Neuroimaging in insomnia: Review and reconsiderations

Over the last decades, neuroimaging has become a substantial component of insomnia research. While theoretical underpinnings of different studies vary just like methodological choices and the experimental design, it is suggested that major features of insomnia disorder rely on the impaired function, structure, metabolism and connectivity of brain areas involved in sleep generation, emotion regulation, self‐processing/‐awareness and attentional orientation. However, neuroimaging research on insomnia often suffers from small sample sizes, heterogeneous methodology and a lack of replicability. With respect to these issues, the field needs to address the questions: (1a) how sufficiently large sample sizes can be accumulated within a reasonable economic framework; (1b) how effect sizes in insomnia‐related paradigms can be amplified; (2a) how a higher degree of standardisation and transparency in methodology can be provided; and (2b) how an adequate amount of flexibility/complexity in study design can be maintained. On condition that methodological consistency and a certain degree of adaptability are given, pooled data/large cohort analyses can be considered to be one way to answer these questions. Regarding experimental single‐centre trials, it might be helpful to focus on insomnia‐related transdiagnostic concepts. In doing so, expectable effect sizes (in between‐subjects designs) can be increased by: (a) comparing groups that are truly distinct regarding the variables examined in a concept‐specific paradigm; and (b) facilitated, intensified and precise elicitation of a target symptom.

field, at first with the development of magnetic resonance imaging (MRI), which provides a detailed imaging of internal body and brain structures (Nofzinger et al., 2013).The foundation on which MRI is built is represented by nuclear magnetic resonance (NMR), which focuses on the hydrogen highly contained in the human body, using a magnetic field to rebuild different types of images (Shifteh & Lipton, 2013).The development of MRI also allowed to collect functional and metabolic data in the human subjects in addition to structural data (Gomes & Lipton, 2013).Indeed, functional MRI (fMRI) is now the most frequently used functional neuroimaging technique distinguished by a high spatial resolution and the ability to assess brain activity by exploiting the different magnetic properties of deoxygenated and oxygenated blood (Gomes & Lipton, 2013).In this field, an important combined method was developed that simultaneously involves electroencephalogram (EEG) and fMRI (EEG-fMRI), mainly used to investigate the fMRI correlates of spontaneous brain activity observable with EEG (Laufs & Krakow, 2013).
In addition to MRI, another important structural method is diffusion tensor imaging (DTI), which permits to non-invasively evaluate the structural arrangement of white matter tracts, which can provide a better understanding of normal and pathological brain structure by using the direction-dependent (anisotropic) diffusion properties of water contained by axons and their myelin sheaths (Gomes & Lipton, 2013).
Furthermore, neuroimaging methods used to assess metabolism in the human brain also exist: magnetic resonance spectroscopy uses the slight difference in radiofrequency signals detected by MRI to distinguish between metabolites, used then to non-invasively investigate brain metabolic processes in vivo (De Graaf, 2007); then, positron emission tomography (PET) and single-photon emission computed tomography (SPECT) use radiotracers for assessing the metabolism of the human brain.The first one functions through the detection of annihilation photons emitted when radionuclides from a radiolabelled tracer that is introduced into the body decay and release positrons, allowing to map the tracer concentration in each voxel (Bybel et al., 2013).The SPECT, instead, works through a gamma camera movement around a circular or elliptical orbit around the patient, which converts the photons emitted by the radioactive decay into a light pulse and then into a voltage signal (Lorberboym, 2013).The most used radiotracers for each technique are, respectively, 18F Fluorodeoxyglucose (Bybel et al., 2013) and Technetium99m (Lorberboym, 2013).
The results coming from neuroimaging studies on the topic of insomnia are many and are constantly evolving.It is important to note that the heterogeneity of the techniques used for investigating the brain correlates of this sleep disorder, as well as the different analysis approaches that may focus on the whole brain or on specific regions of interest depending on the direction of the study, can make it difficult to compare the huge number of results in the field of neuroimaging, as well as to understand the actual weight that each finding has on the whole picture.
Below, a brief theoretical overview will be presented as an attempt to summarise the current state of the art in ID from a neuroimaging perspective, focusing on the brain areas most qualitatively reported in the literature and their relationship to the most accredited models used to explain the neurobiology of insomnia.

