Abstract
Craving is a central feature of substance use disorders and disorders due to addictive behaviors. Considerable research has investigated neural mechanisms involved in the development and processing of craving. Recently, connectome-based predictive modeling, a data-driven method, has been used in four studies aiming to predict craving related to substance use, addictive behaviors, and food. Studies differed in methods, samples, and conceptualizations of craving. Within the commentary we aim to compare, contrast and consolidate findings across studies by considering conceptual and methodological features of the studies. We derive a theoretical model on the functional connectivity-craving relationships across studies.
Craving, intense urges to engage in a behavior, apply importantly to substance and behavioral addictions (Antons, Brand, & Potenza, 2020; Mallorquí-Bagué, Mestre-Bach, & Testa, 2023; Tiffany & Wray, 2012; Vafaie & Kober, 2022). Understanding brain networks underlying craving thus has implications for treatment development.
Previous work has predominantly focused on inference-statistical and theory-driven analytic approaches where a-priori defined hypotheses on the activity or connectivity of specific brain regions and networks hypothetically related to craving were tested. A large meta-analysis on the neural correlates of cue-reactivity and craving (as well as other addiction-related mechanisms) suggests consistent involvement of the reward network (including the ventral striatum, rostral anterior cingulate cortex, and orbitofrontal cortex) (Zilverstand, Huang, Alia-Klein, & Goldstein, 2018). In addition, the analyses show that while cue-reactivity responses might be anchored in the ventral striatum, the neural responses extend to multiple large-scale networks. Similar results were found for behavioral addictions (Antons et al., 2020). Studies suggest craving as being linked to both altered neural activity and resting-state functional connectivity (Zeng, Han, Gao, Sun, & Yuan, 2023; Zilverstand, O’Halloran, & Goldstein, 2018). One recent meta-analysis investigating resting-state functional connectivity, for example, implicates changes within and between the fronto-parietal network, the ventral attention network and the default-mode network (Zeng et al., 2023). However, these results had not been previously related to craving experiences.
Newer machine-learning approaches generate brain-behavior models from, for example, whole-brain functional connectivity data (e.g., Shen et al., 2017). Driven by data, machine-learning algorithms (e.g., support vector machines/regressions, partial least squares regressions, neural networks, or random forests) have been used to generate brain-behavior models within training datasets. In a second step, the predictive power or performance has been tested in independent/novel datasets. This step of out-of-sample prediction is important to generate more accurate and generalizable models. Connectome-based predictive modeling (CPM; Finn et al., 2015; Shen et al., 2017) is a data-driven machine-learning approach that derives brain-behavior models based on whole-brain functional-connectivity data. First, features are selected by correlating all edges of the functional-connectivity matrix with the behavioral measure (e.g., measure of craving) over all participants of a training dataset. Second, edges whose correlation-coefficient reach a specific threshold, are used as features. Third, a sum value is estimated across all selected edges and a linear model is fitted based on the brain (sum over all selected connectivity measures) – behavior (behavioral measure) relationship. CPM generally differentiates between positive networks (all positive correlation coefficients) and negative networks (all negative correlation coefficients). Fourth, the model is then applied to novel subjects (or one novel subject in a leave-one-out cross-validation), and the performance is estimated by correlating the predicted values with the actual values. Often, permutation testing is used to determine the level of significance. Due to the linear approach, CPM is not a total black-box but allows inspection of features and networks used for the prediction. While findings from such connectome-wide approaches are also complex, raising challenges for simplistic interpretations, this information can be compared to theoretical models and may also be valuable for updating existing theoretical models or deriving new hypotheses, testable in theory-driven or data-driven future work, based on mechanisms implicated in the brain-behavior relationships (Yip, Kiluk, & Scheinost, 2019).
