Introduction

A range of modern interventions show promise for facilitating alcohol use disorder (AUD) treatment, including pharmacotherapies (acting at opioid, glutamate, GABA, 5HT, and acetylcholine receptors) and behavioral-psychological interventions (e.g., brief interventions, motivational interviewing, contingency management, cognitive behavioral therapy, mindfulness-based approaches, and computerized, mobile, and web-based methods); for recent reviews, see [1, 2]. Although the newest methods await further testing, contemporary modern techniques will still require considerable improvement to achieve high efficacy for long-term relapse prevention. A sizable portion of AUD patients still return to drinking within 6 months post-treatment, with disappointing progress on efficacy over the past 40 years [3,4,5,6]. The critical need for effective treatments has driven the creative development of a range of interventions. Identifying common brain areas sensitive to such interventions could facilitate refining these treatments, lead to innovative new treatments, and provide better spatial targeting for neuromodulatory manipulations (e.g., as required for transcranial magnetic stimulation; TMS). The present review is aimed at identifying potential brain targets for therapeutic interventions in AUD.

A key element of AUD is a biased preference toward highly salient alcohol rewards and away from more abstract rewards such as human relationships, health, and career opportunities [7]. Therefore, promising treatments should act to reverse these tendencies. Recent neurofunctional domain-based nosology identifies three key domains of AUD and other addictions: incentive salience, executive function, and negative emotionality [8, 9]. Insofar as targeting two interrelated domains should increase the therapeutic potential of any putative interventions, we review the nature of brain systems governing i) the power of conditioned cues to motivate behavior, ii) the ability of these cues to distort executive function (influence decision-making), and iii) myopic reward decision-making. We limit the scope of the present review to the two domains—incentive salience and executive function—most implicated in positive reinforcement-related AUD. These two domains are especially implicated in positive reinforcement-related AUD subtypes, e.g., high novelty-seeking and impulsivity subtypes such as the previously defined type II (early-onset, familial alcoholism, high novelty-seeking, low harm avoidance; [10]), type B (early-onset, familial alcoholism, polydrug use, greater psychopathology; [11]), or “reward drinking” phenotype [12,13,14], predicting poorer recovery outcomes [15]. Further, these domains interact in brain regions key to reward decision-making, as recently demonstrated in PCC/precuneus of heavy drinkers with alcohol taste cues and delay discounting [16]. Using spatial meta-analytic techniques, we report brain regions responding to interventions on cue reactivity and delay-of-gratification, representing the incentive salience and executive function domains, respectively.

Incentive Salience: Cue Reactivity

Through Pavlovian learning and the repeated pairing of intoxication with alcohol’s associated sensory stimuli, alcohol’s conditioned cues acquire the capacity to alter motivational systems in the brain (particularly in mesostriatal dopamine systems). As a result, such learning can create lasting changes that induce “wanting” and invigorate seeking behavior (for reviews, see [17, 18]). Alcohol-associated cues can thus increase drinking [19, 20], elicit physiological responses in AUD that predict later drinking frequency [21], and provoke greater skin conductance and craving in AUD versus healthy controls [22].

Brain responses are a special class of conditioned responses insofar as they are in a position to modulate decision-making and behavioral action. The dominant method for studying this in humans is fMRI, which is but a hemodynamic proxy for neuronal activation. Nonetheless, there is evidence from positron emission tomography that alcohol-associated sensory stimuli do promote striatal dopamine transmission in humans and that the fMRI response in the striatum at least partially reflects this [23,24,25]. More broadly, alcohol cue-activated brain reward systems predict drinking [26] and relapse [27], and differentiate subjects who transition to heavy drinking [28]. Cue-induced brain activity also correlates with AUD severity [29] and reflects abstinence [30]. Therefore, any successful treatment for AUD might be expected to alter cue-elicited responses, particularly attenuating reward-linked limbic responses, and/or enhancing executive regions [31]. The extant literature thus indicates that neuroimaging of cue reactivity is a promising index of treatment efficacy in the incentive salience domain.

