Brain state stability during working memory is explained by network control theory, modulated by dopamine D1/D2 receptor function, and diminished in schizophrenia

Alessandro Bertolino; Anais Harneit; Andreas Meyer-Lindenberg; Axel Schaefer; Daniel Durstewitz; Danielle S. Bassett; Emanuel Schwarz; Fabio Pasqualetti; Giulio Pergola; Giuseppe Blasi

arxiv: 1906.09290 · v1 · pith:P2WR4FOBnew · submitted 2019-06-21 · 🧬 q-bio.NC · physics.bio-ph

Brain state stability during working memory is explained by network control theory, modulated by dopamine D1/D2 receptor function, and diminished in schizophrenia

Urs Braun , Anais Harneit , Giulio Pergola , Tommaso Menara , Axel Schaefer , Richard F. Betzel , Zhenxiang Zang , Janina I. Schweiger

show 10 more authors

Kristina Schwarz Junfang Chen Giuseppe Blasi Alessandro Bertolino Daniel Durstewitz Fabio Pasqualetti Emanuel Schwarz Andreas Meyer-Lindenberg Danielle S. Bassett Heike Tost

This is my paper

Pith reviewed 2026-05-25 18:05 UTC · model grok-4.3

classification 🧬 q-bio.NC physics.bio-ph

keywords working memorybrain state stabilitynetwork control theorydopamine D1 D2 receptorsschizophreniaenergy landscapestate transitions

0 comments

The pith

Working memory involves brainwide switching between activity states whose stability is set by D1 dopamine receptors and whose transitions depend on D2 receptors.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that working memory requires the brain to switch between distinct activity states distributed across many regions. Stability of each state tracks levels of dopamine D1 receptor gene expression, while the energy cost of moving between states tracks D2 receptor expression and responds to drugs that act on D2 receptors. Patients with schizophrenia display a broader range of possible energy costs and maintain working memory states less reliably than controls. A reader would care because the account ties a specific molecular signal to the network mechanics that let cognition remain flexible yet stable.

Core claim

Working memory entails brainwide switching between activity states. The stability of states relates to dopamine D1 receptor gene expression while state transitions are influenced by D2 receptor expression and pharmacological modulation. Schizophrenia patients show altered network control properties, including a more diverse energy landscape and decreased stability of working memory representations.

What carries the argument

Network control theory metrics computed on brain imaging data that quantify the energy needed to hold or change activity states during a working memory task.

If this is right

Higher D1 receptor expression predicts greater stability of the activity pattern that holds a working memory item.
D2 receptor modulation changes the energy cost of moving from one working memory state to another.
Schizophrenia is marked by both greater diversity in possible transition energies and lower average stability of task-relevant states.
These control properties supply a network-level explanation for why dopamine signaling supports flexible maintenance of information.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Genetic differences in D1 expression could forecast individual variation in how steadily people hold items in working memory.
Drugs that selectively alter D2 signaling might be tested for their ability to normalize transition costs in patient groups.
The same control framework could be applied to other tasks that require rapid shifts between cognitive states, such as task switching or attention.

Load-bearing premise

The network control metrics derived from brain scans accurately capture the real stability and transition costs of working memory activity states.

What would settle it

Absence of any correlation between regional D1 receptor gene expression and measured state stability in an independent group of participants scanned during working memory performance.

read the original abstract

Dynamical brain state transitions are critical for flexible working memory but the network mechanisms are incompletely understood. Here, we show that working memory entails brainwide switching between activity states. The stability of states relates to dopamine D1 receptor gene expression while state transitions are influenced by D2 receptor expression and pharmacological modulation. Schizophrenia patients show altered network control properties, including a more diverse energy landscape and decreased stability of working memory representations.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Applies network control theory to fMRI working memory data to tie state stability to D1 receptor expression, transitions to D2, and reduced stability in schizophrenia, but lacks visible methods to judge the mapping.

read the letter

The main point is that this paper takes network control theory and applies it to brain imaging during working memory, claiming that state stability tracks D1 dopamine receptor gene expression while transitions track D2 expression and can be shifted by drugs, with schizophrenia patients showing less stable states and a broader energy landscape. They frame working memory as brainwide switching between activity states explained by these control properties. What the work does is pull together control theory, receptor maps, and patient data into one story, which is a step beyond pure theory or pure imaging studies. The abstract suggests they have task fMRI, genetic correlations, and a clinical comparison to support the claims. The soft spots sit in the core mapping. Without methods shown, it is unclear how activity states were defined from the imaging, how control energies were calculated from the networks, or whether the correlations survive proper controls for multiple testing and confounds. Gene expression is almost certainly from public atlases rather than the same subjects, so any link stays correlational and could reflect many other variables. The assumption that control metrics directly capture working memory stability rather than some other signal in the data needs explicit validation against behavior or pharmacology, and that step is not visible here. This paper is aimed at people who work at the intersection of network neuroscience and computational psychiatry. A reader already following control theory applications to cognition might find the combination useful if the analyses check out. It deserves peer review because the framing is coherent and the topic matters, even though the current presentation leaves the key steps unexamined.

