Explaining tip-of-the-tongue experiences in older adults – contributions of brain function, structure, and perfusion in relation to older adults’ cardiorespiratory fitness

Cognitive decline associated with healthy ageing is complex and multifactorial: brain-based and lifestyle factors uniquely and jointly contribute to distinct neurocognitive trajectories of ageing. To evaluate existing models of neurocognitive ageing such as compensation, maintenance, or reserve, we explore how various known brain-based and cardiorespiratory fitness factors intersect to better understand cognitive decline. We tested 73 (Mage = 65.51) neurologically healthy older adults and collected neuroimaging (functional, structural, and perfusion MRI), cardiorespiratory fitness, and behavioural performance data to investigate a well-documented, prominent cognitive challenge for older adults: word-finding failures. We aimed to uncover associations between predictors, which have previously been theoretically-implicated, in explaining age-related tip-of-the-tongue rates. Commonality analyses revealed that functional activation of language networks associated with tip-of-the-tongue states is in part linked with age and, interestingly, cardiorespiratory fitness levels. Age-associated atrophy and perfusion in regions other than those showing functional differences accounted for variance in tip-of-the-tongue states. Our findings can be interpreted in the context of the classic models of neurocognitive ageing suggesting compensation. Our findings moreover suggest that brain health indices in concordance with cardiorespiratory fitness measures have the potential to provide a more holistic explanation of individual differences in age-related cognitive decline. Highlights The incidence of word-finding failures is associated with brain health and cardiorespiratory fitness factors Language network activation associated with word-finding failures is linked to age and cardiorespiratory fitness levels Distinct contribution of brain structure and perfusion are also associated with word-finding Brain health indices in concordance with cardiorespiratory fitness measures have the potential to provide a more holistic explanation of individual differences in age-related cognitive decline

• Brain health indices in concordance with cardiorespiratory fitness measures have the potential to provide a more holistic explanation of individual differences in age-related cognitive decline

