Aging is accompanied by changes in several memory domains (Desgranges et al., 1994; Giffard et al., 2001). One of these domains is working memory. Many studies have reported lower working memory (WM) performance in older adults compared to their younger counterparts (Craik et al., 1990; Reuter-Lorenz & Park, 2010; Sylvain-Roy et al., 2015). Given the crucial role of WM in various complex cognitive activities, its decline in older individuals has been suggested to negatively impact both cognitive functioning and daily life (Vaughan & Giovanello, 2010). Many authors have proposed the use of cognitive training to enhance the WM of older adults, potentially leading to significant cognitive benefits. Most WM training programs adopt a process-based approach offering adaptive and repeated exercises aimed at improving the foundational processes underlying WM deficits (Boujut & Belleville 2019; Matysiak et al., 2019).

Several studies have indicated improvements resulting from WM training programs when assessing group performance in older adults (Boujut, Verty et al., 2020; Dahlin et al., 2008; Li et al., 2008; Matysiak et al., 2019; for meta-analysis see Karbach & Verhaeghen, 2014). However, substantial differences exist at the individual level (Borella et al., 2017; Matysiak et al., 2019). For example, Matysiak et al. (2019) reported that, after 25 sessions of adaptive WM training, 21.4% of older participants did not surpass their initial baseline performance level (Matysiak et al., 2019). Therefore, while WM training benefits many older adults, some individuals do not seem to exhibit improvement.

A few studies have attempted to understand whether these inter-individual differences in training efficacy are attributed to individual factors (for reviews see Teixeira-Santos et al., 2019; Ophey et al., 2020). Identifying a coherent set of underlying factors to explain inter-individual differences holds crucial clinical and theorical implications (Desgranges et al., 1998). On a clinical level, it could help maximize efficacy for responders while offering alternative training approaches for non-responders, facilitating more personalized and efficient interventions. From a theoretical perspective, cognitive training can serve as an interesting experimental model for investigating brain plasticity processes in older adults (Belleville et al, 2023). Identifying the brain characteristics and factors behind individual differences in training responses may shed light on the underlying brain mechanisms that help maintain cognition in late adulthood.

Notably, one of the key advancements in cognitive neuroscience research related to aging is the demonstration of substantial inter-individual variability in functional activation changes with age (Nyberg & Bäckman, 2011; Cabeza, 2002). These observed differences may serve as explanatory factors for individual variations in cognitive training outcomes (for review see Baykara et al., 2021). A crucial inter-individual difference lies in the concept of maintenance, which refers to the brain’s ability to retain its neural resources as individuals age (Nyberg & Bäckman, 2011). According to maintenance models, variations in age-related cognitive changes correspond to differences in brain alterations, with a more youth-like activation pattern associated with better cognitive performance (Fandakova et al., 2015; Verty et al.; Nagel et al., 2009, 2011; Nyberg & Bäckman, 2011). The notion of compensation can also contribute to a better understanding of interindividual differences in brain activation. Compensation refers to neuroplastic processes that come into play to adapt to changes in the brain, ultimately supporting cognitive performance (Reuter-Lorenz & Cappell, 2008; Cabeza, 2002). Accordingly, differences in activation patterns observed in some older adults when compared to younger ones might indicate underlying compensation. It is important to note that the concepts of maintenance and compensation are not necessarily antithetical. The CRUNCH model, for example, suggests that brain maintenance and compensation for brain decline are two possible paths to cognitive success in older adults (Reuter-Lorenz & Cappell, 2008). Hence, a youthlike activation might reflect either activation maintenance or lack of compensation. Given the potential implications of maintenance and compensation in explaining cognitive aging, a crucial question arises: Do inter-individual differences in youth-like functional activation explain inter-individual differences in WM training gains?

Two studies have offered supportive evidence suggesting that older adults with a youth-like activation pattern may be the most responsive to WM training. Heinzel et al. (2014) examined this question by examining activation network efficiency and capacity in 20 older adults through WM-load-dependent activation. Neural efficiency was defined as the ability to maintain lower activation levels during a low WM-load (1-back) condition, while neural capacity referred to the ability to increase activation in response to higher WM-load (e.g., 3-back) conditions compared to the low WM-load condition. Given that existing literature indicates that younger adults typically exhibit superior efficiency and capacity compared to older adults, this activation pattern was interpreted as reflecting a more youth-like activation pattern. The authors observed that both greater efficiency and greater capacity of the frontoparietal network were associated with higher behavioral gains in older adults after a 12-session WM training program (Heinzel et al., 2014). Vermeij et al. (2017) reported similar results in 30 older adults with and without mild cognitive impairment (MCI) using near-infrared spectroscopic imaging before a 25-session WM training program.

