Mnemonic discrimination of object and context is differentially associated with mental health

Episodic memories are formed by hippocampal binding of the "what" and "where" features of everyday events. The hippocampus minimizes interference between related similar episodic memories by pattern separation. Stress and psychopathology are associated with lowered pattern separation. While current behavioral paradigms typically use correct rejections of single object or context lures rather than composite stimuli, it is not known if object and context pattern separation differentially associate with mental health. We reasoned that an object-in-context paradigm would be more sensitive to mental health state than current implementations, given increased task demands. We found that non-clinical depression and anxiety symptom severity were associated with reduced lure rejection for both object and context and that only the object domain was associated with a concomitant increase in lure overgeneralization. Therefore, we argue that reduced lure rejection and increased overgeneralization must not be conflated. Although our object-in-context paradigm was not more sensitive to variation in mental health, we show that lure rejection and overgeneralization rate in one domain (e.g. object) was affected by the status of the other domain (e.g. context target versus lure). Finally, as several metrics of pattern separation exist in the literature, we evaluated the association of different metrics with mental health.


Introduction
Episodic memory allows us to mentally travel in time to relive specific, personally experienced events. Such memories are formed by hippocampal binding of the "what" and "where" features of a given event (Davachi, 2006;Eichenbaum, Yonelinas, & Ranganath, 2007). A defining characteristic of episodic memory is its ability to establish distinct memories of events sharing common features (e.g. the many birthdays celebrated in one's family home), allowing for recall of a particular memory with minimal interference from related similar events. Deficits in this minimum-interference storage have been linked to common psychiatric disorders (Kheirbek, Klemenhagen, Sahay, & Hen, 2012). Yet, standard paradigms usually only employ one of the above episodic features to study it. As episodic memories are better understood as composites, we herein present a new paradigm that combines both features to potentially serve as a more sensitive task.
In human studies of pattern separation, the "Mnemonic Similarity Task" (MST) is the most commonly employed task. It is an incidental encoding paradigm wherein subjects rate pictures of everyday objects along a given dimension (e.g. outdoor vs. indoor) before being presented with identical targets, similar lures (i.e. visually similar variants of the original target; e.g. a blue toothbrush instead of a red), and novel foils, which are to be identified as "old", "similar", and "new", respectively (Kirwan & Stark, 2007; see Stark, Kirwan, & Stark, 2019 for a review). Correct rejection of a lure as "similar" (mnemonic discrimination) is thought to represent successful pattern separation and correlates with DG/CA3 activity (Bakker, Kirwan, Miller, & Stark, 2008;Kirwan & Stark, 2007) and volume (Doxey & Kirwan, 2015). Variations on the MST have been implemented to investigate the effects of age (Toner et al., 2009), cognitive impairment (Stark, Yassa, & Stark, 2010), and depression (Déry et al., 2013(Déry et al., , 2015Shelton & Kirwan, 2013) on pattern separation. Recent installments have also included spatial information in the form of on-screen location of objects (Reagh & Yassa, 2014) or in the use of spatial contexts as stimuli .
Although episodic memory formation is widely understood as the binding of "what" (objects) and "where" (contexts) (Davachi, 2006;Eichenbaum et al., 2007), with few exceptions (e.g. Libby, Reagh, Bouffard, Ragland, & Ranganath, 2018;Reagh & Yassa, 2014), pattern separation paradigms only employ one stimulus type at a time . Moreover, while object and spatial information are processed along anatomically segregated streams in cortical and parahippocampal structures, both streams converge on the DG (Reagh & Yassa, 2014). Yet, it is not known if pattern separation demand in one domain could affect computational processing in the other as a result and thus bias behavioral output. Furthermore, because perturbation of neurogenesis and DG plasticity have been demonstrated to cause overgeneralization of conditioned fear (McHugh et al., 2007;Tronel et al., 2012), some speculate that dysfunctional pattern separation might underlie the intrusive memory symptomatology seen in anxiety disorders, such as post-traumatic stress disorder (PTSD) (Kheirbek et al., 2012;Lissek et al., 2014). Generally, it is reasoned that stress, through its detrimental effects on neurogenesis, disrupts pattern separation of events sharing similar features (e.g. the sound of New Year's Eve fireworks and the sound of gunfire), such that inappropriate pattern completion results instead (e.g. intrusive recall of gun-related trauma). Yet, it is unknown whether mental health state differentially affects behavioral pattern separation of objects and contexts. For instance, dysfunctional context pattern separation could potentially explain the observed deficit in contextual modulation of memories that is associated with overgeneralization in PTSD (Garfinkel et al., 2014). Therefore, we sought to investigate relationships between mental health and mnemonic discrimination of both object and context (putative pattern separation, but see Santoro, 2013). We reasoned that a combined object-in-context task might be more sensitive to this end than previous implementations, while also allowing for a more faithful representation of the composite nature of episodic memory.
In the literature (e.g. Stark, Yassa, Lacy, & Stark, 2013;Déry et al., 2013;, mnemonic discrimination is most commonly quantified and reported as the "Lure Discrimination Index" (LDI): p("Similar" | Lure) − p("Similar" | Foil). Analogous to signal detection theory of recognition memory (Snodgrass & Corwin, 1988), LDI seeks to correct for response bias to yield a more valid metric of mnemonic discrimination. However, several variations on LDI have been reported (e.g. Cunningham, Leal, Yassa, & Payne, 2018;, and some disagreement exists with respect to its most valid operationalization (Loiotile & Courtney, 2015). Importantly, it is not known if any particular instantiation serves as a better predictor of mental health. Therefore, we decided to compare the association of different versions of LDI with measures of depression and anxiety.
In summary, herein, we investigate whether measures of mental health differentially associate with (a) object and context lure discrimination and (b) different metrics of mnemonic discrimination. We chose depression and trait anxiety symptom severity as proxies for mental health, since these have been successfully employed by previous studies. Specifically, we hypothesized that symptom severity would correlate more strongly with a measure of combined object and context lure discrimination than with separate metrics that take only one domain into account.