Introduction
The proportion of older adults in the global population is rapidly increasing, a trend most pronounced in high-income countries (WHO, 2022).Despite this, there are considerable gaps in our understanding of how we can promote maintenance of cognitive function in later life.Even in the absence of pathology in older age, cognitive decline-including difficulties with memory, processing, and communication skills-can limit everyday functioning and restrict independent living (Clouston et al., 2013;Njegovan et al., 2001).One prominent example of this is the increasing difficulty in older age with word finding and a related phenomenon known as tip-of-the-tongue states.These are impermanent failures to retrieve and produce intended words; lapses where we cannot call upon a word that we know.
Research has shown that incidences of tip-of-the-tongue states increase with age (Huijbers et al., 2017;Salthouse & Mandell, 2013;Schwartz & Frazier, 2005).Contrary to common belief, however, tip-of-the-tongue states are not indicative of significant memory decline.
Older adults typically possess a larger vocabulary than younger adults (Salthouse, 2019).
Instead, tip-of-the-tongue states are indicative of a breakdown in the transmission process between two sequential speech production stages: first retrieving word meaning and then its associated phonology, or sound (Burke & Shafto, 2011).
Virtually all speakers experience tip-of-the-tongue states during regular, everyday conversation but they are of particular significance to older adults, who regard them as their most frustrating and troubling cognitive failure (Ossher, Flegal, & Lustig, 2013).
Understanding why tip-of-the-tongue states increase in ageing is important for several reasons, including identification of pathology and social participation.Discerning what constitutes 'normal' variability in tip-of-the-tongue states across different age groups might be helpful in classifying when these lapses may signal pathological declines, such as mild cognitive impairment or Alzheimer's Disease.In fact, difficulties with naming are sometimes the most apparent and limiting symptom during the early stages of some dementias (e.g., Savage et al., 2013).Additionally, there is potentially a socio-emotional cost with tip-of-thetongues: if there is recurrent interference with conversation, older adults are more likely to withdraw from and avoid social interaction, ultimately leading to a poorer quality of life.
Overall, while tip-of-the-tongue states are highly prevalent and troubling for older adults, the exact mechanisms underlying them are not fully understood.
Existing neurocognitive accounts show that tip-of-the-tongue states are associated with structural and functional changes in the brain (Huijbers et al., 2017;Resnik et al., 2014;Shafto et al., 2007;2010), and behavioural performance studies have found that cardiorespiratory fitness levels modulate individuals' abilities to resolve tip-of-the-tongue states (Segaert et al., 2018).Indeed, more generally, cardiorespiratory fitness has been shown to be an important factor in successful neurocognitive ageing across studies of brain function (Kawagoe, Onoda, & Yamaguchi, 2017;Prakash et al., 2011;Voss et al., 2010), brain structure (Colcombe et al., 2003;Szabo et al., 2011), and cognitive performance (Barnes et al., 2003;Hayes, Forman, & Verfaellie, 2016).However, returning to the issue of tip-of-the-tongue states, an integrated explanation of these highly prevalent cognitive failures, which includes cardiorespiratory fitness as a variable in addition to brain-based measures, is currently lacking.To address this gap, it is essential to examine how fitness and brain-based factors, including function, structure and perfusion, contribute to age-related changes in tip-of-the-tongue states.These insights need to be aligned with established models of cognitive ageing, which we turn to next.Key models of healthy ageing centre around mechanisms of maintenance, compensation, or reserve and typically assume contributions of both genetic and lifestyle influences, such as an individual's fitness level, to the neurocognitive mechanisms of healthy ageing (Cabeza et al., 2018).Models of maintenance are established on maintaining or returning neural resources to their former levels in response to the typical 'wear and tear' associated with non-pathological ageing (Habeck et al., 2017;Nyberg et al., 2012).As such, fitness levels are predicted to promote brain maintenance by preserving youthful brain structure and cerebrovasculature, potentially supporting brain function and cognition through grey matter preservation and optimal cerebral perfusion.Models of compensation entail functional recruitment of additional resources to meet high cognitive demands when there is an insufficiency in neural resources (e.g., brain atrophy or hypoperfusion) (Reuter-Lorenz & Cappell, 2008).Typically, observations of increased recruitment of particular neural regions in older (vs younger) adults are taken as evidence for compensation, with the presumption that these increases offset neural decline elsewhere in the brain (Knights et al., 2024).
Nevertheless, a key assumption with this model is that the degree of compensatory involvement is directly related to the degree of cognitive performance, an assumption which is not tested directly in most previous studies.Lastly, models of reserve typically posit the idea that fitness, education, or crystallised intelligence (which are accepted to be proxies for reserve) can contribute to accumulated neural reserve, which can be structural and/or functional and develops over time (e.g., through childhood or early adulthood) (Stern et al., 2020;Tucker & Stern, 2011;Whalley et al., 2004).This reserve is thought to protect against age-related neural pathology.Consequently, proxies of cognitive resilience are expected to correlate with brain functional resources involved in specific cognitive processes, rather than structural changes.As alluded to above, tip-of-the-tongue states signify difficulties with accessing phonology (i.e., sound form representations).When one produces a spoken word, first the meaning of the word is retrieved, and then the associated phonology is accessed (Abrams & Davis, 2016;Levelt, 2001).For most people, most of the time, this process is effortless.
However, in older age, connections throughout the lexical system are weakened, and the transmission from meaning to phonology is thought to be compromised, resulting in increased occurrences of tip-of-the-tongue states (Burke, MacKay, Worthley, & Wade, 1991).The processes contributing to this type of cognitive failure thus involve contributions that are distinct from general cognitive slowing or memory (Salthouse & Mandell, 2013).
Complementing these psycholinguistic accounts, several neuroimaging studies have demonstrated neural markers associated with tip-of-the-tongue states: lower grey matter density in brain areas implicated in phonological production, in particular the left insula is related to greater frequencies of tip-of-the-tongue states (Shafto et al., 2007), and reduced functional activation in this same brain region has been demonstrated for older adults (Shafto et al., 2010).Several studies have demonstrated distinct neural signatures for successfully recollected responses compared to tip-of-the-tongue states (Galdo-Alvarez, Lindın, & Diaz, 2009), including recruitment of the wider lateral prefrontal cortex (Huijbers et al. 2017).In addition to the neural accounts of tip-of-the-tongue states, modifiable lifestyle factors, such as fitness, have also been shown to decrease the probability of experiencing tip-of-the-tongue states in healthy older adults (Segaert et al., 2018).However, these individual and isolated accounts have led to a fragmented picture; how different measures of brain health (structural, functional, and vascular) intersect with each other as well as with cardiorespiratory fitness in determining age-associated cognitive decline remains largely an underexplored area.
The present research aims for a more comprehensive account of how ageing leads to cognitive lapses, such as tip-of-the-tongue states.A multi-modal approach, integrating brain-based and lifestyle assessments, is necessary to examine how each factor uniquely and collectively explains this phenomenon.A suitable approach to address this is commonality analysis (Nimon & Oswald, 2013;Nimon et al., 2008), as it partitions explained variance into unique and shared contributions, providing a clearer understanding of predictor relationships compared to traditional regression analysis, which can be distorted by multicollinearity, especially when age is a factor (see Nimon, 2010;Nimon et al., 2008;see Wu et al. 2023 for a recent and comprehensive example of applying commonality analysis to neuroimaging data).
Using commonality analysis, we assessed the unique and shared effects of brain function (networks revealed by task-based fMRI), grey matter volume, cerebral perfusion, cardiorespiratory fitness via a gold-standard lab-based incremental treadmill test, education and age to explain tip-of-the-tongue behaviour.Together, this approach allowed us to test the following predictions, associated with key neurocognitive ageing models: (1) Brain maintenance may be indicated by shared signals among grey matter, perfusion, and brain activity associated with tip-of-the-tongue incidence.This supports the notion that preserving youthful brain structure and function is key to maintaining cognitive health.(2) Compensation may be reflected by unique associations between activity in newly recruited regions and performance during tip-of-the-tongue states, not shared with grey matter (i.e., this functional recruitment would offset atrophy elsewhere).( 3) Evidence for reserve could be reflected in shared coefficients between education or crystalised intelligence and brain activation with tip-of-the-tongue performance.( 4) Importantly, if fitness levels also share signals with indicators of either maintenance, compensation or reserve, it suggests that fitness promotes these adaptive brain conditions.
As such, different neurocognitive models of ageing, centring around maintenance, compensation, or reserve, make different predictions about which aspects of brain health will be modulated by fitness levels in explaining older adults' abilities to resolve tip-of-the-tongue states.It is not clear if one, multiple, or indeed any of the models align with age-related changes in tip-of-the-tongue states.Our objective in the current contribution, therefore, is twofold: 1) to provide a holistic account of tip-of-the-tongue states in ageing, incorporating brain health and cardiorespiratory fitness measures, and 2) to embed our findings within and provide support for existing frameworks of neurocognitive ageing.To test the suppositions of these models in a single contribution requires the capture of multimodal data, ideally in a single design.Here, for the first time, we provide such a contribution.