While these two studies suggest that maintaining a youth-like activation pattern in WM-related regions among older adults is associated to greater gains in WM training, several critical questions still require further investigation. When considering individual differences, a major challenge lies in identifying a single and meaningful metric to reflect the degree of youth-like activation at the individual level. For example, in the two previous studies, the authors did not compare a particular older participant to a group of younger adults to establish his/her youth-like pattern. Instead, they used scores reflecting how closely the activation difference between the high and low load WM conditions in a particular older participant resembled the pattern observed in younger adults (Heinzel et al., 2014; Nagel et al, 2011; Vermeij et al., 2017). Thus, the youth-like scores obtained by older adults expressed how similar their activation pattern was to the activation pattern observed or expected in younger adults, rather than being computed from an actual comparison with the scores of younger adults.

The Goodness of Fit method (GOF; Düzel et al., 2010), employed in this study, offers a promising approach to enhance the interpretation of inter-individual functional activation differences in the context of a youth-like activation pattern. This method assesses the degree of youth-like activation by comparing an individual older adult’s activation pattern and the WM neurofunctional network observed in a reference group of younger adults who underwent the same fMRI protocol under similar conditions. The GOF method considers differences in activation both within the regions typically activated by younger adults and in regions outside of these areas. This is advantageous compared to previous approaches, as compensation models suggest that older adults may compensate by activating brain regions not engaged by younger adults (Cabeza et al. 2018). Notably, the two earlier studies selected regions of interest (ROIs) based on those reported in the literature for comparable experiments, thus excluding activation outside of pre-determined WM-related areas, limiting their interpretation of the observed youth-like pattern.

Another critical consideration is the timeline of training gains, which typically follows a non-linear pattern characterized by an initial stage of rapid improvement followed by a slower learning stage (Belleville et al., 2022; Belleville et al., 2023; Belleville et al., 2018; McAvinue et al., 2013; Boujut, Verty, et al., 2020). Learning models propose that these early and late learning stages rely on different cognitive and brain mechanisms (Chein & Schneider, 2012) and may be influenced by distinct individual characteristics (Belleville et al, 2022). The two previously mentioned studies focused on whether a youth-like activation was associated with change from pre-training to post-training, involving a relatively large number of training sessions. It is plausible that different mechanisms explain inter-individual differences in WM training efficacy in the early rapid stage vs the later stage of training. Therefore, this study aimed to investigate which neurofunctional activation is associated with gains during either the early- or late-stage of training.

Lastly, an additional aim was to examine whether neurofunctional characteristics of older participants could be associated with inter-individual differences in transfer outcomes. This could offer valuable insights, as transfer to untrained tasks is often not observed following WM training (Dahlin et al., 2008; Ji et al., 2016; Lawlor-Savage & Goghari, 2016). Similar to training gains, the extent of transfer gains might depend on inter-individual variability in baseline activation.

The primary objective of this study was to use the GOF metric, which quantifies youth-like activation at the individual level, to investigate whether youth-like functional activation in older adults is associated with greater WM training gains. Additionally, we aimed to determine whether a similar association is found for the early- and late-stage of training. The GOF value was calculated during the performance of the N-back task under low and high task loads. This was done to allow interpretation in terms of efficiency and capacity. Efficiency is typically examined under lower task load conditions. This is because lower levels of activation to complete the task under low load conditions is interpreted as reflecting better efficiency. Conversely, capacity is typically examined under high task load conditions, as higher recruitment under demanding conditions is typically interpreted as reflecting superior capacity (Barulli & Stern, 2013). Finally, a secondary objective was to examine whether youth-like activation is associated with greater transfer gains and, once again, whether this applies to the early- and late-stage of training.

Based on previous findings, we hypothesized that a larger GOF, reflecting a more youth-like brain activation, would be associated with greater training gains. While the specific effects pertaining to the early or late training stages are unknown, we expected a stronger association between brain activation and training gains in the early-stage given that training gains have been found to be more pronounced in the early-stage and dependant on individual characteristics. In younger adults, lower activation during lower WM task load reflects efficiency, whereas higher activation during high WM task load, reflects capacity. Consequently, a positive correlation between training gains and youth-like neurofunctional activation patterns at low WM task load in older adults was hypothesized as a link between neural efficiency and training gains. In contrast, a positive association between training gains and youth-like neurofunctional activation patterns at higher WM task loads was hypothesized as a link between greater neural capacity and training gains. Finally, we hypothesized that older adults with a youth-like activation pattern would exhibit greater transfer gains.