Introduction

Due to the brain’s potential for neuroplasticity, atypical functional organisation of language and memory networks can be observed in response to neurological insults such as epilepsy. An atypical organisation observed in adulthood can result from either abnormal development (eg, atypical network development in the presence of pathology) or reorganisation (eg, atypical recovery from an insult to a previously established typical network)—two possibilities that we jointly refer to as (re)organisation. This putatively compensatory shift is thought to reflect the transfer from pathological to less-affected tissue and is thought to preserve cognition.1 2 In most studies, (re)organisation is studied at the level of a single cognitive modality. However, understanding how cognitive networks interact as they (re)organise is of theoretical and clinical interest, especially in conditions such as temporal lobe epilepsy (TLE), in which surgery can disrupt both language and memory function.

Neurosurgical planning synthesises many factors including clinical characteristics, preoperative neuropsychological status and neuroimaging.3 In these considerations, hemisphere of language dominance is an important predictor of memory outcomes.4 Modern proposals of a functional unification of language and episodic memory,5–7 coupled with functional MRI studies in which language and verbal memory tend to colocalise in epilepsy,8 9 together suggest that (verbal) memory dominance can be inferred from language dominance8–13--though this is not beyond dispute.14–16 While memory and language are undeniably intimately connected, how they interact during (re)organisation in response to epilepsy, and what this means for memory outcomes following surgery is less clear. A better understanding of the frequency and patterns of joint (re)organisation on an individual level is particularly important in an era of expanding surgical options, and given high, but individually variable, rates of postsurgical cognitive decline.

Here, we gathered language and memory lateralisation data from three surgical centres to assess the interrelation of these two cognitive domains using a simulated lesion approach (ie, Wada testing). To our knowledge, this represents the largest multicentre study to directly examine patterns of joint (re)organisation of language and memory in epilepsy. We ask: How frequently do language and memory (re)organise together, and what are the dominant patterns of (re)organisation? What factors drive concordance versus discordance? And does concordance/discordance provide insight into baseline memory function and risk for postsurgical decline? Based on previous fMRI findings8–10 and modern theories of language/memory unification,5 17 we hypothesised that (1) language and memory would be strongly colateralised, even following a rightward shift, with the strongest colateralisation for verbal memory and (2) the degree of language and memory concordance would be associated with unique patterns of preoperative and postoperative memory function.

Materials and methodsParticipants

162 patients with epilepsy received technically valid bilateral Wada tests as part of a presurgical evaluation across three comprehensive epilepsy centres—University of California, San Diego (UCSD), University of California, Los Angeles (UCLA) and Yale University. Basic demographic and clinical variables were available for all sites; however, some clinical variables were limited only to UCSD and Yale cohorts (eg, mesial temporal sclerosis (MTS)). We excluded 29 patients from the final analysis due to technical complications (eg, somnolence during one or both injections, lack of EEG slowing and substantial cross-fill of the anaesthetic). At all centres, a Wada was attempted for any patient considered for resective or ablative medial temporal lobe surgery between 2006–2023 (UCSD), 1999–2013 (UCLA) and 2007–2012 (Yale). Data were acquired retrospectively via a review of medical records.

Procedure

Wada testing technical procedures were generally consistent across centres and are described in online supplemental material. Cognitive protocols varied across centres and are described in detail in online supplemental material.

Language was tested via a combination of naming, repetition, reading and comprehension of verbal commands.

Memory encoding was tested by presenting the following for each site—UCLA: three-dimensional (3D) objects named by the patient and examiner; UCSD: a combination of verbally presented words, verbally and visually presented sentences, verbal commands, 3D objects and non-famous faces; Yale: black-and-white line drawings of common objects (named by the patient and examiner) and abstract designs.

Memory recall: UCLA and Yale: Recall of items was tested using progressive cueing (ie, free recall followed by cued recall followed by recognition). Recognition testing included the target stimulus plus three (UCLA) or five (Yale) foils. A memory score for each hemisphere was calculated as a percentage based on a number of items correct out of the total number of items. UCSD: Recall was tested with yes/no recognition (with an equivalent number of foils as targets). A memory score for each hemisphere was calculated using the following formula: 1−[(total misses+total false positives)/total items].