Executive Function: Delay-of-Gratification

Evocation of the brain’s motivational system is, however, but one aspect of a larger behavioral ensemble. Adaptive behavior in the environment requires that the urges elicited by alcohol’s (and other reward’s) cues be appropriately restrained and regulated. Impaired regulatory capacity may reflect one aspect of an inherited predisposition to AUD, which entails an altered set of executive abilities (e.g., 32, 33,34,35), in addition to any impairment induced by alcohol itself (36 for review). While tempting to regard AUD as largely a function of alcohol reward—i.e., that alcohol’s elevated appetitive value in some people drives compulsive use—one key observation highlights the primary role of adaptive executive regulation: the majority of Americans derive reward from alcohol (71% past-year drinking), but only a small fraction engage in pathological use (6% AUD; 37). Considerable work implicates impaired delay-of-gratification in AUD [38,39,40,41,42,43,44] and other addictions. Beyond alcohol, this tendency scales with addiction severity across a range of substances and risk measures (for meta-analysis, see 45), suggesting a phenotypic marker for addiction [46]. Delay discounting tasks simultaneously measure immediacy preference and delay aversion, which together with drug over-valuation, represent a high-risk phenotype [47]. Discounting robustly predicts post-treatment smoking relapse [48], and treatment response [49,50,51,52], and thus offers a potentially viable therapeutic target [53]. Given the clinical importance of behavioral discounting processes to AUD and recovery, we extend the work of several comprehensive meta-analyses of brain regions associated with delay discounting [54,55,56,57] by focusing specifically on interventions targeting delay-of-gratification.

Rationale

With increasing interventions for AUD and other substance use disorders that are designed to target specific brain loci [58], and the concomitant need for spatial targets, we endeavored to map brain locations most likely to respond to interventions targeting cue reactivity and delay discounting, given their importance in addiction processes and recovery.

Objectives

We used spatial meta-analytic techniques for fMRI to examine studies employing manipulations designed to attenuate alcohol cue reactivity and delay discounting in a range of drinkers, from healthy controls to heavy drinkers. Both pharmacological and behavioral interventions were employed in both domains. We prioritized manipulations designed to effect long-term change, and so selected studies using pre- versus post-intervention designs with stringent inclusion criteria, assuming that these represented manipulations with potentially lasting efficacy, and therefore clinical utility. Unfortunately, the nascent field of enhancing delay-of-gratification does not yet contain pre- versus post-intervention studies; therefore, we opted to include acute manipulations of discounting as a first-pass assessment of brain regions likely to be involved in this critical executive process. Although the methods and samples of the reviewed manuscripts differ in important ways, we will conduct these analyses in parallel, and compare the independent results for convergence and divergence of these anatomic systems supporting related domains. Our outcome of interest, in all cases, was a change in brain activation corresponding to the manipulation, specifically within-subject or group interactions in altered brain activity. Our present review of findings from interventions that change activation is therefore designed to identify brain regions jointly sensitive to manipulations of conditioned alcohol reward and increased delay-of-gratification.

Methods

We report this review in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelinesFootnote 1 [59], although note that the lack of fMRI variability estimates precluded heterogeneity estimates.

Eligibility Criteria

To be considered for inclusion, studies had to meet the following criteria: 1) performed an intervention targeting conditioned alcohol reward or monetary delay discounting; 2) used human laboratory paradigms with alcohol cue exposure or delay discounting during fMRI; 3) reported pre- versus post-intervention fMRI results in within-subjects or mixed designs, i.e., studies using a single session intervention were excluded: studies using single session interventions on delay discounting were included; 4) reported whole-brain general linear model-based voxel-wise fMRI results in standardized space (only the whole-brain results were used in studies reporting both whole-brain and a priori region of interest [ROI]-based analyses); 5) published in a peer-reviewed English language journal. The intent was to identify brain locations responding after an intervention, i.e., pre- versus post-intervention for both cue reactivity and delay discounting; however, an initial search revealed that the published delay discounting fMRI intervention studies used within-session experimental manipulations only (except [60]). Although imperfectly matched, these comparisons should inform spatial activation patterns for both classes of interventions.

Sources, Searches, and Study Selection

Our primary database searches were conducted in PubMed (https://www.ncbi.nlm.nih.gov/pubmed/), with secondary searches of Google Scholar (https://scholar.google.com/), and follow-up from references listed in relevant reviews.