Results
Participants (N = 73, age range 60-81; see Table 1) reported to the research centre on separate days, first for neuroimaging (structural, functional, and perfusion MRI) and then for cardiorespiratory fitness testing (V O 2 peak).Prior to this, they received extensive health screening (see Methods).Structural anatomical images were acquired using a T1-weighted sequence and cerebral perfusion (cerebral blood flow, CBF) was obtained using a pseudocontinuous arterial spin labelling (pcASL) protocol.Functional brain activity during a tip-ofthe-tongue task-which comprised definitions where participants responded with either Know, Don't Know, or TOT (i.e., tip-of-the-tongue)-was recorded using blood-oxygen-leveldependent (BOLD) fMRI, with TOT > Know as the primary statistical contrast.Fitness testing was performed using an incremental treadmill test where respiratory gases and heart rate were continuously monitored as participants advanced through progressive stages until volitional exhaustion.
The data in the current report were acquired as part of a larger intervention study titled Fitness, Ageing, and Bilingualism (www.fab-study.com),which was detailed in a single preregistration on the OSF: https://osf.io/d7aw2.The preregistration contains a multitude of hypotheses and analyses plans that form the basis of other, separate reports (e.g., Fernandes et al., 2024;Fosstveit et al., 2024;Markiewicz et al., 2024).No one specific hypothesis in the preregistration is related to the current contribution, though several hypotheses related to Research Question 3 pertain to the relationship between fitness-levels and each of the brain-based measures included.All datatypes and several analysis steps used in the present report were specified in this preregistration.Note: V O 2 peak (mL/kg/min).BMI = Body Mass Index (kg/m 2 ).Education: Compulsory = mandatory education under the British (English) system; Further = Post-16 study after secondary education (and prior to university); Undergraduate = bachelor's degree or equivalent; Postgraduate = qualification undertaken after a bachelor's or undergraduate degree.Next, we analysed all data modalities with one joint approach using commonality analysis, to determine a parsimonious combination of factors-which were all individually theoretically-motivated, as shown above-in explaining healthy older adults' tip-of-thetongue rates.Complementing the bivariate functional analyses, commonality was found between age and function in explaining variance of tip-of-the-tongue rates in regions including the cerebellum, middle temporal gyrus, inferior frontal gyrus (pars triangulars), and superior frontal gyrus (medial part).See Table 3 and Figure 3B for details of significant contrasts and their associated coefficients.Note that the regions reported for the age and function common effect show some degree of dissimilarity relative to the regions reported previously in the pre-commonality analysis.The reasons for this are twofold.First, only regions that expressed commonality in age and function and survived stringent thresholding are reported for the commonality analyses; and second, clusters containing fewer than 150 voxels were not reported in the pre-commonality analyses for the sake of brevity but are reported here due to overall smaller effects.Interestingly, and key to the present research question, we also found commonalities between age, function, and Vȩ O 2 peak in explaining age-related tip-of-the-tongue variability.
While these effects were not widespread, with the effect being localised to the left lingual gyrus, the finding was robust across multiple iterations of commonality modelling (see below for more information on models including education and vocabulary size).Building on age and function above, the concomitant effect of age, function, and Vȩ O 2 peak suggests that agerelated functional recruitment of processing tip-of-the-tongue states is linked to differences in cardiorespiratory fitness.It is worth noting, however, function did not produce a shared effect with brain structure or perfusion, suggesting that the functional effects are over and above any differences in atrophy and/or perfusion.
Further, a shared effect of age and structure in the right middle temporal pole also emerged as significant in explaining ageing tip-of-the-tongue states, which suggests a link between age-related atrophy and age-related increases in tip-of-the-tongue rates.Finally, a shared effect of age and perfusion emerged as significant in explaining ageing tip-of-thetongue states, implying a link between age-related vascular changes and age-related increases in tip-of-the-tongue rates.The distinct effects of brain structure, perfusion, and function suggest that the mechanisms by which these three indices of brain health contribute to tip-of-the-tongue in healthy ageing are independent.
In addition to the core model of interest outlined above, we ran additional analyses which included the variables education and vocabulary size.We considered these variables as indicators of crystallised intelligence and found a moderate-to-strong positive correlation between them (r = .638,p <.001).However, neither education nor vocabulary were significant-uniquely or jointly with other predictors-in explaining variance in tip-of-thetongue occurrences (see Supplementary Materials for details and output of additional models).