Participants
As pattern separation tasks are generally sensitive to age, we only recruited participants within the youngest age span defined in Stark et al. (2013). We did not exclude participants based on any other criteria than age. Study 1 included 30 participants (mean age = 22; range = 19-31; 11 males), while Study 2 included 87 participants (mean age = 22; range = 18-38; 16 males). All participants were recruited from Lund University and given monetary compensation for their participation (approx. 20 USD). Informed consent was obtained from all before participation. The study was conducted in accordance with The Code of Ethics of the World Medical Association (Declaration of Helsinki) and was approved by the Regional Ethics Board at Lund University. Demographic data are summarized in Table 1.

Task design -MDOC
The "Mnemonic Discrimination of Object-in-Context" (MDOC) task was designed as an explicit memory test divided into blocks, each containing a study and test phase (Fig. 1). During study phase trials, participants were presented with visual stimuli consisting of a background image depicting a natural or man-made scene (the "context") filling the screen, and a smaller picture of an everyday object (the "object") superimposed on the scene. To encourage equal attention to both stimuli, participants were asked to judge (yes or no), using computer keys, whether the object fit the context semantically (e.g. a rubber duck in a bathroom; however, objects were randomly assigned to contexts independent of congruity). In the subsequent test phase, participants were presented with further object-in-context stimuli and informed that both object and context could be identical repeats (targets), new (foils), or similar but non-identical (lures) with respect to the last Abbreviations: BAI (Beck Anxiety Inventory); BDI-II (Beck Depression Inventory); IPAQ (International Physical Activity Questionnaire); STAI-T (State-Trait Anxiety Inventory -Trait version); SD (standard deviation). study phase and asked to identify these as "old", "new", and "similar", respectively. The test phase was self-paced, and participants gave separate answers for object and context (response order and keys counterbalanced across subjects). Based on piloting with 40 participants (for results, see Supplementary Tables S1,2), it was found that an exposure time of 4000 ms followed by a 500 ms ITI during study trials would avoid ceiling effects at test and allow performance above chance level. Visual stimuli were adapted from Brady, Konkle, Alvarez, and Oliva (2008), Brady, Konkle, Alvarez, and Oliva (2013), Konkle, Brady, Alvarez, and Oliva (2010a), and Konkle, Brady, Alvarez, and Oliva (2010b). MDOC was presented using E-Prime 2.0 Professional. In line with previous studies, we calculated "corrected recognition", P r , for target hits [p("Old" | Target)p("Old" | Foil)] and LDI for lure correct rejections [p("Similar" | Lure)p("Similar" | Foil)], henceforth called LDI similar . In addition, two alternative LDIs were also calculated (herein named according to their unique characteristics): p("Old" | Target)p("Old" | Lure) (Loiotile & Courtney, 2015), henceforth called LDI old ; p("Similar" or "New" | Lure)p("Similar" or "New" | Target) (Cunningham et al., 2018), henceforth called LDI new . Finally, we calculated an "Overgeneralization Index" (OI) for lure false alarms to quantify erroneous behavioral pattern completion: p("Old" | Lure)p ("Old" | Foil) . These scores were calculated both for object and context responses alone as well as for integrated object-incontext (OiC) responses. For the OiC scores, the relevant answer (e.g. "similar") had to be given to both object and context in response to a trial where both object and context had the same status (e.g. lure): for instance, p("Similar" | Object & Context Lure)p("Similar" | Object & Context Foil). Importantly, because of the three-choice response protocol, LDI old and LDI new are perfectly correlated for the separate object and context measures but not for OiC.