Referee Report

2 major / 2 minor

Summary. The paper claims that working memory involves brainwide switching between activity states whose stability is explained by network control theory, modulated by dopamine D1 receptor gene expression for stability and D2 for transitions, with pharmacological effects and diminished stability plus a more diverse energy landscape in schizophrenia patients.

Significance. If the mapping from control-theoretic quantities to empirical state stability and transitions holds after proper validation, the work would link network control metrics to dopaminergic gene expression and clinical phenotypes in a falsifiable way, offering a mechanistic account of working-memory dynamics.

major comments (2)

[Abstract] The abstract supplies no methods, state definitions, imaging modality, task details, statistical thresholds, or validation analyses, so it is impossible to evaluate whether the reported correlations with D1/D2 expression or the schizophrenia differences are load-bearing or artifactual; this must be addressed with explicit sections on data acquisition, state identification, and control-metric computation.
[Results (presumed)] The central claim that network control quantities accurately reflect stability and transitions rests on an untested mapping (weakest assumption in the stress-test note); without explicit validation against independent behavioral or electrophysiological measures of state persistence, the interpretation remains circular with the definitions of energy landscape and control energy.

minor comments (2)

Notation for control energy, minimum control energy, and energy landscape should be defined with equations and referenced to prior network-control literature.
Gene-expression correlations require reporting of effect sizes, multiple-comparison correction, and spatial specificity tests against null atlases.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their comments. We address each major point below and indicate where revisions have been made.

read point-by-point responses

Referee: [Abstract] The abstract supplies no methods, state definitions, imaging modality, task details, statistical thresholds, or validation analyses, so it is impossible to evaluate whether the reported correlations with D1/D2 expression or the schizophrenia differences are load-bearing or artifactual; this must be addressed with explicit sections on data acquisition, state identification, and control-metric computation.

Authors: We agree the abstract is brief by design and omits methodological specifics. The full manuscript contains dedicated Methods sections on fMRI data acquisition during the n-back working memory task, state identification from dynamic functional connectivity, and control energy computation on the structural connectome. Statistical procedures and thresholds are reported there. We have revised the abstract to include a concise statement on imaging modality, task, and network control approach to facilitate evaluation of the D1/D2 and schizophrenia results. revision: yes
Referee: [Results (presumed)] The central claim that network control quantities accurately reflect stability and transitions rests on an untested mapping (weakest assumption in the stress-test note); without explicit validation against independent behavioral or electrophysiological measures of state persistence, the interpretation remains circular with the definitions of energy landscape and control energy.

Authors: The mapping is not circular: empirical states are identified from fMRI time series, while control metrics are derived from an independent structural connectome and linear dynamical system. We demonstrate that these metrics account for observed state persistence and transitions, with additional support from correlations to behavioral accuracy. The stress-test section examines robustness. Direct electrophysiological validation is not feasible with the available dataset, but the separation of modalities and behavioral correlations provide non-circular support for the interpretation. revision: partial

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The provided abstract and context contain no equations, derivations, or explicit claims that any network control theory metric reduces to a fitted parameter, self-definition, or self-citation chain. The central claims are empirical correlations between control-theoretic quantities, dopamine receptor expression, and working memory state stability in patients versus controls. No load-bearing step is shown to be equivalent to its inputs by construction, and the derivation (if present in the full text) is treated as self-contained against external benchmarks such as imaging data and pharmacological modulation. This is the expected honest non-finding when no specific reduction can be quoted.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no identifiable free parameters, axioms, or invented entities; all such elements would require the full methods and results sections.

pith-pipeline@v0.9.0 · 5676 in / 1012 out tokens · 24769 ms · 2026-05-25T18:05:44.666125+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

12 extracted references · 12 canonical work pages

[1]

Participants and study design All participants provided written informed consent for protocols approved by the Institutional Review Board of the medical faculty in Mannheim. For the first study including healthy controls and patients with schizophrenia, a total of 202 subjects (178 healthy controls, 24 schizophrenia patients) were included (see Table S1)....

work page
[2]

fMRI For the first study, BOLD fMRI was performed on a 3T Siemens Trio (Erlangen, Germany) in Mannheim, Germany