Discussion
Here, for the first time, we bring together multiple brain-health measures and a goldstandard assessment of fitness to explain the differences of word-finding difficulties and more broadly cognitive decline in older adults.Commonality analyses revealed that functional recruitment of language networks associated with resolving tip-of-the-tongue states differed with age.Crucially, functional activation of these language networks was also associated with individuals' fitness levels, in explaining tip-of-the-tongue states.Finally, ageassociated loss of brain structure (or atrophy) and age-related cerebrovascular differences in brain regions other than those showing functional differences accounted for the variance in tip-of-the-tongue state incidence.
Our findings can be interpreted in the light of key neurocognitive models of ageing, such as maintenance, compensation, or reserve.We found no evidence supporting maintenance and reserve indicators: neither grey matter nor perfusion, nor indicators of reserve such as education, showed shared effects with task activity on predicting tip-of-thetongue behaviour.We did, however, find evidence for compensation: there was a shared effect of functional recruitment, cardiorespiratory fitness, and age predicting tip-of-the-tongue incidence.The unison of these factors in explaining variance in age-associated tip-of-thetongue states suggests that both fitness and age modify the functional recruitment of wordfinding areas in the brain.While some previous studies found evidence of fitness levels contributing to altered brain function (Douw et al., 2014;Prakash et al., 2011;Voss et al., 2016;Wong et al., 2015), these studies did not formally explore the connections to cognitive performance as examined in the present study.Future studies could use our present holistic approach integrating over multimodal brain-health combined with lifestyle measures to examine other indicators of language abilities as well as other domains of cognition.Our findings indicate that older adults with higher fitness levels display distinct behaviourallyrelevant brain activity compared to their less fit counterparts, providing a brain-based explanation for the bivariate relationship observed in previous research between fitness and tip-of-the-tongue states (Segaert et al., 2018).
Our findings also build upon previous work which has found that atrophy (Shafto et al., 2007) was associated with age-related tip-of-the-tongue rates.Specifically, age-related increases in tip-of-the-tongue states have been associated with atrophy in the left insula, a region implicated in phonological production (Shafto et al., 2010).Similarly, in the current contribution, we report commonality between age and brain structure, albeit not in the insula region but in the right middle temporal pole.Further to functional and structural brain effects, we also observed that individual differences in cerebral perfusion, as a function of age, explained variation in tip-of-the-tongue states.The literature on the association between age-related changes in cerebral blood flow, cognition, and wider brain health is unresolved (for a review on perfusion and cognition in ageing, see Ogoh, 2017): while some accounts report perfusion maintenance in old age supporting brain function and cognition (De Vis et al., 2018;Wu et al., 2023), others have shown mixed results (Espeland et al., 2018;Leeuwis et al., 2018;Poels et al., 2008).With our data, we bring further clarity by extending the literature to show that cerebral blood flow is an important contributor to age-related tip-ofthe-tongue states.
We focused on tip-of-the-tongue variability as an example of cognitive decline in older age.Failure to produce known words is a common human experience; it often leads to frustration and disrupts the natural flow of verbal communication.While it is not unique to ageing, healthy older adults tend to display more tip-of-the-tongue occurrences relative to their younger counterparts.This age-related pattern cannot be explained by memory deficits alone-indeed, semantic and phonological features of the intended yet elusive word can often still be articulated (Brown, 1991;Brown & McNeill, 1966;Miozzo & Caramazza, 1997).
Moreover, previous work has shown the independence of memory and word retrieval/production systems (Salthouse & Mandell, 2013).We demonstrate, for the first time, that tip-of-the-tongue variability in ageing can be potentially explained by variability in brain function, structure, and perfusion in addition to fitness, in one analytical framework, and provide a theoretical and methodological step forward in bringing together seemingly disparate streams of data to profile a recognisable, everyday age-related irritation.Our approach could be applied to other aspects of cognitive function which are relevant to characterising age-related cognitive decline.