Primary outcomesLanguage lateralisation

Language lateralisation (ie, left, right or bilateral hemispheric dominance) was a clinical categorical decision at each site based on speech arrest, paraphasic errors and comprehension.

Global memory lateralisation

Global memory was assessed both as (1) a clinical categorisation as left, right or bilateral at each site (see online supplemental material) and (2) a continuous asymmetry score derived by the authors by subtracting the right hemisphere memory score from the left (ie, subtracting the score obtained during the left injection from that during the right). Here, positive asymmetry scores indicate greater left hemisphere memory support and negative values represent greater right hemisphere support.

Verbal memory lateralisation

A subset of UCSD patients had available item-level raw data (n=34), which allowed us to perform an analysis of verbal memory alone by excluding objects and faces.

Neuropsychological data

Verbal memory was assessed with a measure of delayed prose recall (ie, stories) using raw scores from the logical memory II (ie, LM-II) subtest from the Wechsler Memory Scale-III. Visuospatial memory was assessed with the delayed recall raw score from the Brief Visuospatial Memory Test-Revised (BVMT-R). Preoperative LM-II was available for n=61 (UCSD: n=36, UCLA: n=18); BVMT-Delayed Recall was available for n=49 (UCSD: n=31, UCLA: n=18). Postoperative LM-II and BVMT-Delayed Recall were available for n=18 (UCSD). Postoperative scores were obtained approximately 1 year following anterior temporal lobectomy (n=13) or laser ablation of the hippocampus. Preoperative to postoperative memory change was calculated by subtracting preoperative from postoperative raw scores (ie, negative scores correspond to decline).

Statistical analysisColateralisation of language and memory

Categorical agreement between language and memory lateralisation was assessed with Cohen’s kappa coefficient (κ) which accounts for percent agreement by chance. The following interpretations were used based on the most recent guidelines for level of agreement18: κ<0.20=none; 0.21–0.39=minimal; 0.40–0.59=weak; 0.60–0.79=moderate; 0.80–0.90=strong. Fisher’s exact tests examined the relationship between language and memory lateralisation in a 3×3 contingency table (ie, based on bilateral/left/right categorical classification).

Whereas categorical analyses are based on common clinically used thresholds, continuous associations allow verification of categorical memory results obtained with site-specific cut-offs. Therefore, we performed univariate analyses of variance (ANOVAs) with continuous memory asymmetry as the dependent variable and language lateralisation as a fixed variable. The amount of variance in memory lateralisation explained by language lateralisation was examined in linear regressions.

Predictors of concordant versus discordant lateralisation

ANOVAs and Fisher’s exact tests were used to examine differences in clinical and demographic characteristics among concordant versus discordant language/memory groups.

Neuropsychological correlates of colateralisation

To examine the effect of concordance/discordance of language and memory on neuropsychological memory outcomes, we reclassified patients based on (1) Hemisphere of dominance (eg, left hemisphere language lateralisation paired with left hemisphere epilepsy=dominant; left hemisphere language lateralisation paired with right hemisphere epilepsy=non-dominant) and (2) Concordance of language/memory colateralisation (eg, left hemisphere language lateralisation paired with left hemisphere memory lateralisation=concordant). Patients with bilateral language dominance or bilateral epilepsy were excluded from this analysis. We conducted 2 (Dominance: language-dominant vs language-non-dominant)×2 (Concordance: concordant vs discordant) models with memory score (ie, preoperative or preoperative to postoperative change) as dependent variables, with appropriate covariates (eg, age for preoperative models and preoperative memory score for preoperative to postoperative change models).

Results

Table 1 presents clinical and demographic characteristics of the overall sample. Online supplemental table S1 shows these data by site.

Table 1

Clinical and demographic characteristics of the overall sample (N=162)

Overall lateralisation rates

As expected, most patients showed left-hemisphere language lateralisation (79%), followed by right (12%) and bilateral lateralisation (9%). Memory was left-lateralised in 41% of patients, bilateral in 39% and right in 20%. Importantly, there were no significant differences in lateralisation rates across sites (3×3 Fisher’s exact test ps>0.05; online supplemental table S1).