To date and to our best knowledge, CPM has been used in four studies with the aim to predict craving (Antons, Yip, et al., 2023; Garrison et al., 2023; Yang et al., 2023; Zhou et al., 2022). In these studies, craving was predicted for alcohol (Garrison et al., 2023), cocaine (Antons, Yip, et al., 2023; Garrison et al., 2023), opioids (Yang et al., 2023), food (Garrison et al., 2023), gambling (Antons, Yip, et al., 2023) and gaming (Zhou et al., 2022). In two studies (Antons, Yip, et al., 2023; Garrison et al., 2023), the prediction was done across different types of addictions. Given that all studies used the same data-driven methodological CPM approach (despite smaller differences), consolidating the findings of these studies could provide insights into the general neural networks relevant for craving processing. Therefore, we compared, contrasted, and integrated findings across studies by considering conceptual and methodological features of the studies (Table 1). We aimed to add to existing theories by investigating the functional connectivity-craving relationship across studies and in particular across addictions.
Methodological characteristics of studies and main results
| Study by Antons, Yip, et al. (2023) | Study by Garrison et al. (2023) | Study by Yang et al. (2023) | Study by Zhou et al. (2022) | |
| Sample |
|
|
|
|
| fMRI task |
|
|
|
|
| Craving measure |
|
|
|
|
| CPM specificities |
|
|
|
|
| External validation |
|
|
|
|
| Number of features identified |
|
|
|
|
| Highest degree nodes |
|
|
|
|
All four studies were successful in predicting measures of craving with similar performance but different relevance of positive and negative networks: Antons, Yip, et al. (2023) (positive: r = 0.16, p = 0.143; negative: r = 0.38, p = 0.009; both: r = 0.27, p = 0.038, permutation testing), Garrison et al. (2023) (both, r = 0.41, p = 0.37, q2 = 0.14, mean square error = 34.22, p < 0.001, permutation testing 1,000 iterations, one-tailed), Yang et al. (2023) (positive: r = 0.027, p = 0.149, negative: r = 0.42, p < 0.001, both: r = 0.14, p = 0.100) and Zhou et al. (2022) (positive: r = 0.49, p < 0.001; negative: r = 0.02, p = 0.46; both: r = 0.31, p = 0.045, permutation test).
The number of edges identified as features and the highest degree nodes differed between studies (Table 1, associations with functions described here were derived from Neurosynth if not identified by the authors, Yarkoni, Poldrack, Nichols, Van Essen, & Wager, 2011). Antons, Yip, et al. (2023) identified prefrontal regions implicated in executive functions and inhibitory control and regions implicated in emotion, reward processing, and face recognition. Garrison et al. (2023) identified regions implicated in memory performance and visual and sensory processing. Yang et al. (2023) identified regions relevant in processing of coordination, movement, action, sensory processing, working memory, inhibitory control and visual processing. Zhou et al. (2022) identified regions implicated in object recognition, motor planning and emotion and reward processing.
Considering canonical networks, the studies show wide overlaps of networks identified as highly relevant for the prediction of craving either because of the number of nodes selected as features or by virtual lesion approaches. A summary is presented in Fig. 1. The relevant networks identified by each study are represented in the colored circles. The default mode, motor/sensory, subcortical/basal ganglia, and visual I networks were identified in all four studies. These networks might be relevant in the prediction of craving across different methodologies and may represent the core neural networks of craving processing. Using the information from the positive and negative networks in the study by Antons, Yip, et al. (2023), there might be one network consisting of the motor/sensory, subcortical/basal ganglia, and medial frontal networks with high synchronous activity. This network, might reflect more automatic and reactive mechanisms of craving. The medial frontal network has not been identified in the study by Yang et al. (2023). As this study did not use a direct confrontation with addiction related cues but a resting-state approach, the medial frontal network as regulatory instance might have been less relevant. A second, higher-order network consisting of the fronto-parietal network integrating with salience/limbic and visual I networks may potentially reflect cognitive mechanisms of craving. The fronto-parietal-hub network (cognitive mechanism) was less consistently represented in the features of the four studies. The visual I network, that has been shown to be relevant in all four studies might represent the imaginal prefiguration as important feature of desire thinking (Brandtner, Antons, Cornil, & Brand, 2021).