Limitations
There are several other potential extensions to the present work.Our cross-sectional study cannot directly show individual progression over time (i.e. the ageing process).
Longitudinal studies are needed to clarify the conditions and sequences of events behind our findings.This could include longitudinal intervention work, as changes in lifestyle, for example, increasing fitness through changes in regular physical activity accumulated over time, would help to elucidate the framework we have set out in the present work.Moreover, while we examined brain activations and co-activations, we did not quantify functional connectivity (Bethlehem et al., 2020;Geerligs et al., 2017;Liu et al., 2023;Tsvetanov et al., 2016).Both aspects may change independently during cognitive ageing, requiring further investigation (Tsvetanov et al., 2018).We believe that our current contribution provides a sound theoretical basis and analytical framework to motivate future longitudinal designs.
Neurocognitive models of ageing are continually evolving, as are the ways in which these constructs are best studied (Stern et al., 2020).Future work should validate our findings using diverse approaches to define and assess these constructs.Our method of recording education data -a proxy for cognitive reserve-was suboptimal, relying on selfreported categorical data with low variability.The lack of effect with education does not suggest that education is an unimportant variable in characterising cognitive reserve.Future studies should use more sophisticated and sensitive methods to measure education and crystallised intelligence, as proxies of cognitive reserve.Moreover, our understanding of what constitutes cognitive reserve is evolving.Factors such as physical activity (and resultant changes in fitness levels) or bilingualism may also contribute to individual variations in cognitive reserve (Cabeza, 2018).
We also note that while our study included participants aged 60 to 81, 84% of participants were aged 60-69, 13% were aged 70-79, and only 3% were aged 80 or above.
While this inequality in age distribution is understandable given the stringent nature of our inclusion criteria, it nonetheless means that the current analysis perhaps lacked sensitivity at the highest bands of age in our sample, and we were unable to test whether the reported effects differed across older age groups.We do, however, contend that chronological age is only one component that determines cognitive ageing; here, a much wider perspective on how the effect of ageing is potentially modified by cardiorespiratory fitness, and how this then downstreams into cognition, is provided.Nevertheless, sampling more widely in the future would be an improvement on the current work.Lastly, in the present work, individual participants' arterial transit time (i.e., the time it takes blood to travel from an arterial input to a brain region) measures were used to the correct our cerebral blood flow assessment for arterial transit time-differences, improving the estimation accuracy of cerebral blood flow.
The measure of arterial transit time could be added in future work to supplement the standard cerebral blood flow measure (Feron et al., 2023).The inclusion of arterial transit time may highlight potential commonality with functional BOLD signal (e.g., if blood takes longer to reach an 'activated' area due to a prolonged arterial transit time, it may delay the onset of, and therefore distort, the BOLD signal response).

Conclusion
In sum, our multi-modal approach, which integrates brain structure, function, perfusion, and fitness measures offers a useful framework for advancing research on cognitive ageing.We demonstrate that a widespread and recognisable feature of language processing is explained by shared signals of brain activation and fitness levels, without significant contributions from grey matter volume and perfusion differences.This suggests that cardiorespiratory fitness is associated with brain function and contributes to cognitive performance in older adults, potentially promoting compensatory mechanisms.

