Forgetting can be highly functional when being reminded of unpleasant or even traumatic autobiographical experiences. Studies employing the think/no-think paradigm show that it is possible to forget unwanted memories by repeatedly attempting to suppress them (Anderson and Green, 2001; Anderson et al., 2004; Anderson and Levy, 2009; Anderson and Huddleston, 2012; but see Bulevich et al., 2006; Waldhauser et al., 2011). In this experimental approach, subjects learn paired associates and are subsequently presented with one member of those stimulus pairs (cue) and instructed to either suppress (no-think condition) or retrieve (think condition) the other member of the pairs (target; Anderson and Green, 2001). In subsequent recall tests, memory for suppressed targets is often significantly diminished when compared to baseline items that have been learned, but not cued to be either suppressed or retrieved between study and test phase.

The suppression of unwanted memories correlates with the recruitment of inhibitory control areas in the prefrontal and parietal cortices, and reduced retrieval-related activity in the hippocampus and sensory processing areas (Anderson et al., 2004; Depue et al., 2007, 2010; Butler and James, 2010; Dieler et al., 2010). In event-related potentials (ERPs) studies, memory suppression is reflected in a series of negative amplitude peaks that predict later forgetting, a reduction of the recollection-related late parietal positivity, and a modulation of ERP slow-waves at anterior frontal electrodes (Bergström et al., 2007, 2009a,b; Hanslmayr et al., 2009; Mecklinger et al., 2009).

Based on these data, current theoretical accounts assume that the decrease of memory performance is due to inhibitory control mechanisms that act by dampening memory traces of no-think items (Anderson and Green, 2001; Levy and Anderson, 2002; Anderson and Levy, 2009). The present study put the inhibition-hypothesis to a direct test by combining the think/no-think procedure with an old/new item recognition test. Item recognition can be achieved on the basis of familiarity, which reflects the general memory strength of the target item (Ratcliff, 1978; Yonelinas, 2002). If intentional suppression in the think/no-think paradigm leads to an inhibition of the no-think target, this should result in reduced strength of the item representation, and should lead to decreased recognition performance for no-think compared to baseline items (cf. Hicks and Starns, 2004; Spitzer and Bäuml, 2007; Spitzer et al., 2009). However, such an effect has not yet been demonstrated (Tomlinson et al., 2009).

By measuring ERPs during the think/no-think phase we investigated whether the presumed inhibition of no-think items also leads to a reduction of neural signals of memory strength. Previous think/no-think studies showed that memory suppression affects the ERP correlates of conscious recollection, occurring around 500 ms after onset of a memory cue (Bergström et al., 2007, 2009a,b; Mecklinger et al., 2009). This suggests that conscious retrieval is under strategic control and can be intentionally avoided during suppression attempts. However, recollection can be independent of the strength of the actual memory trace (Bergström et al., 2009a). Thus, it remains to be shown that memory suppression reduces the neural signature of the memory representation, which would indicate the inhibition of the target memory trace (cf. Spitzer et al., 2009). In studies on recognition memory, dissociable ERP correlates of item memory strength typically occur between 300 and 500 ms at frontal electrode sites (Gonsalves et al., 2005; Rugg and Curran, 2007; Stenberg et al., 2009; Woroch and Gonsalves, 2010). Likewise, studies on cued recall report a modulation of ERPs with successful retrieval after 300 ms (Allan et al., 1996; Friedman and Johnson, 2000). Contrasting think and no-think trials, we expected ERPs related to memory strength to be more negative-going in the no-think condition. These analyses should also reveal the abovementioned correlates of inhibitory control and reduced recollection (Bergström et al., 2007, 2009a,b; Mecklinger et al., 2009). We expected a decrease of memory strength to be related to later memory impairment in the final item recognition test. Crucially, inhibition should gradually decrease the strength of the to-be-suppressed memory trace with repeated suppression, ultimately resulting in the inaccessibility of target representations. Such a gradual reduction is suggested by studies showing a linear decrease of recall performance in dependence of the number of suppression attempts (Anderson and Green, 2001; Hanslmayr et al., 2009). In the present study, a reduction of memory strength was expected to result in progressively more negative-going ERPs to no-think trials over frontal electrode sites in later parts of the experiment.

The present study investigated the effect of memory suppression in the think/no-think paradigm on behavioral and electrophysiological correlates of memory strength. The repeated intentional suppression of unwanted memories led to memory impairment in an item recognition test and was reflected in more negative-going ERPs at frontal electrode sites. These findings are in line with inhibitory accounts of forgetting (Levy and Anderson, 2002; Anderson, 2005), suggesting that memory suppression can lead to the attenuation of memory trace strength at an item level. Novel methodological approaches, the employment of a recognition test, investigating memory impairment even for correctly recognized items by means of RT analysis, and showing repetition effects on ERP correlates of memory suppression, gave important insight on the different processes underlying the inhibition of unwanted memories.