Study 1
In the first study, 189 study and test trials were divided equally into 7 blocks. The visual stimuli were divided into 9 conditions (all possible combinations of old, similar, and new objects and contexts), resulting in 21 trials per condition, and 3 trials per condition per block. Stimuli were assigned randomly to conditions across subjects.

Study 2
In order to increase the sensitivity of the experiment, the second study excluded four conditions, leaving five: old object and context; new object and context; similar object and context; old object and similar context; similar object and old context. We reasoned that the latter three would entail the greatest demand on pattern separation and thus be of the highest interest in investigating the effects of mental health on pattern separation. In this version, 200 study and test trials were divided equally into 8 blocks. The visual stimuli were divided into 5 conditions, resulting in 40 trials per condition, and 5 trials per condition per block. Stimuli were assigned randomly to conditions across subjects.

Mental health questionnaires
In order to investigate the association between mental health and mnemonic discrimination, both studies included depression and anxiety questionnaires for participants to fill out upon completing MDOC. Both studies included the Beck Depression Inventory (BDI). Study 1 included the State-Trait Anxiety Inventory -Trait (STAI-T), while Study 2 included the Beck Anxiety Inventory (BAI). Given the smaller sample size of Study 1, we opted against BAI because of its possible limitations in non-clinical samples (Creamer, Foran, & Bell, 1995). Finally, we wanted to see if we could establish an association between self-reported physical activity level and lure discrimination. Therefore, the International Physical Activity Questionnaire (IPAQ) was included in Study 1; however, it had to be excluded in Study 2 for purposes of time.

Statistical analyses
All statistical analyses were carried out using IBM SPSS statistics version 25. All data were tested for normality using the Shapiro-Wilk Fig. 1. The MDOC task. The experiment was divided into blocks containing sequential study and test phases. During study phases, participants viewed object-in-context stimuli and rated whether the object was appropriate in the background context (Yes/No). Pictures were displayed for 4000 ms with a 500 ms ITI. During test phases, a series of objectin-context stimuli were presented, where object and context could be identical repeats (targets), novel foils, or similar but non-identical lures. Participants were asked to identify these as "old", "new", and "similar", respectively, with separate answers for object and context. Trials were subject terminated.
test. Because we recruited healthy young adults, mental health questionnaire data were skewed toward lower values; therefore, we used Spearman's rank correlation to test associations between mental health and memory performance. Tests for differences between correlation coefficients were performed using Hotelling's t-test, as part of the "cocor" package (Diedenhofen & Musch, 2015). As we found high levels of inter-correlation among the different lure discrimination metrics (data not shown), we did not make any corrections for multiple comparisons in our analyses, since correlation among dependent variables introduces statistical dependency. Thus, all results were considered significant at the p < 0.05 level. This would further make our analyses and results more readily comparable to those in the literature.

Study 1
Mean response proportions for each stimulus and response type are summarized in Table 2; mnemonic index scores are summarized in Table 3.
In order to investigate the association between mental health and mnemonic discrimination, we correlated participants' LDI scores with BDI and STAI-T data. However, no significant correlations were found (Supplementary Figs. S1,2; Supplementary Table S3). The same results were obtained when using OI (OiC scores had to be left out due to a floor effect) and P r scores. Furthermore, no association was found between self-reported physical activity level (IPAQ) and mnemonic scores. The study was likely underpowered given the small sample size and few trials per condition.