Integration of the results from Antons, Yip, et al. (2023), Garrison et al. (2023), Yang et al. (2023), and Zhou et al. (2022) regarding networks predicting craving. The red, blue and grey lines were derived from the positive and negative networks observed in Antons, Yip, et al. (2023). Red lines – foremost positive connectivity, blue lines – foremost negative connectivity, grey line – both positive and negative connectivity
Citation: Journal of Behavioral Addictions 13, 3; 10.1556/2006.2024.00050
As CPM is a data-driven approach, differences in results may be explained by the different tasks, stimuli, samples, craving assessments, types of addictive behaviors, and other factors. For example, the stimuli used in the studies rely on different sensory modalities (audio-visual vs. aural vs. visual only). While individualized imagery scripts as used in Garrison et al. (2023) might require consolidating memory resources (hippocampus as a high degree node), the videos shown in the study by Antons, Yip, et al. (2023) may require the processing and recognition of faces (fusiform gyrus as a high-degree node) and the video-game scenes shown in the paradigm by Zhou et al. (2022) may require identifying objects related to the game League of Legends (inferior temporal gyrus as a high-degree node). In the study by Yang et al. (2023) a resting-state paradigm was used without any cues. However, regions involved in visual processing, such as the primary visual cortex and frontal eye field, were found to be among the ten nodes with the highest degree. These may represent top-down visual processes related to imaginal prefiguration (Brandtner et al., 2021). Differences in models may also reflect samples. For example, previous studies have indicated gender effects in neural correlates of cue-reactivity and craving (Dong et al., 2018; Hallam, Boswell, DeVito, & Kober, 2016), and the CPM studies reviewed here differ in gender distributions. In addition, CPM studies focusing on other aspects of addiction, such as risky alcohol-use in youth, have demonstrated significant sex differences in model performance and in the anatomy of identified networks predictive of female vs. male alcohol use (Yip et al., 2023). The studies by Garrison et al. (2023) and Zhou et al. (2022) included control participants without clinically defined addictive disorders. By including individuals without addictions in a data-driven method, it is possible that the subjective craving measured may in part reflect affective and motivational components involved in non-problematic urges/desires. Cognitive components as identified in Antons, Yip, et al. (2023) might become particularly relevant to individuals when “getting caught up” in severe forms of craving in addictions and might not be relevant when predicting non-pathological desires/urges. On the other hand, the individualized (versus generalized) cues used in Garrison et al. (2023) may more robustly elicit craving responses which especially involve the salience/limbic system and visual components. In addition, learning mechanisms might differ between different types of addictions. Natural rewards such as food might affect associative learning differently as compared to substances forming the foundation of current substance use disorders or non-substance-related addictive behaviors like gambling or gaming (Hill-Bowen et al., 2021; van Timmeren, van Holst, & Goudriaan, 2023).
While in three of the four studies craving was assessed during cue-reactivity, Yang et al. (2023) predicted craving at 8 month follow up after a detoxification treatment. Both the timing of the craving assessment and the context of treatment might have affected craving responses and its neural correlates. The networks identified by Yang et al. (2023) are especially characterized by a high proportion of subcortical and cerebellar structures as features and their high connectivity to cortical structures (e.g., motor-network, fronto-parietal network). These results could point to the important involvement of these deep brain structures in addiction and long lasting craving responses (Miquel et al., 2016) or more stable features of craving similar to craving as a clinical characteristic (Yu et al., 2022). Additional conceptual considerations are important. The extent to which craving as a feature of substance use disorders is underpinned by the same neural mechanisms as craving in behavioral addictions and food craving in obesity or healthy individuals warrants consideration. Although overlaps could be identified across behaviors, there were also substantial differences.
