamine transmission in 485 the frontal cortex: a simultaneous microdialysis and [11C]FLB 457 PET study. 486 Mol Psychiatry. Mar 2014;19(3):302-310. 487

42. Sulzer D, Maidment NT, Rayport S. Amphetamine and other weak bases act to 488 promote reverse transport of dopamine in ventral midbrain neurons. J 489 Neurochem. 1993;60(2):527-535. 490

43. Bymaster FP, Katner JS, Nelson DL, et al. Atomoxetine increases extracellular 491 levels of norepinephrine and dopamine in prefrontal cortex of rat: A potential 492 mechanism for efficacy in Attention Deficit/Hyperactivity Disorder. 493 Neuropsychopharmacol. Nov 2002;27(5):699-711. 494

44. Mazei MS, Pluto CP, Kirkbride B, Pehek EA. Effects of catecholamine uptake 495 blockers in the caudate-putamen and subregions of the medial prefrontal cortex 496 of the rat. Brain Res. May 17 2002;936(1-2):58-67. 497

45. Abercrombie ED, DeBoer P, Heeringa MJ. Biochemistry of somatodendritic 498 dopamine release in substantia nigra: an in vivo comparison with striatal 499 dopamine release. Advances in pharmacology (San Diego, Calif.). 1998;42:133-500 136. 501

46. Siciliano CA, Calipari ES, Ferris MJ, Jones SR. Biphasic mechanisms of 502 amphetamine action at the dopamine terminal. The Journal of neuroscience : the 503 official journal of the Society for Neuroscience. Apr 16 2014;34(16):5575-5582. 504

47. Tanda G, Pontieri FE, Frau R, Di Chiara G. Contribution of blockade of the 505 noradrenaline carrier to the increase of extracellular dopamine in the rat 506 prefrontal cortex by amphetamine and cocaine. Eur J Neurosci. Oct 507 1997;9(10):2077-2085. 508

48. Yamamoto BK, Novotney S. Regulation of extracellular dopamine by the 509 norepinephrine transporter. Journal of neurochemistry. Jul 1998;71(1):274-280. 510

49. Howes OD, Williams M, Ibrahim K, et al. Midbrain dopamine function in 511 schizophrenia and depression: a post-mortem and positron emission 512 tomographic imaging study. Brain. Nov 2013;136(Pt 11):3242-3251. 513

50. Guillin O, Abi-Dargham A, Laruelle M. Neurobiology of dopamine in 514 schizophrenia. Int Rev Neurobiol. 2007;78:1-39. 515

51. Toda M, Abi-Dargham A. Dopamine hypothesis of schizophrenia: making sense 516 of it all. Curr Psychiatry Rep. Aug 2007;9(4):329-336. 517

52. Miyake N, Thompson J, Skinbjerg M, Abi-Dargham A. Presynaptic dopamine in 518 schizophrenia. CNS Neurosci Ther. Apr 2011;17(2):104-109. 519

53. Eshel N, Tian J. Dopamine gates sensory representations in cortex. Journal of 520 neurophysiology. 2014;111(11):2161-2163. 521

54. Xu TX, Yao WD. D1 and D2 dopamine receptors in separate circuits cooperate to 522 drive associative long-term potentiation in the prefrontal cortex. Proceedings of 523 the National Academy of Sciences of the United States of America. Sep 14 524 2010;107(37):16366-16371. 525

526

527

528

Figure Legends 529 530

Figure 1. Scatterplot of DLPFC ∆BPND (left) and regression of DLPFC ∆BPND onto fMRI 531

BOLD increase during SOWT (right). HC data is in black circles, drug-free SCZ data in 532

red triangles and drug-naïve SCZ in blue triangles. Left: Group means are given by 533

horizontal lines. Right: the lines represent the best linear model fit of the data, with slope 534

equal to 53% for both groups and intercepts of 6% in HC and -4% in SCZ. 535

536

Tables 537 538

Table 1. Demographics. 539

HC (n = 21) SCZ patients (n = 20;

1 schizoaffective, 19

schizophrenia)

p*

Age 32.6 ± 8.1 33.1 ± 10.2 .887

Sex (F/M) 11/10 10/10 .879

Ethnicity (C/AA/Hisp/As/mixed) 2/8/6/2/3 1/9/7/1/2 .913

Parental SES 35.9 ± 11.3 42.6 ± 14.5 .133

Participant SES 37.4 ± 14.2 21.4 ± 8.0 <.001

Nicotine smoking (No/Yes) 18/3 15/5 .638

Drug-naïve/drug-free - 6/14

Onset psychotic symptoms (yrs) - 17.3 ± 7.8

Duration of psychotic illness (yrs) - 13.2 ± 11.3

Drug-free interval (months) - 38.4 ± 67.3, (n = 14)