Data acquisition 2.1. fMRI For the first study, BOLD fMRI was performed on a 3T Siemens Trio (Erlangen, Germany) in Mannheim, Germany. Prior to the acquisition of functional images, a high-resolution T1-weighted 3D MRI sequence was conducted (MPRAGE, slice thickness=1.0 mm, FoV = 256 mm, TR = 1570ms, TE = 2.75 ms, TI = 800ms, α = 15°). Subsequently, funct...

work page 2000
[3]

We transformed a recently published multimodal atlas (8) into a volumetric format by projecting its FreeSurfer pial cortex coordinates into standard MNI space

Atlas construction To combine structural and functional brain imaging data, we first constructed a brain atlas that equally well respects functional and anatomical features. We transformed a recently published multimodal atlas (8) into a volumetric format by projecting its FreeSurfer pial cortex coordinates into standard MNI space. A grey matter prior pro...

work page
[4]

After estimation of the diffusion tensor, we performed deterministic whole-brain fiber tracking as implemented in DSI Studio using a modified FACT algorithm (11)

Connectome construction For the DTI data, the following preprocessing steps were performed with standard routines implemented in the software package FSL (9): i) correction of the diffusion images for head motion and eddy currents by affine registration to a reference (b0) image, ii) extraction of non-brain tissues (10), and iii) linear diffusion tensor f...

work page
[5]

Brain state definition Because we were interested in investigating how the brain controls and transitions between global brain states underlying circumscribed cognitive processes (such as those supporting working memory, attention, and motor behavior), we defined brain states as stationary patterns of activity during execution of these processes. It is im...

work page
[6]

Network control theory 6.1. Optimal control theory framework To model the transition between 0-back and 2-back brain states, we used the framework of optimal control, following prior work (15-17) implemented in MATLAB. Based on individual brain states X=[x1,…xn] (in our case simplified to n = 2 states: 0-back and 2-back, see above) and a structural brain ...

work page
[7]

v3.macGT1

Gene based polygenic co-expression indices 7.1. Genotyping, imputation and quality control In this study, we used human GWAS data of 63 healthy subjects who were genotyped using HumanHap 610 and 660w Quad BeadChips. For all subjects, standard quality control (QC) and imputation were performed using the Gimpute pipeline (Chen, Lippold et al. 2018) and the ...

work page 2018
[8]

Control analyses 9.1. Null models of structural brain networks To study the impact of structural brain networks on control properties, we repeated the computation of control energy using a randomized null model of the individuals’ structural brain networks that preserves the average weight, the strength distribution and the degree distribution. Null model...

work page
[9]

Figure S1: N-back task design Design of the N-back task: Stimuli were presented in blocks of either 0-back (left) or 2-back (right) conditions

Supplementary figures 10.1. Figure S1: N-back task design Design of the N-back task: Stimuli were presented in blocks of either 0-back (left) or 2-back (right) conditions. There was no additional control or resting condition. In the 0-back condition, subjects were instructed to press the button on the response box corresponding to the number currently dis...

work page
[10]

Supplementary tables 11.1. Table S1: Characteristics of the healthy control and patient samples Healthy controls (n = 178) Matched controls (n = 80) Schizophrenia patients (n = 24) t or χ² value P value Demographic information Age (year) 33.05 ± 10.98 35.49 ± 10.55 32.25 ± 10.33 1.32 0.188 Sex (male / female) 93 / 85 46 / 34 18 / 6 2.39 0.122 Years of edu...

work page
[11]

Tombaugh TN (2004) Trail Making Test A and B: normative data stratified by age and education

Supplementary references 1. Tombaugh TN (2004) Trail Making Test A and B: normative data stratified by age and education. Archives of clinical neuropsychology 19(2):203-214. 2. Lehrl S, Triebig G, & Fischer B (1995) Multiple choice vocabulary test MWT as a valid and short test to estimate premorbid intelligence. Acta Neurologica Scandinavica 91(5):335-345...

work page arXiv 2004
[12]

(2018) Prefrontal co-expression of schizophrenia risk genes is associated with treatment response in patients

Pergola G, et al. (2018) Prefrontal co-expression of schizophrenia risk genes is associated with treatment response in patients. bioRxiv. 28. Zhang B & Horvath S (2005) A general framework for weighted gene co-expression network analysis. Statistical applications in genetics and molecular biology 4:Article17. 29. Colantuoni C, et al. (2011) Temporal dynam...

work page 2018

[1] [1]

Participants and study design All participants provided written informed consent for protocols approved by the Institutional Review Board of the medical faculty in Mannheim. For the first study including healthy controls and patients with schizophrenia, a total of 202 subjects (178 healthy controls, 24 schizophrenia patients) were included (see Table S1)....