Data suggest that craving might have different components involving predominantly cognitive/obsessive, automatic, or physiological elements, respectively (Antons, Liebherr, Brand, & Brandtner, 2023; Flaudias, Heeren, Brousse, & Maurage, 2019). While craving across addictive disorders and potentially related conditions (e.g., substance use disorders, behavioral addictions and obesity/binge-eating disorder) seem to have common components related to automatic, reactive processes such as action impulses (involving for example the motor/sensory network and subcortical/basal-ganglia network), they might differ in cognitive/obsessive or physiological components (involving for example a network centered around the fronto-parietal network), as can be seen by the comparison of canonical networks identified in the studies (Fig. 1). Future studies and theories should consider automatic, reactive components of craving as central mechanisms that might be represented by interactions between the subcortical/basal-ganglia network, the motor/sensory network and the visual I network in the processing of craving. These results add to theories that center around the relevance of the ventral and dorsal striatum in addictive behaviors (Brand, 2022; Brand et al., 2019; Everitt & Robbins, 2013). The interplay between the subcortical/basal-ganglia network and further networks (motor/sensory and visual I networks) might be highly relevant as they combine, for example, the reward mechanisms with motor execution that could explain lapses during abstinence. In previous work where CPM was used to predict cocaine abstinence, a high integration between the motor/sensory network, the subcortical/basal-ganglia network and the salience network has been proposed to relate to the coordination of salience encoding and reward responding (Yip, Scheinost, Potenza, & Carroll, 2019). Based on the current and prior results, a core neural foundation of craving might reflect reward-driven motor impulses. It should be noted that the comparison of the studies presented here were of qualitative nature. However, the comparison and consolidation of findings promotes a deeper understanding of the construct of craving and which factors have been identified when using data-driven methods.
The consolidation of studies using CPM to predict craving across different behaviors indicate that processing within subcortical/basal-ganglia, motor/sensory and visual I networks may represent the neural core of craving experiences. When using data-driven methods such as CPM, it is important to consider possible effects of groups, tasks and other factors when investigating clinical constructs like craving. However, similar findings across the studies suggest common elements that might most strongly reflect networks implicated in craving, providing more robust and rigorous insight into brain networks underlying craving.
Conflict of interest
The authors report no conflicts of interest with respect to the content of this manuscript. SA receives funding from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation); has edited journals and journal sections; has given academic lectures in clinical or scientific venues. MNP has consulted for Opiant Therapeutics, Game Day Data, Baria-Tek, the Addiction Policy Forum, AXA, Boehringer Ingelheim, and Idorsia Pharmaceuticals; been involved in a patent application with Yale University and Novartis; received research support from Mohegan Sun Casino, Children and Screens and the Connecticut Council on Problem Gambling; participated in surveys, mailings, or telephone consultations related to drug addictions, internet use, impulse-control disorders, and other health topics; consulted for or advised gambling, non-profit and legal entities on issues related to internet use, impulse-control and/addictive disorders; and given academic lectures in grand rounds, continuing medical education events, and other clinical or scientific venues. MB receives funding from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation), the EU, and the German Federal Ministry of Education and Research; has performed grant reviews for research-funding agencies; has edited journals and journal sections; has given academic lectures in clinical or scientific venues; and has generated book chapters for publishers of mental health texts. SA, MB, and MNP are associate editors of the Journal of Behavioral Addictions.
Funding sources
The work by SA and MB on this article was conducted in the context of the Research Unit ACSID, FOR2974, funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – 411232260.
References
Antons, S., Brand, M., & Potenza, M. N. (2020). Neurobiology of cue-reactivity, craving, and inhibitory control in non-substance addictive behaviors. Journal of the Neurological Sciences, 415, 116952. https://doi.org/10.1016/j.jns.2020.116952.
Antons, S., Liebherr, M., Brand, M., & Brandtner, A. (2023). From game engagement to craving responses – The role of gratification and compensation experiences during video-gaming in casual and at-risk gamers. Addictive Behaviors Reports, 18, 100520. https://doi.org/10.1016/j.abrep.2023.100520.
Antons, S., Yip, S. W., Lacadie, C. M., Dadashkarimi, J., Scheinost, D., Brand, M., & Potenza, M. N. (2023). Connectome-based prediction of craving in gambling disorder and cocaine use disorder. Dialogues in Clinical Neuroscience, 25(1), 33–42. https://doi.org/10.1080/19585969.2023.2208586.