We conducted literature searches (up to August 29, 2019) and identified potential studies with the following two-step procedure. For cue reactivity studies: 1) PubMed searches of the title/abstract using the following search terms: “fMRI” AND “cue reactivity” AND “alcohol”; “imaging” AND “cue reactivity” AND “alcohol”; “training” AND “alcohol cues” AND “alcohol dependent”; “cues” AND “reduced” AND “alcohol” AND “fMRI”; 2) on Google Scholar searching within the whole article using the following search terms: “MRI”, “alcohol dependent” OR “alcohol use disorder”, “alcohol cue reactivity.” The delay discounting papers were identified with 1) PubMed searches of the title/abstract using the following search terms: “discounting” AND “intervention” AND “fMRI”; “delay discounting” AND “clinical trial”; “reduces” AND “delay discounting” AND “fMRI”; “fMRI” AND “intertemporal” AND “episodic”; “intertemporal choice” AND “future rewards” OR “later rewards” AND “imaging” AND “task” AND “decision”; 2) on Google Scholar searching within the whole article using the following search terms: “delay discounting,” “reduce,” “decision,” “treatment,” “fMRI,” “comparison,” “pre,” “post,” “coordinates.” These papers were carefully screened, and all suitable papers were subjected to reverse lookup (“cited by”) on Google Scholar to identify additional papers. Relevant reviews and meta-analyses were consulted to ensure complete inclusion. One author (YIS) conducted both full literature searches, with two authors (BGO and YIS) reviewing all final records for inclusion. Records raising questions for inclusion required consensus to finalize. Data extraction and entry was independently verified by two authors.

Data Collection and Summary Measures

Activation peak coordinates resulting from the intervention were the outcomes of interest and were collected from published tables. In all but one study, Z, t, or F statistics were reported for the corresponding peaks; these were used to calculate Cohen’s d. One effect size was estimated [61] for the study that did not report it [60].

fMRI Outcomes: Activation Likelihood Estimation

Coordinate-based meta-analysis permits estimating the likelihood of a region’s involvement in a particular phenomenon and increases the generalizability of findings from diverse neuroimaging studies. ALE models activation peaks from multiple studies as spatial probability distributions in the whole brain—permitting random effects inference to the study populations. We performed ALE analyses with GingerALE [62] v3.0.2 (http://brainmap.org/ale/) in MNI space, with a threshold of puncorr < 0.001, and extent of 100 mm3. We relied on ALE primarily to identify the loci of peak spatial convergence, but we additionally performed permutation-based cluster correction. Although we recognize that there is an insufficient number of published studies meeting our selection criteria for a well-powered activation likelihood estimation (ALE) meta-analysis [62, 63], we present findings from a range of intervention studies and use ALE to estimate the centroids of the results reported. Coordinates originally reported in Talairach space were transformed to MNI prior to ALE.

Risk of Bias

The presence of publication bias was assessed separately for cue reactivity and delay discounting with the visual detection of asymmetry in funnel plots (study ns against effect sizes). Cohen’s d effect sizes were calculated from published Z or t-statistics using the highest activation peak statistic in each study, representing the greatest potential for bias (e.g., 64). Right-skewed plots indicate the possibility of publication bias favoring significant results, particularly with smaller sample sizes. This was followed by Orwin’s fail-safe N analysis [65] to determine the number of missing studies with null results required to reduce the study sample mean to a small effect size (d ≤ 0.2; [66]). Given that the outcomes of interest are activation peaks from fMRI analyses, the threats to individual study bias exist largely in fMRI thresholding and analyses, which is discussed in narrative form.

Results

Study Characteristics and Subjects

The study identification and screening flow chart is shown in Fig. 1. The searches returned 472 results for cue reactivity, and 233 results for delay discounting in total from both PubMed and Google Scholar. Reasons for exclusion included no relevant sample, review paper, not an fMRI study, no pre-/postintervention comparison for cue reactivity, and no targeted intervention for delay discounting. Other exclusions listed in Fig. 1 are 1) not reporting significant peaks in the pre/post alcohol–neutral contrast and thus therefore unusable for ALE [67], and 2) presented results obtained during operant behavior rather than simple cue reactivity [68]. After filtering all potential studies for our inclusion criteria, the literature search yielded a total of 10 studies (284 subjects and 87 foci) for cue reactivity and 9 studies (206 subjects and 81 foci) for delay discounting; details are summarized in Tables 1 and 2, respectively. Note that one study [69] using neurofeedback in a single session was included among the cue reactivity studies, as they measured activation outcomes pre-/postintervention in a first versus last scan analysis. Note that a subset of n = 13 in one study [70] received a separate intervention and fMRI in addition to the main results in n = 23, so the peaks for those were separately included in ALE as n = 13. Studies reporting only functional connectivity results were not included (e.g., [71]), as these cannot be used in ALE analyses.