Colateralisation between language and global memoryCategorical analysis

There was a significant association between language and global memory lateralisation (3×3 Fisher’s exact=70.0; p<0.001) but with minimal overall agreement between domains (κ=0.28; SE=0.05). Follow-up 2×2 comparisons were significant (ps<0.05) and showed the following agreement: (1) bilateral and right groups: κ=0.86; strong, (2) left and right groups: κ=0.63; moderate and (3) bilateral and left groups (κ=0.11; no agreement) (figure 1A and online supplemental table S2). Strict concordance was observed in 56% of cases (eg, left language/left memory; n=90), mixed discordance in 35% (eg, left/bilateral; n=56), and strong discordance in 9% (eg, left/right; n=14). Strong concordance was observed in individuals with right-lateralised language, of whom 95% (n=18) also showed right-lateralised memory, with a similar pattern of bilateral language showing a higher proportion of bilateral memory (n=10; 67%). Examination by site did not reveal a significant change to the above pattern (online supplemental table S2). Examination by side of seizure onset showed minimal overall agreement for left hemisphere onset (κ=0.21) and weak agreement for right hemisphere onset (κ=0.43 (see online supplemental table S3). However, concordance was notably higher for left language/left memory lateralisation in right (70%) compared to left (35%) hemisphere seizure onsets.

Figure 1

Counts and percentages of left, bilateral and right memory lateralisation as a function of language lateralisation (depicted at the top). Counts are shown separately for (A) global memory and (B) verbal memory. Note that whereas A represents data from all three sites, B is a subset of data from UCSD only, given the availability of raw data coded by stimulus type in this cohort only. UCSD, University of California, San Diego.

Continuous analyses

We next carried out a sensitivity analysis with global memory asymmetry as the dependent variable (figure 2A). The main effect of language lateralisation was significant (F(2,157)=23.2; p<0.001; ηp2=0.23). Follow-up comparisons (Bonferroni adjusted) showed that patients with right-lateralised language had greater right hemisphere memory asymmetry than patients with bilateral (p=0.03) and left-lateralised (p<0.001) language, and those with left-lateralised language trended towards greater left hemisphere asymmetry than those with bilateral language (p=0.08). When adding side of seizure onset (ie, left vs right) to the model, the main effect of side of onset was significant (p=0.003), such that left hemisphere seizures were associated with greater right-lateralised memory asymmetry. However, side of seizure onset did not moderate the association between language and memory colateralisation (p=0.64). In addition, in a linear regression with memory asymmetry as the dependent variable and language lateralisation simplified to typical (ie, left) vs atypical (ie, bilateral/right), language lateralisation explained 20% (R2) of the variance in memory asymmetry (p<0.001). The variance explained was higher for right than left seizure onsets (R2=26% vs 19%, respectively, ps<0.001).

Figure 2

Language lateralisation as a function of global memory laterality (ie, asymmetry). (A) Box plots with individual data points for memory asymmetry (ie, laterality index) as a function of language lateralisation group separately shown for left and right-hemisphere seizure onsets. The horizontal black line denotes median values, and the box displays the IQR (ie, 25%–75% of values). (B) Mean effect size (Cohen’s d) based on one-sample t-tests comparing global memory asymmetry against 0. *p<0.05; ***p<0.001.

Finally, we examined effect sizes of global memory asymmetry (comparing it to 0; ie, completely bilateral) for left and right seizure onsets, separately (figure 2B). Patients with right seizure onset and left-lateralised language showed a more leftward memory asymmetry (p<0.001; Cohen’s d=1.2) and patients with left seizure onset and right-lateralised language had a more rightward memory asymmetry (p<0.001; d=−1.7). The effect size for left seizure onsets with left-lateralised language was small in comparison (p=0.02; d=0.28).

Colateralisation between language and verbal memory

We next evaluated language and verbal memory colateralisation in a subset of UCSD patients (figure 1B).

Categorical analysis

The pattern of verbal memory and language colateralisation was similar, such there was a significant association between the two domains (Fisher’s exact=25.3; p<0.001), but with weak individual agreement (κ=0.44; SE=0.13). Strict concordance was observed in 68% of cases. Discordance was again driven by patients with left-lateralised language and variable verbal memory (68% left; 32% bilateral). In contrast, all seven patients with right-lateralised language showed right-lateralised verbal memory.