Brand, M. (2022). Can internet use become addictive? Science, 376(6595), 798–799. https://doi.org/10.1126/science.abn4189.
Brand, M., Wegmann, E., Stark, R., Müller, A., Wölfling, K., Robbins, T. W., & Potenza, M. N. (2019). The Interaction of Person-Affect-Cognition-Execution (I-PACE) model for addictive behaviors: Update, generalization to addictive behaviors beyond internet-use disorders, and specification of the process character of addictive behaviors. Neuroscience & Biobehavioral Reviews, 104, 1–10. https://doi.org/10.1016/j.neubiorev.2019.06.032.
Brandtner, A., Antons, S., Cornil, A., & Brand, M. (2021). Integrating desire thinking into the I-PACE model: A special focus on Internet-use disorders. Current Addiction Reports. https://doi.org/10.1007/s40429-021-00400-9.
Dong, G., Zheng, H., Liu, X., Wang, Y., Du, X., & Potenza, M. N. (2018). Gender-related differences in cue-elicited cravings in Internet gaming disorder: The effects of deprivation. Journal of Behavioral Addictions, 7(4), 953–964. https://doi.org/10.1556/2006.7.2018.118.
Everitt, B. J., & Robbins, T. W. (2013). From the ventral to the dorsal striatum: Devolving views of their roles in drug addiction. Neuroscience & Biobehavioral Reviews, 37(9, Part A), 1946–1954. https://doi.org/10.1016/j.neubiorev.2013.02.010.
Finn, E. S., Shen, X., Scheinost, D., Rosenberg, M. D., Huang, J., Chun, M. M., … Constable, R. T. (2015). Functional connectome fingerprinting: Identifying individuals using patterns of brain connectivity. Nature Neuroscience, 18, 1664–1671. https://doi.org/10.1038/nn.4135.
Flaudias, V., Heeren, A., Brousse, G., & Maurage, P. (2019). Toward a triadic approach to craving in addictive disorders: The metacognitive hub model. Harvard Review of Psychiatry, 27(5), 326–331. https://doi.org/10.1097/hrp.0000000000000225.
Garrison, K. A., Sinha, R., Potenza, M. N., Gao, S., Liang, Q., Lacadie, C., & Scheinost, D. (2023). Transdiagnostic connectome-based prediction of craving. American Journal of Psychiatry, 180(6), 445–453. https://doi.org/10.1176/appi.ajp.21121207.
Hallam, J., Boswell, R. G., DeVito, E. E., & Kober, H. (2016). Gender-related differences in food craving and obesity. Yale Journal of Biology and Medicine, 89(2), 161–173.
Hill-Bowen, L. D., Riedel, M. C., Poudel, R., Salo, T., Flannery, J. S., Camilleri, J. A., … Sutherland, M. T. (2021). The cue-reactivity paradigm: An ensemble of networks driving attention and cognition when viewing drug and natural reward-related stimuli. Neuroscience & Biobehavioral Reviews, 130, 201–213. https://doi.org/10.1016/j.neubiorev.2021.08.010.
Mallorquí-Bagué, N., Mestre-Bach, G., & Testa, G. (2023). Craving in gambling disorder: A systematic review. Journal of Behavioral Addictions, 12(1), 53–79. https://doi.org/10.1556/2006.2022.00080.
Miquel, M., Vazquez-Sanroman, D., Carbo-Gas, M., Gil-Miravet, I., Sanchis-Segura, C., Carulli, D., … Coria-Avila, G. A. (2016). Have we been ignoring the elephant in the room? Seven arguments for considering the cerebellum as part of addiction circuitry. Neuroscience & Biobehavioral Reviews, 60, 1–11. https://doi.org/10.1016/j.neubiorev.2015.11.005.
Shen, X., Finn, E. S., Scheinost, D., Rosenberg, M. D., Chun, M. M., Papademetris, X., & Constable, R. T. (2017). Using connectome-based predictive modeling to predict individual behavior from brain connectivity. Nature Protocols, 12(3), 506–518. https://doi.org/10.1038/nprot.2016.178.