* 2 group t tests for continuous variables, 2 for categorical. Abbreviation: SES = 540

socioeconomic status. 541

Table 2. Clinical and Neurocognitive assessments 542

Assessment HC (n = 21) SCZ (n = 20) 2 grp t (p)

PANSS positive symptoms 7.0 ± 0.2 15.1 ± 4.7 <.001

PANSS negative symptoms 9.0 ± 2.5 13.6 ± 5.59 .001

PANSS general symptoms 17.9 ± 2.5 30.4 ± 8.2 <.001

1-back baseline AHR .998 ± .007 .953 ± .061 .005

1-back post-amphetamine

AHR

.991 ± .025 .947 ± .092 .060

2-back baseline AHR .863 ± .158 .777 ± .248 .209

2-back post-amphetamine

AHR

.916 ± .146 .768 ± .212 .014

3-back baseline AHR .747 ± .229 .654 ± .310 .296

3-back post-amphetamine

AHR

.730 ± .200 .608 ± .238 .088

LNS baseline 16.7 ± 3.3 13.7 ± 3.5 .010

LNS post-amphetamine 17.6 ± 3.9 14.4 ± 3.0 .009

Abbreviations: PANSS = positive and negative symptom scale, AHR = adjusted hit rate, 543

LNS = letter number sequencing task 544

Table 3. Scan Parameters 545

HC (n = 21) SCZ (n = 20) 2 grp t (p)

Base ID (MBq) 175 ± 20 177 ± 21 0.77

Amph ID (MBq) 174 ± 30 179 ± 20 0.50

Base IM (µg) 0.26 ± 0.13 0.24 ± 0.15 0.61

Amph IM (µg) 0.28 ± 0.17 0.26 ± 0.17 0.68

Base SA (MBq/nmol) 400 ± 467 417 ± 229 0.88

Amph SA (MBq/nmol) 344 ± 259 367 ± 176 0.74

fp (Base) 0.30 ± 0.09 0.27 ± 0.08 0.21

fp (Amph) 0.29 ± 0.08 0.26 ± 0.08 0.25

Plasma amph (ng/mL) 82.6 ± 14.2 81.4 ± 24.6* 0.86

Estimated VND 3.27 ± 0.55 2.97 ± 0.86 0.20

Estimated k4 0.023 ± 0.008 0.022 ± 0.005 0.50

Cerebellum VT (base) 4.56 ± 1.08 4.19 ± 0.95 0.23

Cerebellum VT (amph) 4.43 ± 1.11 4.29 ± 1.11 0.68

* Plasma amphetamine available in n = 19 SCZ. Abbreviations: ID = injected dose of 546

radioactivity, IM = injected mass of radiotracer, SA = Specific Activity, Base = baseline 547

scan, amph = post-amphetamine scan, fp, VND, k4 and VT as in Innis et al 200729 548

