Among the 30 participants initially recruited, 27 (8 females) aged from 30 to 85 (M = 56.22 years old, SD = 17.53, variation coefficient = 0.31) were finally retained, divided into two groups: the Younger Group (YG; n = 13, M = 40.07 years old, SD = 8.33, 5 females) and the Older Group (OG; n = 14, M = 71.21 years old, SD = 6.93, 3 females). All participants were right-handed (Edinburgh Handedness Inventory; Oldfield, 1971) and had normal or corrected-to-normal vision. Participants were cognitively unimpaired (Mini Mental State Examination, MMSE; Folstein et al., 1975), had no psychiatric symptoms (Hospital Anxiety and Depression, HAD; Zigmond and Snaith, 1983) or episodic memory deficit (“5 words” test, Dubois et al., 2002). All participants were highly educated (Poitrenaud questionnaire; Kalafat et al., 2003) and were native French speakers. They gave their written informed consent for the study, which was approved by the local ethics committee (CPP no. 2014-A00569-38). Demographic information and inclusion criteria are mentioned in Table 1.

Several neuropsychological tests were administered to evaluate cognitive domains, including language, memory, visual processing, and executive function, that are commonly used to measure cognitive function associated with naming (Duffau et al., 2014). Language was assessed using verbal fluency (generation of words that are related to the same semantic category; Cardebat et al., 1990), the verbal automatisms test (completion of overlearned French expressions by the participants; Beauregard, 1971), and Mill-Hill B vocabulary scale (explanation of words meaning and selection of the appropriate synonym for a word among a list; Deltour, 1993). Memory was assessed using forward and backward digit span tests (short-term and working memory; Wechsler, 1997) and the MacNair questionnaire to assess subjective memory complaints (McNair and Kahn, 1983). Visual processing was evaluated using the version A of the Trail Making Test (TMT-A; Tombaugh, 2004). Finally, executive functioning was assessed using Version B of the TMT (TMT-B; Tombaugh, 2004). We computed a difference score between the TMT-B and the TMT-A, referred as the “executive score”, as mentioned in the next sections of the manuscript (Corrigan and Hinkeldey, 1987). This executive score reflected cognitive flexibility, and removed simple sequencing, visual scanning and psychomotor functioning effects.

The participants performed an object naming task with 80 black and white drawings of objects and animals (DO-80 test; Metz-Lutz et al., 1991) using a blocked design alternating four task and control periods. Twenty images lasting 2 s each with an inter-stimulus interval of 500 ms were presented during each task period, and participants were instructed to name the images as accurately and rapidly as possible. During each control period, 20 simple images (circles and squares) were presented and participants were instructed to respond either “square” or “round.” Stimuli were displayed using E-prime (E-prime Psychology Software Tools Inc., Pittsburgh, PA, USA) and were projected onto a screen behind the magnet using a video projector. Oral responses were recorded via an MRI-compatible microphone (FORMI™ II, version 1.2) attached to the coil. We used Praat software (Boersma and Weenink, 2001) to measure the response times for each item and each participant. Specifically, we delimited 160 periods of 2.5 s from each audio file (one per participant) that correspond to the repetition time for scan acquisition. Within each identified scan, we delimited the beginning of oral response according to the audio-spectrogram (Figure 1). Response latencies were calculated based on the difference between the beginning of oral response and the beginning of the scan, given that pictures were presented at the beginning of each scan. Based on oral responses, the correct response rate (% of CR) and the mean reaction time (RT, in milliseconds) for correct responses were calculated. The run lasted 7.06 min.

The study was performed in a whole-body 3T MR scanner (Philips Achieva) with a 32-channel head coil. For functional scans, the manufacturer-provided gradient-echo/T2*-weighted EPI method was used. Forty-four interleaved axial slices parallel to the bi-commissural plane were acquired. Slice thickness was 3 mm. The in-plane voxel size was 2.3 × 2.3 mm (220 × 220 mm field of view acquired with 88 × 85 pixel data matrix, reconstructed with zero filling to 96 × 96 pixels). The main sequence parameters were TR = 2.5 s, TE = 30 ms, and flip angle = 80°. A T1-weighted high-resolution three dimensional anatomical volume was acquired, via a 3D T1 TFE sequence (field of view = 256 × 240 × 160 mm; resolution: 0.89 × 0.89 × 1 mm; acquisition matrix: 272 × 250 × 160 pixels; reconstruction matrix: 288 × 288 × 160 pixels).

When controlling for naming latencies (ANCOVA analysis), the difference between YG and OG on L_AP indices (see above) remained significant (F_{(1, 26)} = 6.450, p = 0.018). This result indicates that the increase left temporo-parietal asymmetry in OG was not related to increase naming latencies with age.