Study 2
In order to increase statistical power in the second study, we reduced the number of conditions in MDOC from nine to five (see Section 2.2.2. Study 2), which increased the number of trials per condition to 40. Moreover, we nearly tripled participant numbers (N = 87) as compared to Study 1.
Mean response proportions for each stimulus and response type are summarized in Table 4; mnemonic index scores are found in Table 5. We once again investigated the association between mnemonic discrimination and mental health. There were significant negative correlations between object LDI similar and BDI (r s = −0.235; p = 0.028) as well as BAI (r s = −0.260; p = 0.016) ( Fig. 2; Table 6). Importantly, this was paralleled by a positive association between OI and BDI (r s = 0.278; p = 0.009) as well as BAI (r s = 0.249; p = 0.021). These results suggest that the decrease in LDI is specifically associated with an increase in overgeneralization. In contrast, while there was a correlation between context LDI similar and BAI (r s = −0.238; p = 0.027), we found no concomitant association with OI (Fig. 3). This implies that the decrease in context LDI is not paralleled by a concurrent increase in overgeneralization, as was found for objects. Interestingly, erroneous "new" responses were somewhat more common for context lures than for object lures (Table 4). Concerning combined OiC responses, we found significant correlations between LDI similar and BDI (r s = −0.221; p = 0.040) as well as BAI (r s = −0.256; p = 0.017). However, OiC overgeneralization was exceedingly rare, which resulted in an OI floor effect ( Fig. 3; Table 5). Interestingly, the alternative LDI metric (LDI old /LDI new ) produced no significant correlations with BDI or BAI data (Table 6). As predicted, there were no associations between the old-new discrimination measure P r and BDI or BAI (though context P r did approach significance with BAI, p = 0.053). Finally, there were no correlations between age and LDI or OI scores, suggesting that our selection criteria achieved the desired effect of controlling for age.
Next, we wanted to compare the significant correlations found above to ascertain whether any particular metric could serve as a better predictor of BDI and BAI. To this end, we evaluated the Spearman correlation coefficients with Hotelling's t-test (see Section 2.4). We had hypothesized that the combined object and context lure discrimination measure (OiC) would be more sensitive to mental health than the separate object and context LDI. However, there were no significant differences between object and OiC LDI similar correlations with BDI or BAI (max t (84) = −0.235, p > 0.05), suggesting that the supposed added complexity of combined object and context lure discrimination was not more sensitive to mental health than object lure discrimination considered on its own. Moreover, for BAI, there was no significant difference between object and context LDI similar (t (84) = −0.393, p > 0.05). The negative relationship between mental health scores and mnemonic discrimination (LDI similar ) was paralleled by a concomitant increase in overgeneralization (OI) for objects (Table 6). We thus compared object LDI similar and OI but did not find evidence of any difference between these correlations (max t (84) = −0.509, p > 0.05), suggesting neither served as a reliably better predictor than the other.
In summary, higher BDI and BAI scores were associated with decreasing LDI similar for both object and context in our non-clinical sample; yet, only the object domain produced a concomitant increase in lure overgeneralization. Moreover, while LDI similar yielded correlations with both BDI and BAI, the recently suggested LDI old /LDI new showed no such associations. Finally, the OiC metric was not found more sensitive to mental health than the separate object or context metrics.