Continuous analysis

Language lateralisation explained 35% of the variance in verbal memory asymmetry (p<0.001) and was slightly higher for right (40%; p=0.004) than left seizure onsets (33%; p=0.01).

Comparison with global memory

In a follow-up analysis, we investigated whether colateralisation rates were higher for verbal memory compared with global memory. For this analysis, UCLA and Yale cohorts represented global memory, whereas verbal memory was limited to UCSD. We restricted the analysis to left-hemisphere language-dominant patients only, as this group showed the highest discordance rate. A 2 (memory type; global vs verbal)×3 (memory lateralisation: left/right/bilateral); Fisher’s exact test revealed a marginally significant association (p=0.07). As expected, a trend demonstrated greater left-lateralisation for verbal (68%) than for global (43%) memory.

Predictors of concordant versus discordant lateralisation

Four groups with at least 10 cases emerged: left language/left memory, left language/bilateral memory, left language/right memory and right language/right memory. We explored whether these groups differed on any variables (table 2). The effect of handedness was significant, such that the right language/right memory group had a higher proportion of left-handedness (56%) compared with both the left language/left memory (10%; p<0.001) and left language/bilateral memory (8%; p<0.001) groups and was marginally higher than the left language/right memory group (17%; p=0.06). A significant effect of hemisphere of seizure onset suggested that the left language/bilateral memory and left language/right memory groups had a significantly higher proportion of patients with left seizure onsets (77% and 83%) compared with the left language/left memory group (43%). Online supplemental table S3 presents individual demographic and clinical characteristics for the 19 patients with right-lateralised language, of whom 18 showed right-lateralised memory.

Table 2

Clinical and demographic characteristics of concordant versus discordant language-memory groups

Left TLE with left language lateralisation

Given the wide variability of memory asymmetry observed in our largest group of patients with left-lateralised language and left hemisphere seizures, we examined this group more closely. Hippocampal pathology is a known predictor of memory reorganisation in left TLE. Therefore, we tested whether the presence of MTS predicts left concordance (ie, left language/left memory) versus left discordance (ie, left language/bilateral or right memory combined) in a subset of n=46 individuals with a confirmed temporal lobe focus and known MTS status. As predicted, we found a significant effect of MTS (p=0.027), such that the left discordant group had a higher proportion of MTS positive cases (73%) than the left concordant group (37%; figure 3).

Figure 3

Examination by MTS status in temporal lobe epilepsy. Counts and percentages of MTS positive versus MTS negative patients within left discordant (ie, left language paired with bilateral or right memory lateralisation) and left concordant (ie, left language paired with left memory lateralisation) groups. MTS, mesial temporal sclerosis.

Neuropsychological correlates of colateralisationPreoperative memory

For LM-II, a 2 (dominance)×2 (concordance) analysis of covariance (ANCOVA), controlling for age, revealed a significant main effect of Dominance (F(1,55)=4.4; p=0.04; ηp2=0.08), such that those with seizures in the language-dominant hemisphere had lower LM-II scores (figure 4A) and no significant interaction. No significant effects emerged for BVMT-Delayed Recall (ps>0.05) (figure 4B).

Figure 4

Associations with memory function. Plots of means and SEs for 2×2 models of hemispheric language dominance and concordance between language/memory for preoperative memory: (A) LM-II; (B) BVMT delayed recall, as well as preoperative to postoperative memory change: (C) LM-II; (D) BVMT delayed recall. BVMT, Brief Visuospatial Memory Test; LM, logical memory.

Preoperative to postoperative memory change

A 2 (dominance)×2 (concordance) ANCOVA with preoperative to postoperative change in LM-II, controlling for preoperative memory score, revealed a significant interaction (F(1,18)=5.5; p=0.04; η p2=0.30), suggesting LM-II decline in the dominant concordant group only (figure 4C). Post hoc comparisons (Bonferroni adjusted) demonstrated that whereas the dominant concordant group experienced significantly greater decline than the non-dominant concordant group (p=0.02), the dominant discordant group did not differ from either of the non-dominant groups (ps>0.05). This was accompanied by a significant main effect of dominance, such that language-dominant patients overall demonstrated significantly greater LM-II decline (F(1,18)=8.3; p=0.01; ηp2=0.39). For BVMT-Delayed Recall, though the pattern looked similar, the main effect of dominance (F(1,16)=3.4 p=0.09; ηp2=0.24), and the interaction (F(1,16)=1.8; p=0.20; ηp2=0.14) did not reach significance (figure 4D).