To our knowledge, the present study is the first to show below-baseline suppression effects for no-think items in a recognition test. Memory suppression led to reduced discrimination performance for no-think items and this held even when constraining analyses to items that have been learned during study. Moreover, suppression not only affected accuracy measures but also led to longer reaction times for hits. Only two previous studies have investigated no-think effects on recognition memory. Marx et al. (2008) reported a no-think effect in a speeded recognition task. In that study, however, recognition performance for no-think items was compared with the think condition only and could merely indicate less gain for no-think items. Tomlinson et al. (2009) employed a forced-choice recognition test after the think/no-think phase and did not obtain suppression effects. This null effect may have resulted from the fact that recognition memory tests were preceded by recall tests and thus, any effects on recognition may have been counteracted by preceding retrieval attempts. Our finding corroborates the studies by Anderson and colleagues using independent probe tests (Anderson and Green, 2001; Anderson et al., 2004). These studies show that forgetting also occurs in response to recall cues that have a pre-experimental semantic relationship to the target items, but that have not been shown during the experiment (cf. Anderson and Green, 2001). Independent probe tests are considered to rule out that forgetting is due to a lowered effectiveness of the cue-target relationship in eliciting retrieval of a target. However, the study by Tomlinson et al. (2009) indicates that forgetting in independent probe recall tests can be due to interference. Engaging in a motor task upon cue presentation led to comparable forgetting in independent probe tests as intentional memory suppression (Tomlinson et al., 2009). Item recognition, in contrast, is dependent on the availability of the item memory trace alone, as trace strength is the only free variable in the matching computation between the study and test episodes (Ratcliff, 1978; Geiselman et al., 1983; Tulving, 1983; Bjork, 1989; Yonelinas, 2002). Thus, our results support the inhibition account and indicate that intentional suppression in the think/no-think paradigm results in the weakening of individual item memory traces, while circumventing the interpretational problems of independent probe tests (Hicks and Starns, 2004; Spitzer and Bäuml, 2007; Spitzer et al., 2009). As indicated by the no-think effects on RT, memory suppression does not only lead to forgetting, but also affects the retrieval of correctly recognized items.

Repeated retrieval practice for think items did not lead to an increase in recognition performance. For learned items, recognition accuracy was even numerically decreased in the think condition when compared to baseline. Interpretations of this pattern have to remain speculative at this point and await further studies that specifically investigate the effect of repeated retrieval practice on highly overlearned memories. It is possible that the excessive retrieval practice of the association between the cue and the target in the think/no-think phase rendered the test probe in the recognition test ineffective in eliciting a correct response for the single target item (Tulving and Thomson, 1973; Buchler et al., 2008). In line with our findings, memory performance is often not enhanced in independent probe tests that are designed to test memory strength of the target items alone (Anderson and Green, 2001; Anderson et al., 2004; Bergström et al., 2009b; Anderson and Huddleston, 2012). Importantly, as shown by correlation and covariance analyses, behavioral performance in the think condition was independent of forgetting in the no-think condition. The ERP data give further evidence for this independence of effects, since ERPs between think and no-think conditions differed substantially, and these differences predicted later memory impairment.

Memory suppression in the no-think condition led to more negative-going ERPs for the no-think condition over frontal scalp areas. We have strong reasons to assume that this effect reflects a reduction of ERP signals of memory strength. In the comparison of think and no-think conditions, the difference emerged in a time window around 300 ms. This matches studies on recognition memory that show that more positive-going ERPs (attenuated negativity) at frontal electrode sites in a comparable time window are associated with familiarity-based recognition judgments (Vilberg et al., 2006; Rugg and Curran, 2007; Stenberg et al., 2009). In line with this, more negative ERPs in the 300- to 350-ms time window indicate reduced recognition performance, as shown in our analysis investigating item-specific forgetting effects. While this naturally holds true for all items, irrespective of whether forgetting is due to memory suppression or not, the no-think condition entails a stronger reduction of memory strength. The ERP difference between think and no-think conditions correlated with later memory impairment between subjects and highest correlations were obtained at left frontal and parietal electrode sites at which neural activity is known to reflect memory strength (Gonsalves et al., 2005; Rugg and Curran, 2007; Stenberg et al., 2009). The spatiotemporal characteristics of this correlation pattern makes sense when assuming that the less positive-going ERPs during the think/no-think phase reflect the same process as the one that underlies reduced recognition performance (cf. Yonelinas, 2002). In addition to the item-specific effects, this result pattern indicates that a modulation of memory strength does not only relate to forgetting, but also correlates with reduced memory performance for correctly recognized items. Finally, ERPs between 225 and 450 ms at frontal electrode sites were more negative-going in later phases of the experiment, resembling the linear reduction in memory performance with increasing number of no-think trials observed in previous studies (Anderson and Green, 2001; Depue et al., 2006; Hanslmayr et al., 2009; Anderson and Levy, 2010). The behavioral reduction in recognition performance, the electrophysiological patterns, and the correlation between these measures strongly suggest that the no-think condition entails a gradual reduction of memory strength with repeated suppression attempts.