Object lure response is sensitive to context statusand vice versa
The MDOC paradigm was designed according to the premise that a combined object and context lure discrimination metric would be more sensitive to mental health than a single-domain metric. However, as evidenced by Table 5, the OiC scores were generally as high or higher than the individual object and context scores. This would not be expected if object and context responses were independent processes. We, therefore, wondered whether the status of one domain (e.g. context target) could bias the response probability in the other domain (e.g. response to object lure).
To this end, the proportion of correct rejections of object lures [i.e. p ("Similar" | Lure)] was calculated for trials where context status was either target or lure. By pooling data from both studies and our pilot study data (same version of MDOC as in Study 1, see Supplementary Tables S1,2), we achieved N = 157 subjects. Interestingly, correct  Note. For object and context, LDI new and LDI old produce equivalent outputs.  rejections of object lures were higher (t (156) = −2.191, p = 0.030, twotailed) given context lures (targets: mean = 0.71; SEM = 0.01; lures: mean = 0.73; SEM = 0.01). We also calculated correct rejection scores in the same manner for context but given object target or lure. However, there was no significant effect of object status on context lure rejection (t (156) = −0.456, p > 0.05). Since object lure rejection was sensitive to context status, we wanted to know if the results extended to lure false alarms (i.e. overgeneralization). Using our pooled data, we calculated the proportion of "old" responses to object lures during trials where context status was either target or lure. Indeed, object lure false alarm rate was higher (t (156) = 5.100, p < 0.001, two-tailed) given context targets (targets: mean = 0.21; SEM = 0.011; lures: mean = 0.16; SEM = 0.007). We computed false alarm rates in the same manner for context but given object target or lure. Here too, we found that lure false alarm rate increased (t (156) = 2.723, p = 0.007, two-tailed) given object targets (targets: mean = 0.14; SEM = 0.008; lures: mean = 0.012; SEM = 0.007).
Finally, we were unable to establish any significant association between mental health measures and the above object-context interactions (r max = −0.060, p > 0.05).

Object hit rate is sensitive to context statusand vice versa
Given the above results, we investigated whether object target hit rate would also be sensitive to context status. To test this, we calculated the proportion of object target hits for trials where context status was either target or lure. Indeed, hit rate was higher (t (156) = 10.643, p < 0.001, two-tailed) given context targets (targets: mean = 0.86; SEM = 0.009; lures: mean = 0.75; SEM = 0.011). We performed analogous calculations for context but given object target or lure. Here too, hit rate was higher (t (156) = 5.966, p < 0.001, two-tailed) given object targets (targets: mean = 0.89; SEM = 0.009; lures: mean = 0.84; SEM = 0.010).
We note that these results, and those reported in Section 3.3, could potentially be explained by participants adopting a consistency bias across the object and context responses (e.g. an increased object lure rejection because of a tendency to use the same response twice). If participants were basing their second response on their first, the overall accuracy in the second domain should be compromised, as a perseverating response bias would, on average, be wrong more often. However, a paired t-test (N = 157) with estimated Bayes factor revealed no effect of response order on overall accuracy (p = 0.32, onetailed) and favored the null hypothesis (BF +0 = 0.13). Thus, it is unlikely that the reported results reduce to per trial response consistency.