| The hyperarousal theory
The first pioneering neuroimaging data supporting the hyperarousal model of insomnia (Riemann et al., 2010) came from a PET study conducted by Nofzinger et al. (2004), which highlighted that a smaller metabolic decline from waking to sleep occurs in ID in regions that participate in the promotion of wakefulness, such as the ascending reticular activating system, hypothalamus, thalamus, insula, amygdala, hippocampus, anterior cingulate cortex (ACC) and medial prefrontal cortices (Nofzinger et al., 2004).The present state of the art is studded by studies that continue to support this neurobiological model.
The thalamus participates in several mechanisms of sleep-wake control, i.e. cortical activation and synchrony and sleep consolidation (Gent et al., 2018), and plays a fundamental role in modulating arousal functions (Coenen et al., 2012).In ID, a state of hyperarousal seems in fact to be connected to an enhanced synchronisation of the thalamus with emotion-related (Ma et al., 2018), sensory-related (Huang et al., 2022;Kim et al., 2021) and attentional (Perrier et al., 2022) areas during wakefulness, as well as to a lower thalamic deactivation after high-beta brain states increase (Feige et al., 2017) and a negative thalamocortical connectivity during sleep (Zou et al., 2021).Moreover, lower thalamic γ-aminobutyric acid (GABA) levels during wakefulness (Winkelman et al., 2008) and a reduced thalamic volume (Li et al., 2019) have been pointed out.
The ACC, which plays a prominent role within the emotional circuits of the brain (Yan et al., 2018), mostly seems to drive an impaired emotion regulation (Seo et al., 2018;Wassing et al., 2019) and an enhanced emotional hyperarousal in ID, as revealed by resting-state fMRI studies (Cheng et al., 2022;Wang et al., 2020;Yan et al., 2018).Higher indices of arousal in the ACC are also related to its increased volume (Li et al., 2021;Winkelman et al., 2013) and decreased GABA rates (Plante et al., 2012) occurring in patients with insomnia.The enhanced emotional hyperarousal seems to also occur from a higher synchronisation within the emotion circuit, encompassing the thalamus, the ACC and the insula (Ma et al., 2018;Wang et al., 2017), but also the amygdala and the prefrontal lobe, which may represent a higher information flow in emotional-related regions in patients with ID (Ma et al., 2018).
A downregulation of thinking processing exacerbating the cognitive arousal occurring in ID may be linked to a failure to disengage from the attention network during sleep (Guo et al., 2022) and to a reduced functionality in some areas of the default mode network (DMN; Marques et al., 2017;Wang et al., 2020), which is the main resting-state network in human brain visibly active during situations where attention is internally directed (Leech & Sharp, 2014).To date, several nodes of the DMN implicated in self-referential processes such as the posterior cingulate cortex (PCC) and the right precuneus, along with the left middle frontal gyrus (MFG), which takes part in the left executive control network (LECN), also seem to participate in a state-specific arousal, showing an enhanced relative regional cerebral metabolic rate for glucose during sleep (Kay et al., 2016).