Fig. 1
figure 1

Study selection and inclusion

Full size image
Table 1 Alcohol cue reactivity intervention studies
Full size table
Table 2 Delay discounting intervention studies
Full size table

Cue reactivity studies tested currently drinking or abstinent AUD subjects or heavy drinkers (HD), with all subjects sober during fMRI. Subjects were characterized as AUD by DSM-IV, ICD-10, or the Mini International Neuropsychiatric Interview. HD were defined by AUDIT scores above 8 or exceeding 2 heavy drinking days per week, and ≥ 21 or 14 drinks per week for men or women, respectively. Our interest in intervention effects precluded studies evaluating the effects of abstinence alone (e.g., [72]). Just one cue reactivity study included control subjects. Most of the delay discounting studies used only healthy subjects, except one using abstinent AUD [60], and another with pathological gamblers [73]. One delay discounting intervention study otherwise meeting criteria was excluded due to likely nonspecific brain deficits in pre-dementia patients [74]. Finally, these analyses did not require protection of human subjects review, as all data were previously published and contained no confidential information.

Paradigms, Stimuli, and Interventions

All the cue reactivity paradigms used images of alcohol drinks (one with additional olfactory cues) designed to elicit alcohol-associated conditioned reward, and control images of either non-alcohol drinks or a visually matched abstract image, some from the International Affective Picture database [75]. Interventions used on cue reactivity spanned a wide range, including putative pharmacotherapies, magnetic or direct current stimulation, behavioral extinction training, or neurofeedback training. Interventions were performed after the first fMRI scan (except the neurofeedback study, which was concurrent to scanning) and were assessed for effects on cue reactivity by within-subjects comparisons to the postscan.

The delay discounting paradigms used similar binary choice paradigms comprising a choice between a smaller/sooner (e.g., “now”) or a larger/later delayed monetary reward. All non-pharmacological interventions presented stimuli that were designed to increase preference for the delayed reward—these were displayed during, or just preceding, the choice period during monetary delay discounting in fMRI. The majority of the DD intervention studies (6 out of 9) employed versions of episodic future thinking (EFT). By invoking a personalized, relevant future, EFT cues are intended to increase preference for future rewards by eliciting brain activation related to the abstract valuation of a future event. The EFT studies reported here used extensive pre-scan interviews to identify specific events relevant for each subject that were planned for the near future (e.g., “vacation paris,” “friend’s wedding,” “mum’s birthday”; [76]). These were later presented during fMRI as text inserted into the delay discounting paradigm immediately preceding choice. The episodic trials were compared with visually similar control trials with placeholder strings. Other EFT variations used an “Imagine” (spending the money) versus an “Estimate” (what the money could buy) condition [77] or episodic future café meetings [78, 79]. Non-EFT interventions used images of emotional faces versus neutral faces [80] or naltrexone versus placebo [60]. Although some interventions, such as emotional faces, would not likely form the basis of an AUD treatment (but see “Discussion”), our interest was in discovering brain regions sensitive to manipulations of DD by various methods.

Interventions on Cue Reactivity: fMRI

Convergent brain regions showing intervention effects on alcohol cue reactivity were localized mostly to the left hemisphere and included the left dorsal caudate, orbitofrontal cortex/subgenual anterior cingulate cortex, left temporal pole, left frontal operculum, and thalamus; Fig. 2, Table 3. High peak density was observed in the ventromedial prefrontal cortex (Fig. 1, lower left) and ventral striatum (Fig. 1, right), but spatial variability prevented the ventral striatum from achieving ALE convergence. This may be unsurprising, as a large number of exclusively ROI-based cue reactivity studies focused on these regions, particularly the ventral striatum, by extracting averaged activity from previously defined anatomical regions of interest. As above, these studies (e.g., [81, 82,83,84]) were not included as they do not lend themselves to the ALE meta-analytic technique. However, this body of cue reactivity work using a priori ventral striatal ROIs is generally consistent in showing attenuation to alcohol/drug cues using successful interventions (for reviews, see [31, 85]).