Tiffany, S. T., & Wray, J. M. (2012). The clinical significance of drug craving. Annals of the New York Academy of Sciences, 1248(1), 1–17. https://doi.org/10.1111/j.1749-6632.2011.06298.x.
Vafaie, N., & Kober, H. (2022). Association of drug cues and craving with drug use and relapse: A systematic review and meta-analysis. JAMA Psychiatry, 79(7), 641–650. https://doi.org/10.1001/jamapsychiatry.2022.1240.
van Timmeren, T., van Holst, R. J., & Goudriaan, A. E. (2023). Striatal ups or downs? Neural correlates of monetary reward anticipation, cue reactivity and their interaction in alcohol use disorder and gambling disorder. Journal of Behavioral Addictions, 12(2), 571–583. https://doi.org/10.1556/2006.2023.00015.
Yang, W., Han, J., Luo, J., Tang, F., Fan, L., Du, Y., … Liu, J. (2023). Connectome-based predictive modelling can predict follow-up craving after abstinence in individuals with opioid use disorders. General Psychiatry, 36(6), e101304. https://doi.org/10.1136/gpsych-2023-101304.
Yarkoni, T., Poldrack, R. A., Nichols, T. E., Van Essen, D. C., & Wager, T. D. (2011). Large-scale automated synthesis of human functional neuroimaging data. Nature Methods, 8(8), 665–670. https://doi.org/10.1038/nmeth.1635.
Yip, S. W., Kiluk, B., & Scheinost, D. (2019). Toward addiction prediction: An overview of cross-validated predictive modeling findings and considerations for future neuroimaging research. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, 5(8), 748–758. https://doi.org/10.1016/j.bpsc.2019.11.001.
Yip, S. W., Lichenstein, S. D., Liang, Q., Chaarani, B., Dager, A., Pearlson, G., … Garavan, H. (2023). Brain networks and adolescent alcohol use. JAMA Psychiatry, 80(11), 1131–1141. https://doi.org/10.1001/jamapsychiatry.2023.2949.
Yip, S. W., Scheinost, D., Potenza, M. N., & Carroll, K. M. (2019). Connectome-based prediction of cocaine abstinence. American Journal of Psychiatry, 176(2), 156–164. https://doi.org/10.1176/appi.ajp.2018.17101147.
Yu, F., Li, J., Xu, L., Zheng, X., Fu, M., Li, K., … Becker, B. (2022). Opposing associations of internet use disorder symptom domains with structural and functional organization of the striatum: A dimensional neuroimaging approach. Journal of Behavioral Addictions, 11(4), 1068–1079. https://doi.org/10.1556/2006.2022.00078.
Zeng, X., Han, X., Gao, F., Sun, Y., & Yuan, Z. (2023). Abnormal structural alterations and disrupted functional connectivity in behavioral addiction: A meta-analysis of VBM and fMRI studies. Journal of Behavioral Addictions, 12(3), 599–612. https://doi.org/10.1556/2006.2023.00025.
Zhou, W. R., Wang, Y. M., Wang, M., Wang, Z. L., Zheng, H., Wang, M. J., … Dong, G. H. (2022). Connectome-based prediction of craving for gaming in Internet gaming disorder. Addiction Biology, 27(1), e13076. https://doi.org/10.1111/adb.13076.
Zilverstand, A., Huang, A. S., Alia-Klein, N., & Goldstein, R. Z. (2018). Neuroimaging impaired response inhibition and salience attribution in human drug addiction: A systematic review. Neuron, 98(5), 886–903. https://doi.org/10.1016/j.neuron.2018.03.048.
Zilverstand, A., O’Halloran, R., & Goldstein, R. Z. (2018). Resting-state and structural brain connectivity in individuals with stimulant addiction. A systematic review. In H. Pickard, & S. Ahmed (Eds.), The Routledge handbook of the philosophy and science of addiction (pp. 362–379). London, New York: Routledge.