| Increased interoceptive sensitivity and alteration of sensory processing
Insomnia disorder also seems to be related to an alteration of sensory processing across a wide range of interoceptive modalities (Wei & Van Someren, 2020).The brain area representing the centre of salience processing across many sensory and cognitive domains is the insula (Uddin, 2015), which has a major role in the continuous central nervous system's processing of bodily signals defined as interoception (Wei et al., 2016).Moreover, it was also pointed out as a source of low-frequency oscillations (< 1 Hz) occurring during slow-wave activity in non-rapid eye movement (NREM) sleep (Murphy et al., 2009).
Along with the ACC, the insula constitutes a key node of the salience network (SN), a set of neural circuits involved in the conscious awareness of both exteroceptive and interoceptive stimuli (Wei et al., 2016), and it seems to be involved in altered cortical responses to interoceptive signals with the anticipation of aversive events in ID (Seo et al., 2018).Moreover, as shown by EEG-fMRI studies, an increased coactivation of the anterior insula at sleep onset (Chen et al., 2014), as well as an increased connectivity between the anterior insula and frontal and temporal regions during NREM sleep (Li et al., 2022), may represent a fundamental contribution to sleep discrepancy in patients with insomnia.Indeed, it is known that a discrepancy between the subjective perception of the amount of sleep and the parameters objectively recorded by instruments such as polysomnography occurs in patients with insomnia, in a phenomenon called sleep state misperception (Harvey & Tang, 2012;Riemann et al., 2015); in this context, the correlation between a longer sleep-onset latency in NREM sleep and an enhanced metabolic activity in the anterior insula and other several brain regions associated with conscious awareness such as the ACC and the PCC (Kay et al., 2017) seems to support this phenomenon.Moreover, a greater intra-network connectivity of the SN found in the dorsal ACC may represent self-regulatory difficulties and amplified negative affective reactivity, promoting abnormal levels of sensory and information processing, and resulting in difficulties in initiating and maintaining sleep (Cheng et al., 2022).
The thalamus also represents a key structure in the interoceptive pathway (Chen et al., 2021).For this reason, reduced volume of bilateral thalamus (Li et al., 2019) and decreased resting-state thalamic connectivity with several areas (e.g., ACC, orbitofrontal cortex, caudate, putamen and hippocampus) may lead to an impaired emotional processing and dysfunctions in decision-making and memory in patients with insomnia (Li et al., 2019).Cognitive and emotional dysfunctions present in ID are also supported by DTI findings, which suggest insular hypoconnectivity and reduced connectivity between frontal areas (e.g.MFG) and classical limbic areas (e.g., amygdala, hippocampus, and thalamus; Jespersen et al., 2020).

| Difficulties in network modulation
The heightened sensory processing detectable in ID may also be linked to an unbalance in resting-state networks activation, with an emphasis towards the DMN where a decreased activity mainly occurring in the cingulate cortex has been pointed out (Marques et al., 2017;Pang et al., 2017).Moreover, reduced functional integration, functional segregation and lower nodal efficiency involving the ACC, the insula and the precuneus have been pointed out (Wu et al., 2020).Indeed, a decreased modulation of the DMN and the LECN from the SN has been highlighted in ID, which may be connected to a reduced functional connectivity of the insula and the SN during wakefulness (Li et al., 2018;Liu et al., 2018;Wei et al., 2020).This may be justified by an impaired synchronicity between the cores of the SN and the DMN, leading to an increased sensitivity and selfawareness (Wang et al., 2017), and to a reduced ability to disengage from external information processing at sleep onset, along with the potential emergence of cognitive dysfunction (Wang et al., 2022) and negative emotion persistence and emotion inflexibility reported in ID (Seo et al., 2018;Wassing et al., 2019).This seems to result in a difficulty in networks' switching following changes in environment and needs (Wei et al., 2020), as supported by reduced disengagement of the DMN (Drummond et al., 2013) and activation of the LECN (Altena et al., 2008) during cognitive tasks.Such impairment may determine a confusing flow of information, leading also to the presence of intrusive worries and thoughts at bedtime (Fichten et al., 2001;Harvey, 2002;Wicklow & Espie, 2000) due to the failure in inhibiting non-salient information (Wei et al., 2016).

| Summary
As literature suggests, the major features of ID appear to rely on the impaired function, structure, metabolism and connectivity of areas involved in sleep generation (e.g. the thalamus and the insula; Gent et al., 2018;Murphy et al., 2009), emotion regulation (e.g. the ACC; Etkin et al., 2011), self-processing and self-awareness (e.g. the PCC and the precuneus; Brewer et al., 2013;Cavanna & Trimble, 2006), and attentional orientation (e.g. the MFG; Japee et al., 2015).A common interpretation is that the general hyperarousal subjects with ID experience during both sleep and wakefulness becomes apparent when bringing together the specific neuroimaging findings.Moreover, the increased emotional and cognitive reactivity described in the hyperarousal model might be underlined by an impairment in sensory gating and exteroceptive signals processing occurring in ID, possibly resulting in an altered perception of the state of consciousness during sleep.Because most of the areas identified in neuroimaging studies on ID are part of widespread brain circuits showing functional connections and synchronisation, an alteration in the balance of activity and modulation of the main brain networks is assumedin particular, regarding increased sensitivity and reactivity during resting state but also when performing cognitive tasks and at sleep onset.