549

550

551

Table 4. BPND 552

HC SCZ

Baseline Post-Amph ∆BPND Baseline Post-Amph ∆BPND

DLPFC 1.8 ± 0.5 1.6 ± 0.4 -7.5 ± 11.4% 1.8 ± 0.7 1.8 ± 0.6 1.8 ± 11.1%

OFC 1.8 ± 0.5 1.7 ± 0.4 -5.5 ± 13.9% 1.9 ± 0.7 1.9 ± 0.6 2.6 ± 11.3%

MFC 1.9 ± 0.5 1.8 ± 0.4 -4.7 ± 13.2% 2.0 ± 0.8 2.0 ± 0.7 0.9 ± 10.9%

A. CING 2.2 ± 0.6 2 ± 0.5 -4.9 ± 15.7% 2.3 ± 0.8 2.3 ± 0.7 3.6 ± 12.5%

OCC CTX 1.8 ± 0.6 1.6 ± 0.5 -6.5 ± 12.1% 1.8 ± 0.7 1.8 ± 0.6 2.1 ± 10.0%

PAR CTX 2.0 ± 0.6 1.8 ± 0.5 -6.1 ± 9.8% 2.0 ± 0.7 2.0 ± 0.6 1.0 ± 10.4%

TEMP CT 3.4 ± 0.8 3.2 ± 0.8 -5.3 ± 10.0% 3.4 ± 1.0 3.4 ± 0.9 0.2 ± 8.6%

SUB GEN 2.4 ± 0.7 2.3 ± 0.5 -1.5 ± 18.1% 2.6 ± 1.0 2.6 ± 0.8 5.4 ± 18.0%

INSULA 3.4 ± 0.8 3.2 ± 0.7 -4.1 ± 10.2% 3.6 ± 1.1 3.7 ± 1.0 2.3 ± 9.8%

AMYGDALA 5.6 ± 1.7 5.4 ± 1.5 -1.9 ± 16.5% 5.8 ± 2.4 5.9 ± 2.5 2.0 ± 11.0%

HIPPOCAMPUS 2.5 ± 0.9 2.3 ± 0.7 -5.1 ± 14.4% 2.6 ± 1.5 2.6 ± 1.5 0.5 ± 11.9%

SN/VTA 4.6 ± 1.3 4.4 ± 1.2 -4.5 ± 12.7% 5 .0± 2.5 5.1 ± 2.5 2.7 ± 10.8%

THALAMUS 6.2 ± 1.9 5.9 ± 1.6 -3.0 ± 11.7% 6.3 ± 2.5 6.5 ± 2.6 2.8 ± 10.1%

UNCUS 3.8 ± 1.2 3.6 ± 1.2 -5.9 ± 17.5% 4.0 ± 1.8 4.0 ± 1.9 -1.1 ± 11.6%

See PET methods for abbreviations 553

1

Supplement 1 2

eMethods.1 ROI Definitions p2 3 eMethods.2 PET Kinetic Methods and Simulations p2 4 eTable.1 Simulation Results p5 5 eMethods.3 fMRI Methods p6 6 eMethods.4 Adjusted Hit Rate Definition p8 7 eResults.1 Drug Free vs. Drug Naïve p8 8 eTable.2 VT Results p8 9 eReferences. Supplement References p9 10 eFigure.1 Supplement Figure 1 p10 11 12

13 14 15

2

eMethods.1 Operational Definitions of Amygdala and Hippocampus ROIs. 16 The amygdala and hippocampus are both identified by visual inspection on subjects’ coronally oriented 17 T1-weighted structural images as gray matter tissue in the medial temporal lobe. The anterior boundary of 18 the amygdala is the first coronal slice on which the anterior commissure appears. The amygdala continues 19 to be outlined on coronal slices moving in the posterior direction until the first slice on which the lateral 20 horn of the temporal ventricle appears. A horizontal line (parallel to the anterior commissure) is drawn 21 from the most superior part of the ventricle, and amygdala continues to be identified as the medial temporal 22 gray matter superior to this line. The line at the most superior point of the ventricle and the amygdala 23 continue to be drawn on successive posterior slices until the amygdala is no longer visible. The anterior 24 boundary of the hippocampus is defined as the first (most anterior) coronal slice on which the horizontal 25 line superior to the ventricle is drawn. Hippocampus is defined as the medial temporal lobe gray matter 26 inferior to the line. The hippocampus continues to be drawn on successive posterior slices inferior to the 27 line where the line is identified, and posterior to the amygdala and ventricle until the hippocampus is no 28 longer visible. 29

30

eMethods.2 PET Kinetic Method 31 32 To address the issues of low signal-to noise ratio of [11C] FLB457 cortical specific binding as well as the 33 poor suitability of the cerebellum as a reference tissue, a kinetic approach utilizing a modified two tissue 34 compartment model was applied. In this method, the rate constant k4, and VND, the distribution volume of 35 the nondisplaceable compartment, were fitted to a single value across regions, whereas the delivery 36 constant K1 and the rate constant k3 were fitted in each region and condition. The method was applied 37 separately to the 9 cortical regions and the 5 moderate binding subcortical regions examined in the study. 38 The reason for separating the low and moderate binding regions was the possibility that k3 and k4 estimates 39 might be correlated, with moderate binding regions having a lower estimated value of k4 than low binding 40 regions. In theory, k4 is the dissociation rate of the tracer from the receptor and is independent of k3, but in 41 reality the estimated quantities tend to be correlated, and this approach allowed for the optimization routine 42 to converge to fitted values appropriate to the two different ranges of k3. In the simulations, two versions 43 of the method were compared: One in which the same values of k4 and VND were estimated for both the 44 baseline and post-amphetamine scan (the 2 scan method), and a second version in which k4 and VND were 45 fitted separately for baseline and post-amphetamine (1 scan method). To test which approach is more 46 accurate, simulations were performed in which the model assumptions (uniform k4 and VND across 47 regions and/or scans) were intentionally violated and estimated values of VT, BPND and their percent 48 change across conditions were compared to the (simulated) true values. For completeness, standard 2TC 49 fits with all 4 rate constants fitted in each region and condition, using cerebellum as a reference tissue, were 50 performed as well. 51 52 Simulation Methods. Each modeling assumption was tested with a set of 250 Monte Carlo simulations with 53 9 simulated cortical regions, a simulated cerebellum, and mean decrease in BPND following amphetamine 54 of 5%. For each of the 250 runs in a simulation, rate constants [K1, k2, k3, k4] for the baseline condition 55 were generated as the mean observed across cortical regions in the real data, plus a Gaussian random 56 variable with standard deviation of 10% of the mean parameter value. Rate constants for the post-57 amphetamine condition were set to the same value as baseline with k3 decreased across regions by uniform 58 random variables ranging between 0 and 10% of the baseline value (mean decrease of 5%). Pairs of arterial 59 plasma input functions were selected randomly from among the baseline/post-amphetamine pairs of input 60 functions in the real study data, and continuous time activity curves were generated. These were 61 subsampled at the mid-frame times of the collected data to generate discrete time activity curves 62 comparable to the reconstructed PET data, and an additional layer of mean-zero noise with standard 63 deviation equal to 2.5% of the interpolated values was added. These data were then fitted by the 1 scan, 2 64 scan and 2TC methods, with weighting as in the main study. VT was computed in every case. BPND was 65 computed as the estimated k3/k4 ratio for the 1 scan and 2 scan methods and both as k3/k4 and indirectly 66