Discussion
As episodic memories are formed by hippocampal binding of the "what" and "where" features of events (Davachi, 2006;Eichenbaum et al., 2007), we designed an object-in-context paradigm to study the impact of mental health on object and context mnemonic discrimination, reasoning that such an approach might be more sensitive than previous implementations. Moreover, because several metrics of mnemonic discrimination exist in the literature, we evaluated the associations of different variants of LDI with measures of mental health.
In accordance with earlier studies (Déry et al., 2013(Déry et al., , 2015Shelton & Kirwan, 2013), object LDI similar was negatively associated with BDI score. The association also held true for BAI, arguing for a more general association between mental health on mnemonic discrimination. This is in line with current models claiming that psychological stress, in general, has a detrimental effect on neurogenesis, which subsequently perturbs pattern separation (Dillon & Pizzagalli, 2018;Kheirbek et al., 2012). Notably, the functionally equivalent alternative LDI new /LDI old produced no significant correlations with BDI or BAI. LDI old was suggested as a theoretical improvement to LDI similar in the Old/New/Similar response protocol (Loiotile & Courtney, 2015), while its LDI new complement has been employed to study the effects of post-encoding Fig. 3. Relationship between Overgeneralization Index (OI) and depression (BDI) and trait anxiety (BAI) symptom severity. ρ refers to Spearman's rank correlation coefficient. No correlations were performed for Object-in-Context (OiC) due to floor effects. Note: Beck Anxiety Inventory (BAI); Beck Depression Inventory (BDI). stress on mnemonic discrimination (Cunningham et al., 2018). Interestingly, Cunningham et al. found an effect of stress but only for negative valence lures. Though redundant for the separate object and context measures, LDI new and LDI old do produce different results for the combined OiC metric. However, we still found no correlation with BDI or BAI for the pair. This suggests that the standard LDI similar fares better in non-clinical samples when studying mental health.
Our study extends previous results by showing an association between mental health and context mnemonic discrimination. A similar association with age was shown by Stark and Stark (2017), using separate tasks for object and context. When comparing the two, they noted that as well as explaining twice the variance in aging, only object LDI similar correlated with DG/CA3 volume. Although the studies differ somewhat in the operationalization of spatial context, both appear to argue against the possibility that context discrimination would contribute any unique predictive power. This may stem from the fact that the DG (McTighe, Mar, Romberg, Bussey, & Saksida, 2009) and neurogenesis (Clelland et al., 2009) are only necessary for pattern separation of small differences. Object lures are usually more visually similar than context lures, since spatial contexts are complex multi-dimensional stimuli that can vary in more respects. This difference could partly explain the increased proportion of "new" responses to context lures compared to object lures (Tables 2, 4 and Supplementary Table  S1); Stark and Stark (2017) noted the same trend in their study. A possible limitation concerns our definition of "context". Our context stimuli primarily serve as complex single-paired associates rather than functioning like the multi-paired family home alluded to in the introduction (see Stark, Reagh, Yassa, & Stark, 2018 for a discussion). In a recent study by Libby et al. (2018), the hippocampus was shown to generalize across similar events that shared both object and context information, while discriminating events that shared similar information in only one domain, in line with a hierarchical nature of hippocampal representations (McKenzie et al., 2014). Thus, future studies may want to try associating single contexts with many similar objects to study object and context pattern separation. For instance, this could be used to investigate contextualization deficits as an alternative explanation for clinical overgeneralization symptoms (e.g. Garfinkel et al., 2014). Finally, we show for the first time a dissociation between lure discrimination and overgeneralization. While the negative association between object LDI similar and BDI was paralleled by a concomitant increase in object OI, this was not the case for contexts. Therefore, we would argue that overgeneralization must not be conflated with an increase in lure misses. Thus, along with the LDI, future studies may want to consider reporting an overgeneralization metric, like that used by , particularly where clinical associations are considered.
The MDOC was designed according to the premise that combined object and context lure discrimination would be more demanding than previous implementations and thus more sensitive to mental health. However, we did not find evidence of this; rather, OiC LDI similar scores were as high or higher than the individual object and context scores. How might this be explained? While objects and contexts are processed along separate "what" and "where" streams in extrahippocampal structures, they subsequently converge on the DG, granting it domaingenerality (Reagh & Yassa, 2014). Thus, one might expect pattern separation to function with respect to the total study-test overlap. Indeed, this would explain the high OiC scores: all else being equal, the total overlap is lower given two lures than one. Yet, this domain-generality also raises the possibility of the behavioral output being correlated for the streams as a result of inter-dependent processing. Indeed, we found that context status biased the responses to object targets and lures and vice versa. Lure rejection, lure false alarms, and target hits in one domain were all sensitive to the status of the other, with the somewhat perplexing exception of context lure rejection. Nonetheless, these observations are in line with the notion of total study-test overlap dictating the response. Interestingly, the CA1 is thought to receive functionally segregated inputs from the lateral (object) and medial (context) entorhinal streams, along with the amalgamated CA3 signal (Igarashi, Ito, Moser, & Moser, 2014). This could potentially allow CA1 to identify the source of discrepancy between stored and incoming information (Duncan, Ketz, Inati, & Davachi, 2012) and differentially impose pattern separation or completion by feedforward neuromodulation (Hunsaker & Kesner, 2013). Future neuroimaging studies employing high field strengths could potentially elucidate the processing of object and context along the hippocampal transverse axis. Finally, we were unable to establish any effects of mental health on this objectcontext interaction. If such a relationship exists, our statistical power is likely too low to detect it. Future studies could clarify this matter using bigger samples.