| RECONSIDERATIONS
Over the last decades, neuroimaging has become a substantial component of clinical research on psychiatric disorders.Against the background of this general development, it is no surprise that the quantity of research articles on neuroimaging in ID has grown considerably (see PubMed statistics: https://pubmed.ncbi.nlm.nih.gov/?term=% 22neuroimaging%22%20%22insomnia%22&timeline=expanded).
While most of the studies seek to find neurobiological correlates to altered psychological and sleep-related functioning associated with ID, theoretical underpinnings of different studies vary just like methodological choices and the experimental design.

| Consistently inconsistent
This heterogeneity becomes particularly apparent in experimental studies on regional brain activation, brain morphometric alterations and functional connectivity, mostly examined on small samples in betweengroups MRI paradigms.As pointed out in previous epidemiological research on neurobiological processes in ID (Schiel et al., 2022), multiple problems arise from this situation.(1) The statistical power achieved in many studies is limited by sample size in a way that expectable effect sizes can hardly be detected.(2) The comparability of results is, amongst others, reduced by methodological differences (brain atlas/ segmentation choice, data preparation), heterogeneous group compositions (operationalisation of ID, selection of exclusion criteria), variance in data analysis (descriptive or explorative approach) and non-uniform statistical protocols (e.g.multiple testing correction in connectivity analyses).( 3) The listed inconsistencies (within and between neuroimaging studies on ID) lead to a certain lack of replicability (Button et al., 2013;Nord et al., 2017), putting a strain on conclusions drawn from such results.While we think it is important to point out these issues, it seems at least equally important to emphasise thatdepending on the considered concept, model or studiesthe seriousness of problems varies just like the problems themselves.

| Pooled data, large cohorts and open questions
Since it is an almost insurmountable demand on experimental singlecentre trials to accumulate sufficiently large sample sizes with respect to economic and administrative burden, there lies great potential in pooling data or using large cohort databases, respectively.In the context of neuroimaging and insomnia, an example of bringing together data from different sites and studies is the ENIGMA-Sleep Working Group.In a recent analysis, Weihs et al. (2023) examined ID-related morphology differences in more than 1000 individuals from three cohorts.In our studies on associations between sleep health and brain imaging outcomes in the UK Biobank (Holub et al., 2023;Schiel et al., 2022;Schiel et al., 2023), we had the opportunity to analyse data from about 30,000 participants.While these approaches provide an enormous statistical power gainwhich is fundamentally necessary in the context of neuroimagingthey may be confined by either heterogeneous methodology (pooling data) or inflexible/simplistic operationalisation (using large cohort databases).For example, the quality of pooled data analyses highly depends on the comparability of results within the study (see previous paragraph), whereas the flexibility/complexity of database analyses is hard-limited by the availability and operationalisation of relevant variables.
Taken together, neuroimaging research on insomnia faces the following questions.(1a) How can we accumulate sufficiently large sample sizes within a reasonable economic framework, while (1b) amplifying the effect sizes we expect in insomnia-related paradigms?(2a) How can we provide a sufficient degree of standardisation and transparency in methodology, while (2b) keeping an adequate amount of flexibility/complexity in study design?

| Towards answers (1): General considerations
Given that the cost of neuroimaging studies will not decline in the near future, it must be acknowledged that experimental single-centre trials may continue to experience difficulties in detecting small effect sizes.Because most high-resolution neuroimaging methods like (f)MRI are "noisy by nature", relying on indirect measures, non-trivial signal processing and imperfect data acquisition, it might be more expedient to aim for stronger effects (see 1b) than for higher sensitivity within a small sample size.This goal can be approached by optimising the study design, by intensifying the paradigms used to actualise affective or cognitive processes during the scan, and by improving within-group homogeneity and between-group heterogeneity.
With respect to pooled data, it seems particularly important to combine high statistical power (1a) with high comparability of results by paying strict attention to methodological consistency (2a).Furthermore, it must be ensured that statistical analyses are targeted and that explorative calculations are denoted as such (e.g.pre-registration, transparent analysis plans).This also applies for using large cohort databases, which, moreover, profit from a highest possible degree of adaptability (e.g.continual modifiability of paradigms, expandability of variables, see 2b).