3

using the cerebellum VT as an estimate of VND (overestimated by 20%, by design, assuming specific 67 binding in cerebellum) for the 2TC method, for comparison. The tested conditions were i) variability in k4 68 across regions, ii) change in k4 between scan conditions, iii) change of VND across scan conditions and 69 combinations of I, ii and iii. To test i) k4 was varied across regions by a Gaussian random variable with 70 standard deviation equal to 5% of the mean. To test ii) k4 was varied between scans, with post-71 amphetamine k3 readjusted by the same amount such that the post-amphetamine decrease in BPND was 72 preserved. These were tested at the 5% and 10% standard deviation levels for k4, as both Gaussian and 73 uniform random variables, with the latter representing the case in which all amphetamine effect was 74 attributable to reduced affinity, i.e. increased k4. To test iii), VND was varied between scan conditions, 75 again with a Gaussian random variable with standard deviation equal to 5% or10% of the baseline value. 76 Finally combinations of ii and iii, i.e. simultaneous changes of both k4 and VND across conditions, were 77 tested. 78 79 Results. Representative results are shown in Table S1. The table shows the ordinary regression slopes and 80 intercepts of the fitted parameters against the simulated true parameters, along with Pearson correlation 81 coefficients, for each of the 2250 fits for a representative set of simulation conditions. The 2 scan and 1 82 scan methods both outperformed conventional 2TC under all conditions. This includes BPND and 83 ∆BPND where the simulated specific binding in cerebellum could be expected to affect the outcome, but 84 also accuracy and precision of baseline and post-amphetamine VT, where many of the tested conditions 85 would have appeared a priori to favor the greater flexibility of 2TC (data not shown). Under all of the 86 tested conditions, the 2 scan method was more accurate than the 1 scan method. Surprisingly, this included 87 all simulations in which k4 or VND was changed across conditions, which might be expected to favor the 1 88 scan method. We conclude that the simulations support the 2 scan method as being the most sensitive 89 approach for extracting ∆BPND and ∆VT using [11C]FLB457 with amphetamine challenge. 90 91 Discussion. In this study, we applied a kinetic method that would simultaneously address the low signal in 92 cortex, the low expected change in that signal across the tested conditions, and the lack of a suitable 93 reference tissue. Our method fitted single values of VND and k4 across brain regions and scans for each 94 subject. The assumption of single VND is widely accepted, and in fact implicit in the interpretation of 95 ∆BPND as reflecting changes in specific binding only. The use of a single k4 is less conventional, though 96 not unique to this study 1. There is some evidence suggesting that measured k4 may vary across regions 97 due to differing microenvironments 2. Additionally, on statistical grounds, data fitting algorithms could 98 conceivably provide correlated parameter estimates, for example, estimating BPND in high binding regions 99 by a combination of both large k3 and small k4. The simulations here showed that even when the 100 underlying compartment model incorporated all of these conditions that do not conform to the 2 scan model 101 assumptions, the 2 scan method still outperformed all other methods, due to the increased parsimony 102 associated with estimating 38 parameters per subject, rather than 40 for the 1 scan method or 80 for 2TC. 103 One reasonable concern when applying this approach is identifiability of VND and BPND, i.e. that the 104 partition of total distribution volume VT into its component factors VND and (1 + BPND) from direct 105 estimation of these may not represent the true physiological parameters in the absence of additional 106 information to formulate constraints in the data fitting process. To address this concern, we also measured 107 and reported total VT and ∆VT and found these to be most precisely measured by the 2 scan method as 108 well. We expected, a priori, that ∆VT would be less than ∆BPND because a portion of VT is VND which 109 should not be affected by DA release. In fact, in the study data we did see that ∆VT < ∆BPND in HC, but 110 the blunting in SCZ and significant group differences were still detectable. As noted in the main article, 111 average estimated VND, 2.97 ± 0.87 in SCZ and 3.27 ± 0.55, is in approximate accord with the Narendran 112 study 3 in that cerebellum VT measured here was 4.19 ± 0.95 in SCZ and 4.56 ± 1.08 in HC, corresponding 113 to VND equaling 70 ± 13% of cerebellum VT in SCZ and 73 ± 10% in HC. Nevertheless, we cannot claim 114 to have measured physiological VND by this method without further validation, for example through 115