Fig. 2
figure 2

Alcohol cue reactivity interventions. Peaks from individual studies reporting responses to interventions for attenuating conditioned alcohol reward (4 mm radius spheres; yellow); convergent regions from ALE in red (puncorr < 0.001, extent = 100 mm3). White lines (center, sagittal plane) indicate slices displayed in axial (left) and coronal (right) views. MNI coordinates in white (left z, middle x, right y)

Full size image
Table 3 Cue reactivity intervention ALE results
Full size table

Interventions on Delay Discounting: fMRI

Manipulations targeting delay-of-gratification showed convergent results mostly along default mode midline structures, from medial and ventromedial prefrontal, anterior and posterior cingulate, and precuneus, with additional temporal and parietal findings; Fig. 3, Table 4.

Fig. 3
figure 3

Delay discounting interventions. Peaks from individual studies reporting responses to interventions for increasing delayed reward preference (4 mm radius spheres; yellow); convergent regions from ALE in red, (puncorr < 0.001, extent = 100 mm3). MNI coordinates shown in white (z, except x lower right)

Full size image
Table 4 Delay discounting intervention ALE results
Full size table

Co-occurrence of Intervention Effects

Two convergence peaks from the two ALE analyses on cue reactivity and discounting were immediately adjacent in vmPFC but did not spatially overlap (although their extent may be limited by low available power). The pattern of results from these two types of interventions shows overlapping individual peaks (Fig. 4A) and spatial co-localization of ALE results in vmPFC (Fig. 4B).

Fig. 4
figure 4

Intervention effects in vmPFC. (A) Individual peaks from incentive salience and delay-of-gratification intervention studies co-localized in ventral prefrontal cortex from subgenual ACC to frontopolar cortex (blue ellipse); (B) ALE results (puncorr < 0.001, extent = 100 mm3) indicate vmPFC sensitivity to interventions. Interventions on alcohol cue reactivity and delay discounting are shown in red and green, respectively, for individual peaks (A) and ALE clusters (B)

Full size image

Risk of Bias

Visual inspection of funnel plots suggested that both the cue reactivity and delay discounting study samples contained probable publication bias favoring positive effect sizes (Fig. 5). Orwin’s fail-safe N analyses indicated that 53 and 69 studies would be required to reduce the mean effect sizes to 0.2 for cue reactivity and delay discounting, respectively. Note that one outlier in the delay discounting studies with d = 4.73 [77] strongly influenced the mean, and reduced the required study number from 69 to 52 when removed. Two substantial sources of potential bias in neuroimaging studies affecting the weight of evidence are sample size and imaging threshold value. Consistent with the ongoing trend for requiring greater subject numbers to meet stricter thresholds, our reviewed studies show a strong trend for larger samples with time (study n and publication year); r(17) = .45, p = .053. Half of the cue reactivity studies used uncorrected whole-brain thresholds, with the other half corrected either voxelwise or by using a cluster size requirement (garnering significant methodological discussion in recent years, (e.g., [86])); 78% of the delay discounting studies used corrected thresholds. Cue reactivity and delay discounting studies did not differ in sample sizes, p > .2; mean ns 28.4 ± 8.0 and 22.9 ± 10.1, respectively. Methodological differences particularly between randomized controlled trials (RCT; 6 studies) and others (13 studies) may have introduced bias. We might suspect smaller effect sizes in RCTs due to stricter randomization and more evenly distributed variability. Therefore, we tested for effect size differences between RCTs and non-RCTs across all studies reported here; t(17) = 1.39, p = .18, or in the cue reactivity studies alone, t(8) = 1.52, p = .17, but detected no differences. Regarding sample sizes for reported analyses, RCTs did not utilize more subjects than non-RCTs, t(17) = 0.48, p = .64, mean ns = 27.3 ± 5.3 and 25.1 ± 10.7 for RCT and non-RCT studies, respectively.

Fig. 5
figure 5

Funnel plots of Cohen’s d by sample size. Both types of intervention studies showed asymmetry, with underrepresentation of small studies with small effect sizes. Mean effect sizes indicated as dashed lines

Full size image

Discussion

The divergent methods used within these two categories of interventions or manipulations should identify convergent regions sensitive to treatments, yielding loci of responsive brain regions governing, 1) conditioned alcohol reward; 2) intertemporal choice (reward decision-making across the time domain, e.g., immediate vs. delayed rewards, or delay discounting); and 3) areas of overlap. Although these may appear to be divergent processes, both involve immediate reward signaling. However, intertemporal choice alone includes longer delays in pursuit of more abstract rewards. Thus, we may expect overlap of common processes, with divergence in brain regions relating chiefly to alcohol-specific reward, or to reward delay, respectively. Brain regions responding to both types of interventions thus represent putative reward decision-making targets germane to addiction treatment. With neural responses to treatment increasingly recognized as an objective metric for treatment efficacy [31], the loci identified here could facilitate the refinement of current treatments, and spatial targeting for future interventions. Publication bias was evident in both intervention types, but bias did not differ substantially between cue reactivity and delay discounting studies.