Mann-Whitney test comparing OG/RT+ and OG/RT− (see Table 6b) indicates that faster older participants (OG/RT+) showed less inter-hemispheric asymmetry on the A_LR indices (Z = −2.236, p = 0.026) than slower older participants (OG/RT−). The difference was not significant for other HR indices (P_LR, p = 0.902; L_AP, p = 0.620; R_AP, p = 0.805).

Based on cognitive scores, our older participants were less efficient for executive functions and visual analysis, in line with other studies showing aging effects for executive functions (Tomer and Levin, 1993; Cepeda et al., 2001; Ashendorf et al., 2008; Turner and Spreng, 2012), perceptual processing and speed of sensory-motor processing (Cerella et al., 1990; Salthouse, 2004). However, in our study, working memory (backward digit span) and verbal short-term memory (forward-digit span) remained intact in older participants, contrary to other findings (Hultsch et al., 1992; Wang et al., 2011; Logie and Morris, 2014). The fact that, in this study, older adults were well educated and had high socio-cultural levels might explain the preservation of some cognitive abilities (Orsini et al., 1986). Despite adequate scores for both short and long term memory, older participants reported subjective memory complaints more frequently. However, it has been shown that subjective memory complaints may be unrelated to objective deficits (Bolla et al., 1991; Mol et al., 2006). Finally, verbal automatisms increased in older participants, suggesting an increase of overlearned knowledge and automatic retrieval processes with age. Age differences for verbal automatisms scores may point to a cohort effect that could potentially skew interpretations of the results (see Froger et al., 2009 who compared verbal automatisms between older groups only, for example). Overall, our findings are in agreement with previous studies suggesting that older adults show a greater decline in executive functions than in language functions (Glisky, 2007; Harada et al., 2013).

The main contrast, Task minus Control, revealed that object naming elicited activation of the classical cerebral network reported for this task, including bilateral FG (perceptual and semantic processes; Whatmough et al., 2002; Mion et al., 2010; Ding et al., 2016), bilateral ITG (conceptual lexical retrieval; Indefrey, 2011), bilateral hippocampus (semantic memory retrieval; Sawrie et al., 2000; Binder and Desai, 2011), left IFG (lexico-semantic selection and phonological processes; Binder et al., 2009; Indefrey, 2011) and left pre-central gyrus (syllabification and articulatory processes, output phonology; Indefrey, 2011). Based on naming results, older participants were as accurate as younger ones, suggesting unimpaired sized vocabulary and lexical knowledge. This is consistent with other findings (Villardita et al., 1985; Wierenga et al., 2008; Boudiaf et al., 2016; see Goulet et al., 1994 for a review), allowing us to evaluate possible strategic differences between older and younger adults in manipulating lexical information during object naming. Indeed, a first indication of differential strategies according to age is that, despite normal accuracy, older adults were significantly slower than younger adults. The increased response times in older adults have been frequently reported by other studies and are attributable to various causes, including general slowing of processing speed (Salthouse, 1996; Feyereisen et al., 1998), decline of executive functioning (Craik and Byrd, 1982; West, 2000; Lustig et al., 2007), decline of working memory (Kemper and Sumner, 2001; Waters and Caplan, 2005), decline of perceptual processes (Baltes and Lindenberger, 1997; Schneider and Pichora-Fuller, 2000), and a deficit in lexical access (Bowles, 1989; Barresi et al., 2000; Mirman and Britt, 2013). Lima et al. (1991) proposed that a common general slowing of processing speed can explain aging effects for both lexical (domain-specific) and non-lexical (domain-general) processes. In the same vein, Rogalski et al. (2011) showed that longer RTs for naming in older participants were attributable to a slowing down of general processing, rather than to perceptual or contextual deficits. The idea that naming speed is related to executive and/or other general processes is also consistent with our previous findings (Baciu et al., 2016; Boudiaf et al., 2016) on lexical production with aging. Furthermore, we observed that the OG might be divided into two groups based on naming latencies, the faster OG (OG/RT+) and the slower OG (OG/RT−). The two subgroups differed only on naming latencies, raising the question of what mechanisms led to the situation in which the OG/RT+ group showed shorter RTs than the OG/RT− group in the context of cognitive reserve theory. The following section is divided into two parts, (a) the relation between preserved naming accuracy and HR mechanisms with aging (younger vs. older participants); and (b) the relation between naming latencies and HR mechanisms with aging and between older participants.