Sensitivity analysis: association between global memory laterality and postoperative memory outcomes

Whereas the accuracy of language lateralisation assessed with the Wada procedure is less disputed, Wada memory mapping has been criticised for not consistently predicting postoperative memory outcomes. Therefore, to test the validity of our approach for memory lateralisation, we examined whether memory laterality (ie, global memory asymmetry score recoded as ipsilateral minus contralateral to the hemisphere of surgery) was associated with postoperative memory decline, controlling for preoperative memory score (figure 5). Preoperative to postoperative change in LM-II showed a significant association with global memory laterality (partial r(18)=−0.77; p<0.001), such that global memory lateralised towards the hemisphere of surgery was associated with greater verbal memory decline even after taking preoperative LM-II scores into account. The same relationship was significant for BVMT-Delayed Recall (partial r(16)=−0.66; p=0.008), controlling for preoperative memory score.

Figure 5

Associations between Wada memory laterality and postoperative memory outcomes. (A, B) Plot correlations between global memory asymmetry based on the Wada (ie, ipsilateral minus contralateral discrimination scores based on the hemisphere of seizures) and preoperative to postoperative LM-II and BVMT delayed recall change, respectively. Individual data points are colour-coded based on the dominance and concordance groups in figure 4. BVMT, Brief Visuospatial Memory Test; LM, logical memory. ***p<.001; **p<.01

Discussion

The current study represents the first multicentre investigation of joint language and memory (re)organisation patterns in a large and well-characterised sample of epilepsy patients undergoing neurosurgical consideration. We observed (1) modest overall concordance between language and both global memory colateralisation (56%) and, surprisingly, verbal memory colateralisation (68%); (2) However, concordance between language and memory (global and verbal) was high when language was right-lateralised (95%–100%); (3) Discordance was most notable in the presence of left hemisphere language dominance, left hemisphere seizures, and MTS; (4) Whereas language and memory discordance was not associated with impaired pre-surgical memory function, (5) discordance mitigated risk for postoperative memory decline following language-dominant neurosurgery.

Full (re)organisation: concordant rightward language and memory networks

Our most robust finding was that when language lateralised to the right hemisphere, both global and verbal memory nearly always colateralised. Similarly, when language was bilaterally represented, global memory also tended towards bilaterality. This suggests that with disruption severe enough to result in either a complete or partial interhemispheric shift of language, a parallel shift may occur for memory functions. This is predicted by theories proposing a strong interplay between memory and language networks5 7 17 19 20 and thought to be a compensatory, or perhaps necessary, strategy to maintain cognitive function. Supporting evidence for this pattern of concordant reorganisation comes from postsurgical cohorts, where atypical language dominance was found to be protective for postoperative verbal memory outcomes.21 Here, we support this assumption directly with a deactivation approach (ie, lesion simulation via the Wada test) in a large, multisite sample.

This joint rewiring of the frontal and medial temporal networks subserving memory and language may be necessary in the context of disrupted interhemispheric communication. Medial temporal memory and neocortical language networks dynamically interact and overlap in healthy adults22 and in epilepsy.23 24 In typical brain organisation, these overlapping networks predominantly reside in the left hemisphere and include the inferior frontal gyrus, angular gyrus, hippocampus, and medial and ventral white matter tracts.5 Our results suggest that in the presence of neurological disruption severe enough to drive interhemispheric shift in language, the right hemisphere appears capable of taking over both networks to maintain these interconnected processes. As this study focused on adults, the timeline of the observed joint (re)organisation is unknown. Specifically, it is difficult to determine whether (re)organisation of memory is facilitated by pre-existing right hemisphere language lateralisation (ie, whether it ‘follows’ language) or whether language and memory networks truly develop or shift in parallel.16 We note, however, that age of seizure onset did not differ between our groups (table 2), which suggests that the relationship between development and reorganisation is complex.