Definition Correct Answer
Noun Class The termination of the existence of a particular kind of organism or species

Extinction Common
A word borrowed from German that means the "spirit of the times"

Zeitgeist Common
The boy king of Egypt whose tomb was discovered largely intact in 1922 Tutankhamun Proper A true TOT was recorded only when participants indicated they were experiencing a tip-of-the-tongue state and pressed YES when asked "Is this the word you were thinking of?" in a subsequent verification slide (where participants pressed NO in response to the verification slide, these trials were considered in the same manner as Don't Know responses).The definition texts were presented for a fixed duration of 12000 ms which was followed by a 2000 ms ISI and, in the case of TOT responses, verification slides were presented for a maximum of 6000 ms (or sooner with button press).A jittered ITI was presented for an average of 5750 ms (range 3500 -8000 ms; 500 ms increments).A total of 200 unique definitions were presented across four scanning blocks (50 per block); correct answers (targets) pertaining to definitions were arranged so that the number of proper and common nouns, syllable, phoneme, and letter count, as well as word frequency were counterbalanced across the four experimental blocks.The task was written in PsychoPy (v2021.1.4) and text was presented in white Open Sans font on a grey (normalised RGB 0.5, 0.5, 0.5) background projected onto a screen using a ProPixx visual projector system (1440 Hz refresh rate) and viewed by a mirror-system attached to the MRI headcoil.An MRIcompatible button box was used to register Know, Don't Know, and TOT responses.Before entering the scanner, participants performed a practice run of the TOT task with 12 definitions not included in the experimental stimuli-during this practice phase, an explanation of TOT states was provided to ensure participants understood the task.There was no limit on practice; participants could repeat the practice task as many times as they wished.The set-up and parameters of the out-of-scanner practice task was identical to the in-scanner experimental task.

Cardiorespiratory Fitness
Participants completed an incremental exercise test on a treadmill (Pulsar 3p, H/P/Cosmos, Germany).Respiratory gases (V O 2 ; oxygen consumption, V CO2; carbon dioxide production) were recorded continuously using a facemask (7450 V2, Hans Rudolph, USA) and metabolic cart (JAEGER Vyntus CPX, Vyaire, USA), as was heart rate and rhythm using a 12-lead ECG (Cardiosoft, Vyaire, USA).Participants warmed up on the treadmill before completing 4-min walking stages with a 1-min rest period between each stage.Treadmill speed started and remained at 3.8 km/h until either all elevation stages were completed (7 possible stages: 4, 7, 10, 13, 16, 19, and 20%) or individual lactate threshold was reached.
Finger-prick blood lactate was measured during each rest period (Biosen C-Line, EKF Diagnostics, United Kingdom).If all elevation stages were completed, 4-min stages continued with speed increasing 0.5 km/h per stage until lactate threshold was reached or until volitional exhaustion.If lactate threshold was reached, participants immediately began 1-min stages where just the speed increased 0.5 km/h per stage until volitional exhaustion (rest periods were removed).
Participants were asked to exercise to volitional exhaustion unless halted by the researcher due to ECG abnormalities, injury, or the presence of a plateau in heart rate or V O 2 .Cardiorespiratory fitness was measured as peak oxygen consumption (V O 2 peak) recorded during the treadmill test, determined as the mean of the two highest 30 s V O 2 intervals recorded.For participants who only completed a sub-maximal test (N = 8), acquired using these same parameters with the PLD set to 2000 ms and no inversion pulses.Acquisition time was ~17 minutes.The previously described T1-weighted anatomical images were used for differentiation and segmentation of grey and white matter so that resting CBF in only grey matter could be assessed.
Partial volume error correction and adaptive spatial smoothing of the perfusion maps was performed using default settings in oxford_asl (Chappell et al., 2011;Groves et al., 2009).
Participants with abnormal post-label delay maps, often caused by motion but also due to incidental findings such that perfusion in part of the brain could not be detected, were excluded (N = 6).Grey matter masks were thresholded to ensure only voxels containing primarily grey matter were included.Areas within masks containing incorrect assignment to grey matter, primarily around the eyes and nasal cavity, were manually removed (N = 7).
ATT measures are less common in MRI perfusion measures, as often only one PLD is used.
In our work, per-participant ATT measures were used to correct CBF for ATT-differences, improving CBF estimation accuracy.ATT itself was used an outcome measure in another paper (Feron et al., 2023).Both individual-and group-level data were processed using FEAT (fMRI Expert Analysis Tool) in FMRIB's Software Library (FSL, www.fmrib.ox.ac.uk/fsl) v6.0.1.
Registration to the high-resolution T1-weighted structural (BBR [Boundary-Based Registration, Greve & Fischl, 2009]) and standard space images (MNI152 T1 2mm; 12 DOF transformation) was performed using FLIRT.Registration from high-res structural to standard space was then further refined using FNIRT nonlinear registration.After registration, the following pre-statistics processing was applied: motion correction using MCFLIRT (Jenkinson, 2002), slice-timing correction (interleaved) using Fourier-space timeseries phase-shifting, spatial smoothing using a Gaussian kernel of FWHM 5 mm, grandmean intensity normalisation of the entire 4D dataset by a single multiplicative factor, highpass temporal filtering (Gaussian-weighted least squares straight line fitting, with sigma = 50.00s).The time-series were analysed using FMRIB's Improved Linear Model tool with local autocorrelation correction (Woolrich et al., 2001) and Z (Gaussianised T/F) statistics images were thresholded non-parametrically using clusters determined by Z>3.1 and a corrected cluster significance threshold of p = .05.Whole-brain functional univariate analyses were computed within the general linear model using a multi-level mixed-effects design.Each fMRI run was modelled separately at the single-participant level.Trial-by-trial modelling commenced at trial onset and concluded at participant button press within only the definition presentation phase of the task.Participant responses, and therefore statistical modelling, were not captured once definitions disappeared from screen (BOLD signal during ISI/TOT verification slide was not modelled).Trials where participants failed to make a response were discarded.Each condition of interest or exploratory variable (TOT, Know, Don't Know) was convolved using a double-gamma hemodynamic response function.A second-level analysis combined first-level contrast estimates per participant using a fixedeffects model, and higher-level activation (third-level analysis) was computed using a mixedeffects model using FSL's FLAME (FMRIB's Local Analysis of Mixed Effects) tool.