Replicating and extending previous studies, we observed ERP differences between think and no-think conditions in several time windows. We found an N2-like negativity between 450 and 500 ms. Differences in this time window predicted forgetting at an item level exclusively for the no-think condition. This indicates that the ERP peak reflects processes that are related to the active suppression of memory traces, corroborating the assumption that it signifies the recruitment of inhibitory control mechanism (Bergström et al., 2009a,b; Mecklinger et al., 2009). We also found indicators of reduced recollection for no-think items, emerging in less positive-going ERPs over parietal electrode sites between 500 and 600 ms (Bergström et al., 2007, 2009a; Mecklinger et al., 2009). ERP activity in this time window did not differentiate between remembered and forgotten items. This is in line with theories on recognition memory, stating that recollection is not a necessary process for item recognition (cf. Yonelinas, 2002). Contrasting our finding with the 300- to 350-ms effect, forgetting is observed irrespective of whether recollection was successfully avoided during the think/no-think phase or not. As in previous studies, neither the recollection-related effects, nor the inhibitory control component correlated with individual differences in memory impairment (Bergström et al., 2009a,b; Mecklinger et al., 2009). Furthermore, ERPs observed in these time windows did not vary with repeated suppression attempts. This implies that the ERPs observed in these time windows reflect processes that can be strategically employed, irrespective of actual memory strength of the to-be-suppressed target (Herron and Wilding, 2005; Cabeza et al., 2008; Bergström et al., 2009a; Mecklinger, 2010).

We found a frontal, positive-going slow wave for no-think items in the 650- to 900-ms time window (Bergström et al., 2007; Mecklinger et al., 2009). As reviewed by Mecklinger (2010), this slow wave may reflect processes acting in regulating the accessibility of unwanted memories, even if it does not directly correlate with memory suppression (Mecklinger et al., 2009). We also obtained a negative-going slow wave for no-think items in a late time window from 1400 to 1800 ms. This fits the results in a study by Hanslmayr et al. (2009), reporting sustained negative ERPs for no-think items that possibly indicate the strategic control and monitoring of memory retrieval. In addition to these late ERP effects, we found several early components to differ between think and no-think conditions. We observed more negative ERPs to no-think items between 175 and 225 ms. Corresponding to our results, Bergström et al. (2009b) found an early negativity for no-think items between 175 and 225 ms that was exclusively observed in participants that applied a direct suppression strategy to avoid thinking of the target, without retrieving substitutive memories. In contrast to Bergström et al. (2009b) we did not obtain a relationship of this ERP effect with later memory impairment. It is likely that this lack of correlation is due to the fact that we did not explicitly instruct subjects to use a direct suppression strategy. A relationship with a direct suppression strategy was confirmed in a post hoc correlation analysis with a post-experimental self-report questionnaire (Hertel and Calcaterra, 2005), assessing how often participants used a direct suppression strategy versus focusing on substitutive representations. We found that ERP differences between think and no-think conditions correlated with a direct suppression strategy at electrodes F3 and FC3 (r_s > 0.563, p < 0.05). Early differences between think and no-think conditions during the 100- to 140-ms and 225–250 ms time windows likely reflect higher attention to cues of the think condition, as extensively discussed by Bergström et al., 2007; cf. Mecklinger et al., 2009). In contrast to previous studies, these differences occurred earlier in the present data, already with an enhanced N1 peak for think items. The N1 often occurs around 100 ms, is thought to reflect top-down attentional control on early visual processing areas (e.g., Vogel and Luck, 2000), and has been shown to be sensitive to task instructions (Potts et al., 2004). The 225- to 250-ms pattern resembles an effect observed in previous think/no-think studies between 200 and 300 ms, and likely resembles a frontal selection positivity/posterior selection negativity to think items as discussed by Bergström et al. (2007). As indicated by the repetition analysis, the effect in the 225- to 250-ms time window overlapped with a reduction of memory strength for no-think items.