work page

[2] [2]

fMRI For the first study, BOLD fMRI was performed on a 3T Siemens Trio (Erlangen, Germany) in Mannheim, Germany

Data acquisition 2.1. fMRI For the first study, BOLD fMRI was performed on a 3T Siemens Trio (Erlangen, Germany) in Mannheim, Germany. Prior to the acquisition of functional images, a high-resolution T1-weighted 3D MRI sequence was conducted (MPRAGE, slice thickness=1.0 mm, FoV = 256 mm, TR = 1570ms, TE = 2.75 ms, TI = 800ms, α = 15°). Subsequently, funct...

work page 2000

[3] [3]

We transformed a recently published multimodal atlas (8) into a volumetric format by projecting its FreeSurfer pial cortex coordinates into standard MNI space

Atlas construction To combine structural and functional brain imaging data, we first constructed a brain atlas that equally well respects functional and anatomical features. We transformed a recently published multimodal atlas (8) into a volumetric format by projecting its FreeSurfer pial cortex coordinates into standard MNI space. A grey matter prior pro...

work page

[4] [4]

After estimation of the diffusion tensor, we performed deterministic whole-brain fiber tracking as implemented in DSI Studio using a modified FACT algorithm (11)

Connectome construction For the DTI data, the following preprocessing steps were performed with standard routines implemented in the software package FSL (9): i) correction of the diffusion images for head motion and eddy currents by affine registration to a reference (b0) image, ii) extraction of non-brain tissues (10), and iii) linear diffusion tensor f...

work page

[5] [5]

Brain state definition Because we were interested in investigating how the brain controls and transitions between global brain states underlying circumscribed cognitive processes (such as those supporting working memory, attention, and motor behavior), we defined brain states as stationary patterns of activity during execution of these processes. It is im...

work page

[6] [6]

Network control theory 6.1. Optimal control theory framework To model the transition between 0-back and 2-back brain states, we used the framework of optimal control, following prior work (15-17) implemented in MATLAB. Based on individual brain states X=[x1,…xn] (in our case simplified to n = 2 states: 0-back and 2-back, see above) and a structural brain ...

work page

[7] [7]

v3.macGT1

Gene based polygenic co-expression indices 7.1. Genotyping, imputation and quality control In this study, we used human GWAS data of 63 healthy subjects who were genotyped using HumanHap 610 and 660w Quad BeadChips. For all subjects, standard quality control (QC) and imputation were performed using the Gimpute pipeline (Chen, Lippold et al. 2018) and the ...

work page 2018

[8] [8]

Control analyses 9.1. Null models of structural brain networks To study the impact of structural brain networks on control properties, we repeated the computation of control energy using a randomized null model of the individuals’ structural brain networks that preserves the average weight, the strength distribution and the degree distribution. Null model...

work page

[9] [9]

Figure S1: N-back task design Design of the N-back task: Stimuli were presented in blocks of either 0-back (left) or 2-back (right) conditions

Supplementary figures 10.1. Figure S1: N-back task design Design of the N-back task: Stimuli were presented in blocks of either 0-back (left) or 2-back (right) conditions. There was no additional control or resting condition. In the 0-back condition, subjects were instructed to press the button on the response box corresponding to the number currently dis...

work page

[10] [10]

Supplementary tables 11.1. Table S1: Characteristics of the healthy control and patient samples Healthy controls (n = 178) Matched controls (n = 80) Schizophrenia patients (n = 24) t or χ² value P value Demographic information Age (year) 33.05 ± 10.98 35.49 ± 10.55 32.25 ± 10.33 1.32 0.188 Sex (male / female) 93 / 85 46 / 34 18 / 6 2.39 0.122 Years of edu...

work page

[11] [11]

Tombaugh TN (2004) Trail Making Test A and B: normative data stratified by age and education

Supplementary references 1. Tombaugh TN (2004) Trail Making Test A and B: normative data stratified by age and education. Archives of clinical neuropsychology 19(2):203-214. 2. Lehrl S, Triebig G, & Fischer B (1995) Multiple choice vocabulary test MWT as a valid and short test to estimate premorbid intelligence. Acta Neurologica Scandinavica 91(5):335-345...

work page arXiv 2004

[12] [12]

(2018) Prefrontal co-expression of schizophrenia risk genes is associated with treatment response in patients

Pergola G, et al. (2018) Prefrontal co-expression of schizophrenia risk genes is associated with treatment response in patients. bioRxiv. 28. Zhang B & Horvath S (2005) A general framework for weighted gene co-expression network analysis. Statistical applications in genetics and molecular biology 4:Article17. 29. Colantuoni C, et al. (2011) Temporal dynam...

work page 2018