As described in the previous sections, our analyses were mainly based on the calculation of two intra- (L_AP and R_AP) and two inter-hemispheric (A_LR and P_LR) HR indices reflecting patterns of reorganization. Based on previous studies, we initially hypothesized that two strategies could be developed by older adults to perform lexical production: a semantic strategy suggesting supplementary involvement of semantic processes (Ansado et al., 2013; Lacombe et al., 2015) and/or an executive strategy assuming that in case of difficulty in performing the task, older adults would recruit executive functions to retrieve and generate words faster (Wierenga et al., 2008). We observed that L_AP asymmetry was higher for older participants than younger ones, suggesting the use of a semantic strategy to perform naming with aging. Our findings are in line with Ansado et al. (2013) and Lacombe et al. (2015), as these authors suggest that older participants rely on preserved semantic mechanisms by over-recruiting left temporo-parietal regions to perform the task. We posit that this result, referred as the left anterior-posterior aging effect (LAPA), reflects a domain-specific (linguistic) strategy to maintain performance in older adults for tasks involving semantic processes. However, further studies are needed to determine the direct relationship between semantic processing and the LAPA effect. Our study only shows an indirect relation based on anatomical regions, but has the virtue of highlighting a new HR mechanism that was not discussed otherwise. The LAPA effect was not mediated by naming latencies, reinforcing is compensatory role to naming.

In the older group, participants with faster naming latencies (OG/RT+) showed reduced anterior inter-hemispheric asymmetry as compared to participants with slower latencies (OG/RT−). This suggests a compensatory role for the bilateral frontal regions, in line with the HAROLD model (Cabeza, 2002; Cabeza et al., 2002). Similar effects and interpretations were previously revealed for other processes, such as episodic memory, working memory, and inhibitory control (see Cabeza, 2002 for a review). Altogether, these findings indicate that bilateral frontal recruitment reflects domain-general, unspecific compensatory mechanisms occurring with age. Cappell et al. (2010) reported that the recruitment of right prefrontal cortex for a verbal working memory task was dependent on the task demands in older participants. Specifically, the right prefrontal cortex was additionally recruited for low task-demands and was associated with normal performance. With higher task demands, the recruitment of prefrontal cortex decreased in parallel with performance, reflecting its failure to adapt to increased task demands above a certain threshold. In our study, the over-activation of right frontal regions in faster older adults may reflect compensatory recruitment of executive and attentional processes (Weissman and Banich, 2000). However, when task demands become too high, older participants fail to recruit the right hemisphere, resulting in increased naming latencies. In relation to the cognitive reserve theory (Stern, 2002, 2009), we suggest that the ability to recruit bilateral frontal regions during naming reflects higher neural networks flexibility to cope with higher task demands. These preliminary results need to be validated by further experiments to understand better the differential recruitment of frontal executive functions in older adults, according to their performance and the level of task-demands. Indeed, task demands are classically induced by the task (e.g., low demand task conditions vs. high demand task conditions). For example, it would be interesting to evaluate the effect of task-demands induced by the task (e.g., using low and high frequency words) to determine the relationship between bilateral frontal recruitment (HAROLD model) and the ability to cope with task-demands (cognitive reserve).

Overall, our results suggest at least two co-existent HR patterns and two strategic mechanisms at play in object naming, according to age and to performance: the LAPA effect and the HAROLD model. The newly described model, the LAPA effect, entails an intra-hemispheric left asymmetry with greater involvement of temporo-parietal regions explained by supplementary access to semantic resources engaged to perform a lexical production task. In contrast, the HAROLD model, also tested in older adults, is related to more domain-general mechanisms, consisting of supplementary recruitment of frontal executive regions to maintain a good level of performance. Our findings—mainly the LAPA effect—add to other existing models already described in the literature, such as the CRUNCH (Reuter-Lorenz and Cappell, 2008) and STAC (Reuter-Lorenz and Park, 2014 for a revised version) models. However, these models suggest that multidimensional sources may contribute to the diverse mechanisms and strategies developed with aging, such as anatomo-functional properties (e.g., brain volume, white matter integrity, neural specificity), functional reorganization (e.g., bilateral frontal recruitment, neurogenesis), life experience (e.g., intellectual activities, stress), and learning or training. Thus, assessment of aging effects on cognitive tasks, should be considered in the context of many other factors including occupation, daily and social activities, nutrition, and exercise (Fratiglioni et al., 2004; Paillard-Borg et al., 2009; Allès et al., 2012; Small et al., 2012).

A main limitation of this study is related to the sample size, as only 27 participants took part in our experiment, limiting the statistical power of our analyses. As a consequence, we used non-parametric analyses that are adapted for small sample size, allowing us to compare subgroups (YG vs. OG; OG/RT+ vs. OG/RT−). Furthermore, the small sample size limited our analysis approach to selecting subject-specific ROIs for the computation of HR indices. Hence, having only one group of participants, it was not possible to select activity peaks from one sample, and to analyze BOLD-contrast signal changes from the another. The later approach is more conservative in removing serial testing effects (Kriegeskorte et al., 2010), especially when studying age effects. By using the leave-one-out methodology, we posit that the deep circularity issue, although not totally removed, is diminished (Esterman et al., 2010). Further studies are needed to improve detection sensitivity by increasing the number of participants.