Demographically, a higher proportion of these right-lateralised patients were left-handed, consistent with well-documented associations. In terms of clinical factors, patients with the strongest right-lateralised memory asymmetry had left hemisphere seizure onsets (see figure 2), as would be expected if pathology in the language-dominant hemisphere leads to joint interhemispheric reorganisation. But surprisingly, roughly half (9/19) of right/right patients had a right or bilateral seizure onset, though right-lateralised language has been previously reported in right hemisphere pathology.25 26 Notably, five of these nine individuals with right-hemisphere onset were left-handed, consistent with the previously noted strong association between left-handedness and atypical language dominance. These observations suggest that joint (re)organisation is possible even in the presence of right hemisphere seizures. We suggest that joint reorganisation is strongly predicted when language is right dominant, and to a lesser extent, bilaterally dominant and that this pattern of joint (re)organisation may be the brain’s response to form new, compensatory networks linking these two processes.

When language remains left-lateralised: discordance in language and memory

In contrast, when language remained left-lateralised (ie, more ‘typical’), we observed high variability in memory dominance, especially in the case of left hemisphere seizures. In fact, almost 10% of these patients showed fully discordant (ie, right-lateralised) global memory. This observation is in line with a recent fMRI study that reported only moderate correlations (highest r=0.44) between language and memory domains.12 Although language and memory are indisputably intertwined, it is possible that milder forms of pathology do not result in a shift of language but result in memory relying on a paritally independent system. Importantly, we observed weak concordance even with verbal memory in the case of typical language dominance, with approximately one-third of these patients displaying discordant lateralisation. This finding contradicts our hypothesis of strong concordance, which was grounded in the widely held material-specific theory. This theory posits that the type of memory supported by the left and right medial temporal lobes is determined by the verbal versus visual nature of the material respectively.27 In addition, our original hypothesis of strong concordance was based on theories that the left hippocampus plays an important role in establishing language dominance3 23 28 and language reorganisation23 28 29 and may support some language functions10 11 30 31 (eg, retrieval of names from the lexicon). Though our results do not support a strong version of material specificity, a more circumspect version of the theory receives support from our observation that concordance was numerically higher between language and verbal memory than between language and global memory. Nonetheless, the observed concordance was still weaker than predicted by theories proposing a strong link between language and verbal memory dominance.

Our finding of low concordance between language and verbal memory lateralisation for left-language-dominant patients does not support prior fMRI research proposing that verbal memory can be strongly inferred from language lateralisation.8–10 However, our findings do align with some observations from these studies.8–10 12 13 For example, in Binder et al, patients with right language dominance did not show postsurgical declines in verbal memory following left-sided surgery, whereas outcomes for patients with left language dominance were much more variable.8 This is aligned with our own findings of high concordance between rightward language and rightward verbal memory, but lower concordance between leftward language and memory lateralisation. One potential explanation for lower concordance in patients with leftward language is pathology of the ipsilateral hippocampus, which was not taken into account in Binder et al. Indeed, in our cohort patients with left language and left MTS were more likely to have discordant (ie, contralateral) hemisphere memory support. Notably, however, not all patients with MTS showed evidence of memory (re)organisation, and similarly, a decent percentage of patients without MTS showed discordant lateralisation (figure 3). This underscores the idea that in the presence of epilepsy and typical language dominance, the relationship between language and memory networks is ambiguous in the absence of direct measurement of memory laterality.

Implications for cognition and neuropsychology in epilepsy

Established theories predict that verbal memory must be organised in an overlapping manner with language to function properly,32 33 and more modern theories predict a strong overlap for language and episodic memory more broadly.5 6 These theories would predict that a failure of language and memory to jointly (re)organise may contribute to language and memory impairments, which are commonly seen in TLE. However, we found that concordance/discordance of language and (global) memory had no measurable impact on preoperative memory ability. Though null effects are difficult to interpret, these data speak against theories of the necessity of language and memory overlap for function, at least in an epilepsy population.

However, discordance of language and (global) memory did emerge as an important predictor of postoperative memory decline, controlling for preoperative memory score. That is, although as expected, language-dominant surgery was associated with overall worse memory outcomes, outcomes following language-dominant surgeries in patients with discordant memory lateralisation were similar to those following non-dominant surgeries. This unique contribution of discordance even when taking preoperative function into account is notable as preoperative memory score is a well-established predictor of postoperative memory decline.4 This reinforces our findings that memory can be successfully (re)organised in the absence of language (re)organisation. This suggests that beyond language dominance and preoperative memory score, memory dominance benefits from independent measurement in presurgical planning, especially for patients with left language dominance. In contrast, if right-lateralised (ie, atypical) language is observed, then memory appears to also be on the right. This could perhaps provide the clinician with a heuristic to determine whether additional memory lateralisation is necessary.