By examining effects of interventions targeting conditioned reward and decision-making in AUD, we aimed to identify brain areas responding to treatment-relevant interventions in two key domains of addiction—with the additional goal of identifying regions of overlap. Findings included evidence for intervention-related responses in 1) reward and salience networks, and dorsal striatum for cue reactivity interventions; 2) introspection and valuation-related areas for delay discounting interventions; and 3) co-occurrence within a key reward and valuation region: vmPFC. Although the ventral striatum did not reach statistical significance in the cue reactivity studies as we might expect from meta-analysis of cue-induced activation [85], we note that a number of peaks from individual studies occur within this area (Fig. 2, right), albeit with high spatial variability. Unexpectedly, neither frontoparietal network nor ventral striatal convergence was noted for the DD intervention studies, both of which are widely implicated in delay discounting processes [76, 87,88,89]. The general pattern of results here suggests that interventions on alcohol-conditioned cues are more specific to brain systems governing valuation and salience, whereas interventions on delay-of-gratification specifically are more concentrated in midline default mode regions. Importantly, co-occurrence between the two intervention types appears in the vmPFC, which is heavily implicated in subjective reward value [90] that underlies both incentive salience and reward decision-making processes [91,92,93]. Neither alcohol reward (and associated conditioning), nor impaired delay-of-gratification are likely sufficient for addiction, but both appear necessary. For example, a regular drinker who responsibly plans for the future does not necessarily suffer from AUD, nor does a non-drinking but irresponsible spender, necessarily, but the combination of high alcohol reward valuation and myopia for future consequences constitutes serious risk for addiction [94, 95]. Here, brain regions responding to interventions on both traits show common effects in the vmPFC, suggesting a possible common process, while regions specific to each (striatum and dorsal ACC in cue reactivity, and PCC/precuneus for delay discounting) may indicate brain processes specific to those functions.

Cue Reactivity

Over time, and through the process of Pavlovian learning, a reward’s associated sensory stimuli (e.g., sights, smells, tastes) become paired with (conditioned to) its reinforcing qualities (e.g., intoxication). These associations can be strong enough to invigorate reward seeking via “Pavlovian-to-instrumental transfer” (PIT; [96, 97]). The psychological and physiological responses to these alcohol-associated cues are thus implicated in both subjective craving and treatment relapse [21, 27, 98,99,100,101], but also see [102, 103]. Several decades of research in both animals and humans now show that reward- (drug-) cue exposure is itself sufficient to induce activity in the mesolimbic dopamine system [104], which is understood to partly comprise projections from the midbrain’s ventral tegmental area to the ventral striatum (ventral putamen and caudate, to include the nucleus accumbens; see [105] for an overview). Inactivation or dopamine blockade of these regions also eliminates cue-invigorated reward seeking (i.e., PIT; [106, 107, 108]).

Two predominant theoretical camps evolved to explain the behavioral and psychological significance of mesolimbic dopamine transmission from drug conditioned stimuli. The “incentive salience” camp [17, 18] focuses on the motivational (“drug wanting”) significance, while “prediction error” theorists [109, 110] focus on mechanisms of reward learning that is not necessarily linked to wanting. Both nevertheless recognize that activity in these regions reflect a learned relationship between drug (reward) and the drug’s associated sensory stimuli—and that this learning impacts drug-relevant behavior. With this background, it is hardly a surprise that many of the studies we found concentrate on the ventral striatum a priori.

At the same time, the mesolimbic dopamine circuit is only part of this system [111]. In particular, descending (glutamatergic) fibers project from frontal cortex to the striatum, with those from ventromedial and orbitofrontal cortex targeting the ventral and medial striatum. Activity in the ventromedial prefrontal area of convergence detected here (sometimes called medial orbitofrontal cortex) has been linked to perceived reward value [111,112,113,114,115,116,117], with ventral prefrontal cortex proposed to be an important site for “common neural currency” [118, 119] that works in concert with the ventral striatum to update reward value. Thus, the vmPFC’s relevance to conditioned reward and valuation is well-established, and its importance to alcoholism treatment and relapse is also now evident (Seo, et al., 2013).