| Towards answers (2): A little more specific
Transforming these reflections into something more tangible, we would like to exemplify how to possibly address the questions raised above (1a-2b) in future research.When getting creative about sidestepping the natural constraints of experimental (single-centre) trials, it might be worth to identify rather artificial and thus modifiable constraints: for example, most experimental neuroimaging research on insomnia relies on a study design comparing healthy good sleepers (HGS) with patients with ID.While homogeneity within the ID (or HGS) group is given with respect to the presence (or absence) of core ID symptoms as identified in the study (e.g.self-reported problems falling asleep), it often cannot be excluded that individuals within the ID group differ considerably regarding the overall clinical picture they present (e.g.cognitive or affective impairments).When examining ID-related brain activation in comparably small samples, this unmonitored and thus uncontrolled variance within the independent variable might result in highly diffuse effects, most likely of small size.
Because it is key to increase detectability of effects in experimental neuroimaging studies (1b), confining oneself to HGS versus ID-study designs can be a hindrance.An alternative, promising approach might be to address transdiagnostic concepts (TDCs) rather than ID as a diagnose (Cuthbert, 2014).Following this idea ("symptom rather than syndrome"), a first step would be to identify TDCs of relevance for ID.Moreover, when choosing a TDC with respect to theoretical considerations (see Section 2: "Review") and/or clinical experiences, it is advisable to opt for a clear-cut, well-defined concept (e.g. for "rumination" rather than for "dysfunctional cognition").A second step would be to orient the study design towards the chosen TDC (grouping, paradigm, analysis plan).Finally, neuroimaging findings from TDC research can be integrated into a broader picture of neurobiological processes underlying insomnia.For an illustration of the transdiagnostic approach, see Figure 1.
Importantly, this approach moves away from the claim that neurobiological mechanisms in ID can be captured without further differentiation (e.g. with respect to the accentuation of TDCs).However, increased homogeneity within groups and improved operationalisability might be worth the costand, in small-scale studies, even necessary to bridge the gap between attainable sample and effect sizes (1a, b).

| CONCLUSIONS
Neuroimaging has become a substantial component of insomnia research, and major features of ID have been linked to altered brain function/connectivity, structure or metabolism.Novel insights into neurobiological processes are hoped to help develop a comprehensive pathophysiological model of ID, confirming, adjusting or extending preexisting theoretical assumptions about the nature of insomnia.Although structures involved in sleep generation, emotion regulation, selfprocessing and attentional orientation have been repeatedly associated with ID, experimental (MRI) studies on regional brain activation, brain morphometric alterations and functional connectivity often suffer from heterogeneous methodology and small sample sizes.With respect to future research, multiple questions arise from this observation.These are (along with corresponding solution approaches) summarised in Table 1.
While pooled data/large cohort analyses can be considered solution approaches themselves (main issues: balancing methodological consistency and flexibility/complexity), overcoming difficulties inherent in experimental single-centre trials poses a greater challenge.
Aiming to increase homogeneity within groups and improve  operationalisability, it might be helpful to focus on ID-related TDCs.In doing so, expectable effect sizes (in between-subjects designs) can be increased by comparing groups that are truly distinct regarding the variables examined in a TDC-specific paradigm.

| Limitations and outlook
While we deem the issues raised in the "Reconsiderations" section substantial, it must be clarified that previous neuroimaging studies on insomnia should always be regarded individually and with an adequate amount of differentiation.A detailed, study by study review of previous research with respect to the problems described above is neither possible nor the goal of the current overview article.
Much rather, we want to raise awareness of the replication crisis in neuroimaging and of how it affects the insomnia field.While doing so, we refer primarily to fMRI research.The solution approaches listed and discussed are not intended to be exhaustive.
For future considerations, it might be of interest to also take into

F
I G U R E 1 Illustration of an approach addressing transdiagnostic concepts (TDCs) rather than insomnia disorder (ID) as a diagnose.(a) Identifying TDCs relevant for ID; (b) examining TDC-related brain activation (e.g. during rumination, possibly elicited in a therapeutic setting); (c) bringing together ID and neuroimaging findings from TDCs.[Color figure can be viewed at wileyonlinelibrary.com]T A B L E 1 Neuroimaging in insomnia: challenges (referring to the questions raised above, 1a-2b) and solution approaches account the potential of longitudinal study designs, the manipulability of independent variables (e.g. through therapeutic interventions) or the role of technical solutions to signal processing improvement and parsimonious (AI-assisted) data analysis.In light of the clinical relevance of developing a comprehensive pathophysiological model of ID, methodological challenges in neuroimaging research should be addressed actively, systematically andat bestcreatively.