Our finding that language/memory concordance was overall weak, coupled with the finding that discordance away from language-dominant surgery is associated with greater memory preservation, motivates the need for continued development of preoperative memory mapping procedures. Notably, the Wada procedure is much less frequently used in the modern surgical workup given its invasive nature and inconsistent predictive utility,34 with non-invasive (eg, fMRI) and newer invasive (eg, electrical stimulation during memory mapping35 36) methodologies increasing in use. However, mapping memory is quite a challenging process and requires more extensive clinical validation across both invasive and non-invasive measures.37 38 Modern surgical decision-making uses multiple sources of information to assess risk for postoperative decline,39 40 with classic risk factors including higher baseline memory score, surgery in the language-dominant hemisphere, and integrity of the hippocampus and/or medial temporal structures. These factors have been combined in useful risk-stratification-type calculators (ie, nomograms4 41) and other multivariate approaches that allow for examination of a combination of ‘red flags’ for memory decline.3 Though it will be important to investigate the clinical utility of our memory-laterality findings in a larger sample by pitting them against known clinical variables in multivariate models, our preliminary findings suggest that such risk calculators may benefit from the addition of direct measures of memory laterality and/or the concordance of language and memory colateralisation.

Limitations

Several limitations and opportunities for further research merit acknowledgement. Although patients completed Wada testing as part of standard clinical care, this was a retrospective study, so the possibility exists for sample bias (eg, higher rates of atypical language). Our subanalyses of material-specific (i.e., verbal) memory data, known MTS status and other clinical variables, and neuropsychological performance were modest as they were not available across all three sites; therefore, findings require replication. Though prose memory correlates with hippocampal volume in TLE,42 it also relies on lateral frontotemporal cortex more than list-learning or associative memory. However, we included prose recall as this was the most available measure of verbal memory across sites. In addition, caution is warranted when applying our findings to patients from under-researched groups with higher chance of atypicality in their language organisation such as bilingual individuals.43–46

These findings can be expanded upon by additional invasive and non-invasive methodologies. Here, Wada memory laterality was associated with verbal and visual memory post-operative outcomes (figure 5), although our sub-sample of individuals with postoperative memory data was modest. We also acknowledge that the Wada is a procedure with well-recognised limitations47 and has shown an inconsistent relationship with postsurgical memory prediction, with conflicting studies ranging from either no relationship48 to a modest9 49 or strong relationship.50 Several possibilities for these mixed findings include (1) a lack of sufficient material-specificity, (2) dissimilarity between Wada protocols and neuropsychological tests (eg, the use of recall vs recognition) and (3) the possibility that internal carotid artery Wada (ie, the one performed in this study) may incompletely perfuse critical regions such as the hippocampus. Extension with fMRI, stimulation or the posterior cerebral artery Wada will be a critical step in advancing measurements of memory laterality.

Conclusions

In the present study, we found minimal-to-weak overall concordance between language and memory (re)organisation on an individual level, across both global and verbal memory. However, when language (re)organised rightward, memory almost always colateralised in tandem, likely as an adaptive response to atypical language development. Taken together, whereas language (re)organisation appears linked to memory (re)organisation, memory appears to be able to (re)organise independently from language when language remains ‘typical’. Handedness, hemisphere of seizure onset and MTS were associated with discordance/concordance. A fuller appreciation of language and memory concordance/discordance would include a more mechanistic understanding of the drivers of (re)organisation. This would integrate a wider view of (re)organisation such as examination of language and memory ‘hubs’,7 or atypical language dominance being associated with global shifts in cortical organisation (ie, gradient asymmetries).51 Longitudinal studies assessing differences between development of organisation versus reorganisation would be helpful to clarify the mechanisms underlying the observed shifts. A further refinement would be studying more nuanced patterns beyond crude lateralisation, such as intrahemispheric or crossed dominance (eg, reorganisation of frontal but not temporal networks)2 25 as neither language nor memory are ‘all-or-nothing’ constructs.