Delay Discounting

Other recent meta-analytic findings suggest that mPFC and PCC both activate during discounting, and our results indicate that these regions respond to therapeutic interventions for increasing delay-of-gratification. Prior work [57] shows that activation within the impulsive “β system” (regions activating to immediate reward availability; [87]) closely maps to the same mPFC regions, albeit contralateral to the ventromedial prefrontal cortex (vmPFC) result. Similarly, “subjective reward” (the objective reward amount weighted by discount functions; [120]) maps onto both mPFC and posterior cingulate cortex (PCC) in regions close to our findings for regions sensitive to intervention. Contrary to conventional expectations, the locations of brain responses to delay-of-gratification interventions showed no convergence with the classical wanting/reward region (ventral striatum), or with dorsolateral prefrontal cortex/frontoparietal network (linked to executive control), suggesting that reward drive and executive control regions may be less involved than the midline default mode network for increasing delayed reward preference. Here, the finding unique to increased delay-of-gratification is the PCC/precuneus: the hub of the posterior default mode network [121]. Although the default mode network is widely thought to be more active at rest and to oppose the task-positive frontoparietal network, more recent work shows default mode coupling with the frontoparietal network for internally focused cognition [122]. The PCC also appears central to environmental monitoring/change detection, including when it involves reward choice [123, 124], suggesting its key role in decision-making. The PCC/precuneus is specifically involved in episodic memory retrieval [125], with the anterior default mode, i.e., vmPFC, relatively less implicated. Thus, the PCC/precuneus is deeply rooted in a host of functions critical to intertemporal choice—a complex decision process requiring the integration of memory, probability estimation, ambiguity assessment, perspective-taking of the future self, constructing hypothetical future environments, estimating future competing reinforcers, all within the context of dynamically changing contingencies to establish “goal value.” Interventions capable of increasing the subjective value of future rewards would be expected to enhance posterior default mode function, particularly in light of its putative role in change detection and policy selection [123].

Broadly, brain regions implicated in intertemporal choice in healthy subjects fall into default mode, frontoparietal and salience networks (as defined by [126]), with the majority of the peaks for subjective value, objective value, choice difficulty, and the (impatient) “β system” appearing in default mode regions, based on recent meta-analyses [57]. Peaks identified by Schüller and colleagues in the in β system and subjective value studies showed similar convergence to the current findings in midline default mode regions, in both anterior (vmPFC, mPFC, subgenual ACC) and posterior (PCC/precuneus) hubs. Thus, although we did not find frontoparietal or striatal reward/wanting peaks, as identified by Schüller and colleagues, it is plausible that brain effects of intervention are discrete from those used to actually perform discounting tasks. Here, we find intervention effects coalescing along midline anterior and posterior default mode regions, suggesting that these areas both actively participate in discounting, and are amenable to manipulation.

Accumulating evidence identifies impaired delay-of-gratification to be a common mechanism underlying AUD and other addictions as a trans-disease process [42, 45, 46, 127], and one that longitudinally predicts drug use/abuse [128,129,130], as well as treatment outcomes [49,50,51, 131, 132]. Episodic future thinking comprised the majority of the reviewed interventions on delay-of-gratification, and the majority of the EFT studies we reported included explicit instructions for subjects to elaborate and visualize the imagined future. While EFT scenarios often imply the participation of the self, explicitly including the self into intertemporal choice paradigms offers considerable therapeutic promise [133,134,135]. Highlighting the role of midline default mode regions in considering one’s future, subjects’ rostral ACC activation during present- and future-self-description predicted later delay discounting [134], and more similarity in activation to future and present self-description corresponded to less discounting. This suggested the possibility that discounting behavior indexed future self-continuity [135], such that mental constructions of one’s future containing more vivid and integrated future selves indicate greater future self-identification. Given the well-established role of the default network in introspective [121], prospective [136], and self-referential processing [137, 138], it is unsurprising that interventions for increasing delay-of-gratification act in midline default mode regions. The PCC may be particularly involved with introspective delay-of-gratification brain processes (but not necessarily alcohol-conditioned reward processes), and therefore could represent a site of action for techniques designed to enhance the salience of future rewards, e.g., future-oriented motivational interviewing-based methods [139, 140].