4

blocking studies. Our simulations do however strongly suggest that the approach improves estimation of 116 ∆VT and ∆BPND compared to methods that measure within subject regional parameters independently. 117 118 119 120

5

121 eTable.1 Simulation Results for ∆BPND and ∆VT 122

Condition: k4 varies within scan (Gaussian, 5% SD) across regions but not between scans

slope intercept R % Outliers ∆BP 2TC indirect 1.0077 0.0083 0.0073 14% ∆BP 2TC 1.1703 0.0270 0.0333 0.3% ∆BP 1 scan 1.0326 0.0206 0.0405 0 ∆BP 2 scan 1.0157 0.0009 0.5106 0 ∆VT 2TC 1.1550 0.0141 0.0609 0.4% ∆VT 1 scan 1.0234 0.0033 0.2203 0 ∆VT 2 scan 1.0084 0.0004 0.5078 0 Condition: k4 varies within scan (Gaussian, 5% SD) and between scans (uniform between 0 and 10%)

slope intercept R % Outliers ∆BP 2TC indirect 1.1776 -0.0031 0.0104 0.3% ∆BP 2TC 1.0388 0.0205 0.0265 0.0027 ∆BP 1 scan 1.1552 0.0249 0.0478 0 ∆BP 2 scan 1.0238 0.0203 0.4779 0 ∆VT 2TC 1.1816 0.0068 0.0699 0.1% ∆VT 1 scan 1.0018 -0.0014 0.1920 0 ∆VT 2 scan 1.0224 0.0136 0.4781

Condition: VND varies between conditions (Gaussian, SD = 5%)

slope intercept R % Outliers ∆BP 2TC indirect 1.1424 -0.0215 0.0109 15% ∆BP 2TC 0.9151 0.0082 0.0222 0.1% ∆BP 1 scan 1.1581 0.0167 0.0812 0 ∆BP 2 scan 1.0656 0.0057 0.1414 0 ∆VT 2TC 0.9993 0.0041 0.2930 0.1% ∆VT 1 scan 0.9771 -0.0029 0.7182 0 ∆VT 2 scan 0.9626 -0.0017 0.8783 0

Condition: k4 varies within (Gaussian, SD = 5%), between (uniform 0 to 10%) VND varies between (Gaussian, SD = 5%)

slope intercept R % Outliers ∆BP 2TC indirect 0.9094 0.0239 0.0079 15% ∆BP 2TC 0.9851 0.0185 0.0241 0.3% ∆BP 1 scan 0.9996 0.0188 0.0658 0 ∆BP 2 scan 0.9969 0.0108 0.1245 0 ∆VT 2TC 1.0429 0.0051 0.2859 0.2% ∆VT 1 scan 1.0269 -0.0016 0.7280 0 ∆VT 2 scan 0.9731 0.0057 0.8606 0

Regression of parameter estimates for each method against true simulation parameters. ∆BP 2TC 123 indirect used cerebellum VT to estimate regional BPND indirectly as a function of distribution volumes. 124 Cerebellum VT was set to 20% greater than VND. The other 2TC method used direct estimation of k3/k4 125 for BPND. Fits for which |∆BPND| > 80% or |∆VT| > 100% were considered outliers and not included in 126 the regression. 127 128 129