Finding several ERP components that differentiated between think and no-think conditions suggests that widespread control networks are involved in the successful suppression of unwanted memories (Depue, 2012). In order to achieve the identification of several ERP components and to distinguish them in their functional significance, we selected time windows based on previous research and inspection of the current dataset. The analysis strategy chosen for the present study conforms with a common statistical approach to ERP data (Picton et al., 2000; Dien and Santuzzi, 2005; Luck, 2005). However, selecting nine different time windows and statistical testing for each of these time windows, entails that some of the identified obtained statistical significances could in fact be false-positives (cf. Luck, 2005). Our ERP effects are to a large extent backed up by previous findings and the different ERP components appear to have distinguishable functional roles in memory suppression. Nonetheless, studies that selectively investigate some of the components described in the present study and that employ statistical methods that take care of multiple comparison problems are necessary to confirm our conclusions (cf. Bergström et al., 2009a,b).

Our results have important implications for the interpretation of memory suppression effects in the think/no-think paradigm. Most importantly, the observation of reduced recognition memory performance for no-think items, together with frontal phase-sensitive ERP modulation indicate that the no-think instruction can affect the memory strength of suppression targets at an item level. This strongly suggests that forgetting effects in think/no-think studies can be due to inhibition. Our measures of memory strength are robust against alternative explanations for no-think effects (Tomlinson et al., 2009). Neither performance in item recognition tests (Ratcliff et al., 1990), nor early, frontal ERP effects are sensitive to interference (Norman et al., 2008). It is possible that episodic interference, as pointed out by Tomlinson et al. (2009), can affect performance in recall tests that depend on conscious recollection and that are more susceptible to interference (Tulving, 1983; Bjork, 1989; Koriat and Goldsmith, 1996). Importantly, our data suggest that even if this can be the case, forgetting due to reduced memory strength is independent of a reduction of recollection. The attenuated late, parietal ERP effect indicating reduced recollection was not modulated by repeated suppression attempts, whereas the earlier, frontal positivity diminished over the time-course of the experiment. Also, early ERP differences correlated with memory impairment, at an item level and for retrieval of correctly recognized items, whereas ERPs in the recollection-related time window did not, further indicating that ERP reductions in this early time window tap into the memory processes that are important for a reduction in item strength. Our results add to the literature on the ERP correlates of cued recall. Studies in this field report more positive-going ERPs for successfully recalled items after 300 ms (Allan et al., 1996; Friedman and Johnson, 2000). We show that intentional suppression can affect these early correlates of recall for no-think items, but suggest that they are constituted by modulations of item memory strength that dissociate from strategically controlled recollection.

Previous neuroimaging studies on the think/no-think effect did not show such a dissociation. FMRI studies are sometimes not able to clearly dissociate between the subregions of the MTL involved in recollective and strength-based remembering (Kirwan et al., 2007). Thus, decreased MTL activity observed in fMRI studies employing the think/no-think paradigm could reflect both, reduced recollection as well as reduced memory strength (Anderson et al., 2004; Depue et al., 2007). As pointed out above, it is crucial to show a selective reduction of memory strength to conclude that the no-think effect involves inhibition, since a reduction in hippocampus-based recollection can be due to strategic avoidance and/or interference, but not necessarily to an effective attenuation of item memory traces (Norman et al., 2008; Norman, 2010). Previous ERP studies can be interpreted in the light of the present findings. In particular, Bergström et al. (2009b) showed that ERPs between 300 and 500 ms were more negative-going for no-think items in the direct suppression group, but not in participants that retrieved substitutive memories in order to avoid thinking of the target. Crucially, only the direct suppression group showed inhibitory forgetting on independent probe test. The authors interpreted the ERP difference for the direct suppression group as reflecting inhibitory control mechanisms (Bergström et al., 2009b). This is in line with previous studies, directly relating ERP negativities to inhibitory control (Mecklinger et al., 2009). However, it is as feasible to assume that negative ERPs can reflect a reduction of memory strength. Importantly, our findings indicate that ERP correlates of inhibitory control and memory strength may overlap, but differ with respect to sensitivity to repetition. Whereas frontal ERPs related to memory strength are reduced over the time-course of the experiment, centroparietal correlates of inhibitory control are unaffected by repetition and appear to be under strategic control (Bergström et al., 2009a; Mecklinger et al., 2009).

In sum, our data provide a cornerstone for the assumption that intentional suppression in the think/no-think paradigm involves the inhibition of unwanted memories. We show that the strength of memory traces is reduced on measures that are robust against interference as an alternative explanation for forgetting effects. Our results help to dissociate between different cognitive processes related to the suppression of unwanted memories. The activation of executive control networks contributes to the memory impairment of unwanted memories. Forgetting then can result from an inhibition of memory traces at an item level, as reflecting in recognition memory impairment and decreased physiological signals of memory strength.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.