Mnemonic discrimination is understood as a recall-to-reject process in which participants are able to reject lures on the basis of recollecting the original target (Kirwan & Stark, 2007). Our object-in-context paradigm allowed us to study how a parallel layer of information (e.g. the context) can influence the response to such a lure (e.g. the object). Indeed, when subjects were faced with an object lure against a context target, they were more likely to embrace it as "old". This observation may be in line with the findings of Kim and Yassa (2013). They showed that both lure discrimination and overgeneralization were more likely given correct source judgments (i.e. original on-screen location) and high confidence "remember" responses, indicative of recollection. Thus, it would appear that simultaneous recollection of associated, but ultimately non-diagnostic, information can increase the tendency of embracing ambiguous information as previously encountered. Stevenson, Reagh, Chun, Murray, and Yassa (2020) recently showed that source memory and mnemonic discrimination are dissociated in the hippocampal subfields and medial temporal lobe regions. It is tempting to speculate that such dissociated processes converge downstream to inform the final mnemonic decision. Taken together, the current results suggest that no task is process pure; rather, the mnemonic decision in a given domain (e.g. object) is potentially affected by information in another (e.g. context), as determined by total study-test overlap. Accordingly, the above findings of Kim and Yassa monotonically tracked lure similarity. Encoding variability between object and context could further contribute to a differential memory strength to compromise lure rejection (Huffman & Stark, 2017). As we did not collect confidence information (or remember/know), we cannot assess the contribution of general familiarity effects. It would be interesting to more exhaustively investigate the effect of global match on lure overgeneralization.
We were unable to find any association between mnemonic discrimination and self-reported physical activity level. This could simply be due to low power in Study 1; however, a previous study also failed to find this relationship based on self-report measures (Shelton & Kirwan, 2013). Moreover, Déry et al. (2013) only observed increased lure discrimination in subjects who achieved an above-median increase in VO 2 peak , suggesting that the relationship might be more complex.
Our paradigm has some noteworthy limitations. First, due to limited resources, we were unable to gather any subjective similarity ratings of our targets and lures during initial piloting. Thus, unlike the MST, our stimulus material has not been surveyed in an independent population to empirically evaluate its homogeneity. However, we note that our results closely recapitulate those of Stark and Stark (2017;cf. Table 3 in  Stark and Stark and Tables 2, 4 herein). Thus, rather than being an effect of a particular stimulus set, the observed response patterns between the object and context domains are likely to be the result of their intrinsic, relative complexities (see discussion above). Second, given the separate and serial nature of our response protocol, one could argue that it does not promoteor may even disincentivizemnemonic integration of object and context (i.e. the subject does not create a unified memory of the two but rather treats them as separate). This design was necessary to allow the computation of the different metrics separately for object and context. However, the finding that context status affected object response proportions should refute this argument. If subjects were processing the two as separate, it is difficult to see how the above results could emerge. Another possible limitation is that we employed same-day testing with a rather short study-test delay. As such, it is possible that the effects of depression and anxiety amplify with increasing delay (Déry, Goldstein, & Becker, 2015). Moreover, our encoding paradigm is explicit rather than incidental, and our response protocol is self-paced rather than timed. However, Stark, Stevenson, Wu, Rutledge, and Stark (2015) found no effects pertaining to these parameters (including lag) when studying aging-related deficits in pattern separation, suggesting that the results may emerge notwithstanding variations in task-design. While we were primarily interested in the effects of mental health on different metrics of mnemonic discrimination, we recognize that other parameters, such as confidence ratings and reaction times, could also be relevant. Due to data logging issues, reaction times were not available. Previous studies have found longer reaction times with increasing lure similarity (Fujii et al., 2014) and poorer mnemonic discrimination performance (Shelton & Kirwan, 2013), but no significant effects of mental health on lure reaction times have been reported, to the best of our knowledge. Finally, given the already large number of conditions in MDOC, we chose not to include a manipulation of semantic congruency between object and context. Instead, we effectively controlled for congruency by randomly allocating objects to contexts, such that it would highly unlikely for any archetypal pairings to occur (e.g. a toothbrush in a bathroom). As semantic congruency between an object and its associated context is known to impact encoding (van Kesteren, Ruiter, Fernández, & Henson, 2012), it would be interesting for future studies include this as a manipulation in a similar paradigm.
In conclusion, our results highlight the need for further elucidation of pattern separation and completion processes using complex stimuli that more faithfully represent episodic memories. Moreover, future studies of mental health and mnemonic discrimination may want to consider reporting overgeneralization metrics along with the LDI for purposes of greater clarity, especially when using new or complex stimulus sets.