Commonality Analysis
To explore the unique and shared contribution of TOT-related functional activity, grey matter density, cerebral blood flow (CBF), cardiorespiratory fitness (V O 2 peak) and age on behavioural TOT rates, we conducted a commonality analysis (Kraha et al., 2012;Nimon et al., 2008).Commonality analysis partitions the variance accounted for by all predictors in multiple linear regression into variance unique to each predictor and common variance shared between each possible combination of predictors.As such, unique effects specify the (orthogonal) variance accounted for by one predictor above and beyond that accounted for by other factors in the model, while common effects show the combined variance shared between correlated predictors.Coefficients of common effects can show how much variance is accounted for in the outcome variable mutually by two or more correlated predictors.For a more comprehensive account of subjecting neuroimaging and behavioural data to commonality analysis, please refer to Wu et al. (2023).The commonality toolbox used in the present analysis is available at https://github.com/kamentsvetanov/CommonalityAnalysis.
Significant clusters related to effects of interest (as outlined above) were reported, as identified with nonparametric testing using 1000 permutations and threshold-free cluster enhancement (TFCE) with significance level of a = .05(Smith & Nichols, 2009).
Sex differences were not of primary interest in the current contribution but are controlled for because they may covary with multiple components of our data set, including total brain volume (which affects grey matter quantification) and cardiorespiratory fitness levels (males typically show higher V O 2 scores).Sex was included in the modelling as a binary covariate of no interest.Head motion has been shown to lead to spurious outcomes in functional neuroimaging data (e.g., Power et al., 2012) and has been linked with out-ofscanner demographic and lifestyle factors such as age (Madan, 2018) and BMI (Beyer et al., 2021), which in turn correlate with fitness.To control for this, head motion was included as a covariate of no interest in the commonality modelling.Head motion data were derived from FSL MCFLIRT reports where motion was indexed as relative displacement (in mm), which is a measure of the mean head movement in relation to the subsequent volume.These scanspecific values were calculated per FMRI run, and then averaged across runs to compute overall head motion such that a single value represented each participant's total head motion.Recall, however, that the initial pre-commonality fMRI analysis was already motion corrected as part of the standard pre-processing of functional MRI data.
In addition to the primary model, a further model was run where the following variables were included as covariates: education and vocabulary size (depicted in Figure 1 is the primary model with initial covariates, as well as the additional model with two added covariates listed between brackets).Education data were captured as part of the initial screening procedure and vocabulary was captured via an online task where participants were required to select the correct synonym or antonym of a target word.Education was categorised according to the British (English) education system: "Compulsory" = Schooling completed up to and including the age of 16 (N = 21); "College/FE" = Post-16 education including A-Levels and Further Education (N = 25); "Undergraduate" = Undergraduate qualification at university level (N = 14); "Postgraduate" = Postgraduate qualification at university level (N = 13).As education has widespread influences in linguistic and nonlinguistic cognitive function (Zahodne, Stern, & Manly, 2015), is a purported component of cognitive reserve (Tucker & Stern, 2011), and is associated with cardiovascular health in ageing (Schultz et al., 2018), education was included in the modelling.