6

eMethods.3 fMRI Methods 130 131 132 SOWT Task. The participants completed 24 trials of the SOWT, with each trial containing eight steps on 133 which response was required. At the start of each trial, eight simple line drawings of three-dimensional 134 objects were presented in a 3 × 3 grid, with the central position of the grid being empty. The stimuli were 135 the same as those used by Curtis and colleagues 4. Unique stimuli were used on each of the first 12 trials, 136 and stimuli were repeated exactly once during the latter 12 trials. On each step, participants had 7 s to move 137 a mouse cursor to select any object that had not been selected on a previous trial (thus, all responses were 138 correct on the first step). Once a selection was made, a white outline was displayed around the selected 139 object until 9 s had elapsed from the start of the step. At this point, the objects in the display were pseudo-140 randomly rearranged in the grid, with the blank space appearing in the same location as the most recently 141 selected item (to prevent participants from using a spatial strategy or simply responding in the same 142 location on each trial). If no response was made within 7 s, a white outline was displayed for 2 s around a 143 randomly selected object that would have been a correct response; participants were instructed to remember 144 this object as if they had selected it themselves. If an incorrect selection was made, a red box was displayed 145 over the object until 7 s had elapsed from the start of the step, after which the same procedure was followed 146 as in the case when a participant made no response. The ITI was 9 s. 147 Data acquisition. Imaging was carried out on a Philips 1.5 Tesla Intera scanner at the Columbia 148 Radiology MRI Center at the Neurological Institute of New York. Participants lay supine on the scanner 149 bed while viewing stimuli projected onto a screen located at the foot of the scanner bed through a mirror 150 mounted on the head coil. The cursor was controlled with a hand-held fiber optic trackball with buttons on 151 either side. T1-weighted images were obtained with an SPGR sequence with a 256 mm FOV, 200 slices, 152 and 1 mm isotropic voxels. Whole-brain functional EPIs were obtained using an 8-channel SENSE coil 153 with a 2 s TR, 28 ms TE, 77° flip angle, 192 mm field of view, 40 slices, 3 mm isotropic voxels, and 154 SENSE factor of 1.5. The relatively small slice thickness precluded acquisition of data on the ventral most 155 portion of the cerebellum in a large majority of participants; however, this region was not of particular 156 interest in the present study. Participants completed nine runs of 160 volumes each that included either 157 three task trials or two task trials and one control trial between the two task trials. Thirty seconds of rest 158 occurred after each trial, and 32 seconds before the first trial in each run. 159 Data preprocessing. Preprocessing was performed with a combination of SPM8 and custom 160 Matlab scripts. Reconstructed PAR/REC format files were obtained from the scanner and converted to 32-161 bit floating point precision Analyze format files. A rough in-brain mask was computed and in-brain signal 162 values were used to identify artifactual volumes: any volume departing from a sliding window by more 163 than eight mean absolute deviations in terms of either mean global signal or Mahalanobis distance was 164 identified and treated as a nuisance regressor during first-level statistical analyses. 165 Data then underwent slice-timing correction using SPM8 and motion realignment using 166 INRIAlign (Freire, Roche, & Mangin, 2002). Motion realignment parameters were inspected to detect 167 excessive motion; one run from each of two participants (one patient and one control) were excluded from 168 analyses due to translation greater than 2.5 mm or rotation greater than 2.5° from their median position 169 during the run. EPI images were then coregistered to the individual subjects' T1 image used for drawing of 170 ROIs for PET analyses. PET ROIs were then resliced into a 3 mm isotropic voxel space to match the EPI 171 data, and these resliced ROIs were used on an individual-subject basis to extract BOLD activation to the 172 task. For calculation of group-level activation maps, images underwent spatial normalization to the ICBM 173 template using the segmentation algorithm in SPM8 and spatial smoothing with an 8 mm FWHM kernel. 174 First-level statistical modeling. Prior to first-level modeling, the value in each voxel in each 175 volume was divided by the mean value in that voxel over the entire time-series, expressed as a percent, so 176 that the magnitude of the first-level hemodynamic response function (HRF) estimates were equivalently 177 scaled across runs. Data for each participant were modeled in the GLM framework implemented in SPM8. 178 A three parameter HRF model (with temporal and dispersion derivatives) was used to estimate the Blood 179 Oxygen Level Dependent (BOLD) response to each modeled event. An explicit mask was calculated by 180 using the conjunction of the smoothed (6 mm FWHM) gray matter segmentation and the skull-stripped 181 mean EPI for each subject, to restrict the analysis to regions of gray matter and regions not suffering from 182 excessive signal dropout due to susceptibility artifacts, respectively. 183 A separate set of regressors was used to model each of the eight steps of the SOWT and the 184 control task as a nine second boxcar (resulting in 27 regressors in total, due to the three parameter HRF 185