Co-localization

The convergent peaks of intervention effects on cue reactivity and delay discounting co-localize in the right vmPFC, at the locations of decision value and goal value (the net value of a reward decision-making action, and willingness to pay, respectively), as identified in a food purchase task [111]. Interestingly, the individual peaks from the two types of intervention studies lie along the border (z = − 16; Fig. 4) between the orbitofrontal and anterior medial default mode networks, as defined in resting state functional connectivity [126]. The functionally defined orbitofrontal network contains limbic cortex that responds to imagined reinforcer value at the time of choice [113, 114, 141], perceived pleasantness [142], alcohol wanting [23], and more broadly appears to be at the nexus of a range of hedonic experiences governed by primary and secondary reinforcers (for review, see [143]). Our current findings are located at the physical interface of these two key networks governing introspective decision-making, conditioned reward, and valuation.

We restricted our investigation to domains most implicated in positive reinforcement-related AUD, i.e., a “reward drinking” phenotype [12, 13]. Although the emotionality domain may appear to be largely independent [8, 9], important interactions with executive function and cue reactivity have been found. For example, limbic responses to emotional cues are differentially sensitive to modulation by alcohol cues in AUD versus controls [144], and the classically reward-associated midbrain is sensitive to emotional cues during an executive control task in AUD [145]. Germane to recovery outcomes, abstinent AUD showed less hippocampal connectivity than controls, with longer periods of sobriety corresponding with reduced fusiform activation to emotional face stimuli [146]. While there is little extant neuroimaging of negative emotionality and manipulations of delay-of-gratification, one of the reviewed studies indicated that an emotional fear prime was associated with dorsal ACC and PCC activation and increased delay-of-reward, which may be governed by the interaction between emotion and executive function [80].

Limitations

The study designs used in the reviewed cue reactivity studies reflect pre- versus postinterventions that may more closely resemble practicable therapeutic interventions than the delay discounting studies; we were limited by the published literature to acute interventions in these cases. Although the study designs were not truly analogous, we believe the more consistent convergence in the delay discounting interventions argues for confidence in those effects. Future studies of pre- and postintervention effects on delay-of-gratification for addiction to alcohol and other drugs will help to corroborate this observation. Our comparison of results from two populations (AUD/heavy drinkers in cue reactivity and mostly healthy controls in delay discounting) could limit our interpretations insofar as activation in regions governing delay discounting differ between healthy controls and AUD/heavy drinkers. Prior work in abstinent AUD and controls indicate more AUD activation during immediate choices in lateral orbitofrontal cortex relative to controls [147], and other work suggests that delay-of-gratification activation differs as a function of AUD severity in the supplementary motor area, insula/posterior orbitofrontal cortex, inferior frontal gyrus, cuneus [148], and superior frontal gyrus, paracingulate gyrus, and frontal pole [149]. These clusters lie outside the regions identified here, mitigating the concern of differential response by group. The only study reported here potentially informing this question is the Boettiger study [60], which found no group × naltrexone interaction in the ROI analysis. Ultimately, we believe that the approach remains informative insofar as we have no reason to believe that there are substantially different anatomic systems across populations (i.e., the functional anatomy should, by and large, be similar between populations, even if level of function is not). The ALE analyses were somewhat underpowered [63], limited by available studies. Although we acknowledge that robust ALE findings will require additional studies, we believe that this early-stage attempt to locate intervention effects is nonetheless worthwhile—and note that four of the discounting intervention clusters exceeded the recommended threshold. Some studies included in the analyses, particularly the older ones, utilized fMRI statistical thresholds lower than commonly accepted today [150]. While the older studies potentially inject more type I error into our analyses, this effect should theoretically be randomly located, and therefore have little overall effect on analyses of convergence (for excellent commentary on these issues, see [151]).

Conclusion

We identified regions responding to interventions, indicating therapeutic potential, that were common to both incentive salience and (a form of) executive function in the vmPFC. Additionally, we found convergent peaks from interventions specific to delay discounting in PCC/precuneus, indicating particular sensitivity to delay-of-gratification manipulations at this critical network hub. These early findings indicate that manipulations targeting reward decision-making should elicit effects in vmPFC, whereas manipulations designed to change future orientation or prospection should increase activity in vmPFC and PCC/precuneus. Thus, we hope that our findings provide a useful guide to facilitate validation and/or spatial brain targeting for future interventions, and ultimately therapeutic manipulations on AUD and addictions.