Figure 1
Figure1depicts the key bivariate relationships in our healthy older adult sample, corroborating findings from prior studies.With increasing age, the incidence of tip-of-thetongue states increased (Panel A) and cardiorespiratory fitness levels (measured as V O 2 peak) decreased (Panel B).Age did not significantly impact on overall cerebral blood flow (Panel C) (seeFeron et al., 2023 for more detailed discussion).FMRI-BOLD analysis of the TOT > Know contrast produced significant clusters of activation within the precuneus and angular, cingulate, and frontal gyri(Panel D, and Table 2, Huijbers et al., 2017).Voxel-

Figure 1 .
Figure 1.Bivariate analyses of the precursors to commonality analysis.A. Tip-of-the-tongue by age relationship (r = .288,p = .014).B. V O 2 peak by age as a function of sex (Males r = -.566,p < .001,Females r = -.407,p = .014).C. Cerebral blood flow by age (r = -.011,p = .926).D. Significant clusters (determined by Z>3.1 and a corrected cluster significance threshold of p = .05)from the Tip-of-the-tongue > Know contrast.Slices, presented in MNI space, begin (top left) at z = -40 and conclude at z = 80 (bottom right), with each progressiveslice representing an increment of +8.Heatmap depict z scores ranging from 3.1 to 9, with lighter colours representing higher z scores.E. Volumetric analysis depicting grey matter atrophy by age (voxel-based morphometry analysis with 10,000 permutations using threshold-free cluster enhancement).Heatmap represents z scores ranging from 3.1 to 6, with lighter colours representing higher z scores (greater age-related atrophy).All brain images presented in radiological convention.
Figure 2 summarises the main components of the present multimodal dataset, feeding into a commonality analysis.The term functional refers to functional MRI data derived from the contrast in BOLD signal between TOT and Know responses, structure or grey matter are used interchangeably to refer to the quantification of whole-brain grey matter density/integrity, and fitness, cardiorespiratory fitness, and O 2 peak are used to refer to cardiorespiratory fitness.Our core model of interest included the three brain-based datatypes, a measure of cardiorespiratory fitness, and age.As sex and in-scanner head motion can affect brain measures (Ruigrok et al., 2014) and/or are related to core predictors, these variables were additionally entered into the model as covariates of no interest.

Figure 2 .
Figure 2. Summary of the main components of the commonality analysis used to analyse the current multimodal data.For information on how each component was measured and modelled, please refer to Methods.CBF = cerebral blood flow.ASL = arterial spin labelling.

Figure 3 .
Figure 3. A: Venn diagram depicting the relationships, as established by the commonalityanalysis, where, for example, the intersection tip-of-the-tongue states on the one hand with age and function on the other hand, depicts that the commonality between age and function explains a significant portion of the variance in tip-of-the-tongue states.The size of the ellipses is only illustrative, it is not indicative of effect size.Our key finding is the common effect of age, fitness and function (i.e., black intersection) accounted for a significant portion of variance in tip-of-the-tongue states (which is depicted further in Panel B).Age-associated structural atrophy and perfusion in regions other than those showing functional differences also accounted for significant portions of variance in tip-of-the-tongue states (i.e., orange and purple intersections respectively) but these did not share commonality with fitness.B.Three-dimensional render depicting the age, function, and fitness common effect including a scatter presenting the relationship between this contrast and its intensity in the region of interest analysis (lingual gyrus).C. Age by function common effect.D: Age by structure common effect.Age and structure common effect did not converge with function or fitness.A significant age by fitness common effect was observed but is not visualised here.Heatmaps are indicative of t values, with brighter colours representing stronger effects (t range 1.80 -3.00).Commonality effects were observed using threshold-free cluster enhancement (TFCE) correction.∩ = intersection (commonality between stated predictors).

Figure 4 .
Figure 4. Schematic of the tip-of-the-tongue task.Trial lengths varied depending on the volume of tip-of-the-tongue responses (which triggered an additional verification slide that was not included for Know or Don't Know responses).TOT = tip-of-the-tongue.
Brain function.Four fMRI blocks (50 definitions per block) were run with an echoplanar imaging sequence with the following parameters: TR = 1500 ms, TE = 35 ms, slice thickness = 2.5 mm (2.5 mm isotropic voxel size), FOV = 210 mm, flip angle = 71 deg, phase encoding direction = A >> P, and interleaved slices.The total number of volumes per run varied according to the number of tip-of-the-tongue states reported; typically, ~610 volumes were collected per run (~16 min/run).

Table 1 .
Overview of core participant sample characteristics.

Table 3 .
Coordinates (MNI), size, t, and p values of significant commonality effects and their respective brain regions.
A unique effect of age was also present but not reported in the above table because an isolated age effect is uninformative and was not hypothesised.Note: R = Right.L = Left.CBF = Cerebral blood flow.