7

model). Although participants had only seven seconds in which to make a response, we opted to use the full 186 nine seconds of each step in order to capture the period of time in which participants were presumably 187 attempting to maintain previously selected items in WM. Error trials on any of the task steps and on the 188 control task made up two separate sets of regressors (six total), again modeled as nine second boxcars. A 189 two second boxcar was used to model the presentation of textual instructions prior to each trial, and motor 190 responses (button presses) and error feedback (red square appearing over the incorrectly-selected item) 191 were modeled as instantaneous events in order to prevent motor and visual activity from being confounded 192 with other modeled events. Finally, nuisance regressors included all six motion parameters estimates (three 193 translation parameters and three rotation parameters), the squared motion parameters, the first derivative of 194 the motion parameters, the squared derivative of the motion parameters, and dummy regressors for 195 artifactual scans identified as outlined above. Activation at each step of the SOWT and during the control 196 task was quantified as the area under the curve (AUC) in a temporal window ranging from 2 s to 9 s with 197 respect to the three basis functions defining the canonical HRF; this window corresponds to the rise and fall 198 of the HRF following the initial dip and prior to the undershoot. 199 Second-level statistical modeling. Contrast images of overall task-related activation were 200 calculated for each participant by taking the mean activation at each voxel across steps one through eight in 201 the SOWT and subtracting the activation in the corresponding voxel in the control task. These contrast 202 images were then tested for significance using robust regression5 and thresholded at P < 0.05 after false 203 discovery rate (FDR) correction6. 204 Relationship between fMRI BOLD with PET ΔBPND in DLPFC. Voxels showing significant task-205 related activation in either group were returned to individual-subject spaces by using the inverse of the 206 normalization transformation produced during segmentation. BOLD percent signal change values were 207 extracted for each subject from voxels that were significant at the group level and also fell within the 208 DLPFC ROI drawn for a given participant. These percent signal changes values were then used in a 209 regression model with ΔBPND. Model selection was determined based on Akaike information criterion7. 210 211 212 213

8

eMethods.4 Adjusted Hit Rate Formula for the N-Back Task 214 The hit rate (HR) was calculated as the number of correct responses divided by the number of targets 215 (maximum is 1, minimum is 0). The error rate (ER) was calculated as the number of errors divided by the 216 number of non-targets (maximum is 1, minimum is 0). The adjusted HR (AHR) was calculated as HR – ER. AHR 217 ranges from 1 (if the subject provided all the correct responses and no incorrect response) to –1 (if the subject 218 provided no correct response and all the incorrect responses). Operating at chance level corresponds to an AHR 219 of 0.. 220 221 222

eResults.1 Drug Naïve vs Drug Free Patients 223 The drug-naïve (DN) patients had higher BPND (t = 2.96, p = 0.008) and lower ΔBPND (t = -2.12, p = 224 0.048) and ΔVT (t = -2.28, p = 0.035) in the DLPFC than the drug-free (DF) patients. However, these 225 groups differed in age (DN: 25.9 ± 7.2 years; DF: 36.1 ± 10.0 years; t = -2.24, p = 0.038). Given the well 226 documented effect of age on cortical D2R parameters8,9 analyses were repeated with age as a covariate. 227 With age included, there are no significant group differences for BPND, ΔBPND or ΔVT. DLPFC BPND 228 and ΔBPND do not correlate significantly with duration of illness or drug-free interval. DLPFC VT 229 correlates negatively with duration of illness in the total group of patients (r = -0.538, p = 0.014), which is 230 largely driven by the drug-free patients (r = -0.547, p = 0.043). 231 232

eTable.2 VT Results 233 eTable 2. Distribution volumes (VT) 234

HC (n = 21) SCZ (n = 20)

Baseline Post-Amph ∆VT Baseline Post-Amph ∆VT

CEREBELLUM 4.6 ± 1.1 4.4 ± 1.1 -2.9 ± 7.5% 4.2 ± 1.0 4.3 ± 1.1 2.1 ± 7.8%

DLPFC 9.2 ± 2.6 8.7 ± 2.4 -4.8 ± 6.9% 8.1 ± 2.8 8.2 ± 2.8 0.9 ± 7.1%

OFC 9.4 ± 2.6 8.9 ± 2.4 -3.9 ± 8.4% 8.4 ± 3.0 8.5 ± 3.1 1.1 ± 6.8%

MFC 9.5 ± 2.5 9.1 ± 2.4 -3.2 ± 8.0% 8.6 ± 2.8 8.6 ± 2.8 0.2 ± 7.1%

A. CING 10.5 ± 2.9 10 ± 2.7 -3.8 ± 9.6% 9.4 ± 2.9 9.6 ± 3.0 2.2 ± 8.6%

OCC CTX 9.1 ± 2.8 8.7 ± 2.5 -4.3 ± 7.4% 8.2 ± 3.0 8.3 ± 3.2 1.1 ± 6.5%

PAR CTX 9.8 ± 2.8 9.4 ± 2.6 -4.2 ± 6.3% 8.8 ± 3.3 8.8 ± 3.3 0.5 ± 6.7%

TEMP CT 14.6 ± 4.1 14 ± 3.9 -4.1 ± 7.6% 13.1 ± 5.0 13.1 ± 5.1 0.1 ± 6.9%

SUB GEN 11.2 ± 3.3 10.8 ± 2.6 -1.6 ± 11.6% 10.3 ± 3.7 10.5 ± 3.4 3.1 ± 11.9%

INSULA 14.4 ± 3.9 13.9 ± 3.7 -3.2 ± 7.8% 13.4 ± 4.5 13.7 ± 4.9 1.8 ± 7.7%

AMYGDALA 24.5 ± 8.9 23.7 ± 7.9 -1.7 ± 13.8% 21.4 ± 7.2 21.8 ± 8.0 1.8 ± 9.2%

HIPPOCAMPUS 13 ± 4.6 12.3 ± 3.9 -3.9 ± 10.1% 11.2 ± 2.8 11.2 ± 3.3 0.1 ± 8.0%

SN/VTA 20.8 ± 6.5 19.9 ± 6.1 -3.7 ± 10.5% 18.5 ± 5.7 18.9 ± 5.8 2.3 ± 8.9%

THALAMUS 26.8 ± 9.3 25.7 ± 8.4 -2.7 ± 10.1% 23.4 ± 8.7 23.9 ± 9.2 2.4 ± 8.6%

UNCUS 17.9 ± 6.0 16.9 ± 5.7 -4.7 ± 13.1% 15.9 ± 5.6 15.8 ± 6.1 -0.9 ± 9.5% Region Distribution Volumes at baseline, following amphetamine, and their percent change 235 236 237 238

9

Supplement References 239 240 241 242 1. Ashworth S, Rabiner EA, Gunn RN, et al. Evaluation of 11C-GSK189254 as a 243

novel radioligand for the H3 receptor in humans using PET. Journal of nuclear 244 medicine : official publication, Society of Nuclear Medicine. Jul 2010;51(7):1021-245 1029. 246

2. Votaw JR, Kessler RM, de Paulis T. Failure of the three compartment model to 247 describe the pharmacokinetics in brain of a high affinity substituted benzamide. 248 Synapse. 1993;15(3):177-190. 249

3. Narendran R, Mason NS, Chen CM, et al. Evaluation of dopamine D(2)/(3) 250 specific binding in the cerebellum for the positron emission tomography 251 radiotracer [(1)(1)C]FLB 457: implications for measuring cortical dopamine 252 release. Synapse. Oct 2011;65(10):991-997. 253

4. Curtis CE, Zald DH, Pardo JV. Organization of working memory within the 254 human prefrontal cortex: a PET study of self-ordered object working memory. 255 Neuropsychologia. 2000;38(11):1503-1510. 256

5. Wager TD, Keller MC, Lacey SC, Jonides J. Increased sensitivity in 257 neuroimaging analyses using robust regression. NeuroImage. May 15 258 2005;26(1):99-113. 259

6. Benjamini Y, Hochberg Y. CONTROLLING THE FALSE DISCOVERY RATE 260 - A PRACTICAL AND POWERFUL APPROACH TO MULTIPLE TESTING. 261 Journal of the Royal Statistical Society Series B-Methodological. 1995;57(1):289-262 300. 263

7. Akaike H. A new look at the statistical model identification. IEEE Trans. 264 Automat. Contr. 1974;AC19:716-723. 265

8. Inoue M, Suhara T, Sudo Y, et al. Age-related reduction of extrastriatal dopamine 266 D2 receptor measured by PET. Life Sci. Jul 20 2001;69(9):1079-1084. 267

9. Kaasinen V, Vilkman H, Hietala J, et al. Age-related dopamine D2/D3 receptor 268 loss in extrastriatal regions of the human brain. Neurobiol Aging. Sep-Oct 269 2000;21(5):683-688. 270

271 272

10

e Figure.1 Scatterplots of fitted vs true results from a representative simulation where all 3 assumptions 273 were violated: k4 varied across regions and increased between scan conditions and VND changed 274 randomly between scan conditions. BPND decreased following amphetamine randomly between 0 and 275 10%. In this set of simulations the BPND change was always contributed to by increased k4 across 276 conditions but the 2 scan method still gave the best estimates of the changes in BPND and VT across 277 conditions even though the approach fits a single k4 across scans. 278 279 280