Statement of Research Significance
Research Question(s) or Topic(s): Standard neuropsychological tests lack sensitivity in the early stages of Alzheimer’s disease. In this study, we evaluated whether assessment of conjunctive binding abilities could reveal cognitive changes earlier than in standard assessment. Main findings: We demonstrated that impaired binding performance in visual short-term memory was evident as early as the preclinical stage of sporadic Alzheimer’s disease and was linked to tau pathology in the entorhinal cortex. These results suggest that binding performance could detect early tau pathology in the entorhinal cortex in cognitively normal individuals. Study contributions: Assessment of this conjunctive binding abilities could enable diagnosis earlier than standard neuropsychological assessment.
Introduction
Alzheimer’s disease (AD) is characterized by two proteinopathies acting synergistically in the brain: the amyloid (Aβ) and the tau pathologies (Hanseeuw et al., Reference Hanseeuw, Betensky, Jacobs, Schultz, Sepulcre, Becker, Cosio, Farrell, Quiroz, Mormino, Buckley, Papp, Amariglio, Dewachter, Ivanoiu, Huijbers, Hedden, Marshall, Chhatwal and Rentz2019; Nelson et al., Reference Nelson, Alafuzoff, Bigio, Bouras, Braak, Cairns, Castellani, Crain, Davies, Tredici, Duyckaerts, Frosch, Haroutunian, Hof, Hulette, Hyman, Iwatsubo, Jellinger, Jicha and Beach2012). These neuropathological changes accumulate slowly, years before the onset of cognitive symptoms, defining a silent asymptomatic or preclinical phase (Jack et al., Reference Jack, Knopman, Jagust, Petersen, Weiner, Aisen, Shaw, Vemuri, Wiste, Weigand, Lesnick, Pankratz, Donohue and Trojanowski2013). Currently, preventive clinical trials administer anti-Aβ or anti-tau drugs during this window to halt the neurodegenerative processes before cognition starts declining (Rafii & Aisen, Reference Rafii and Aisen2023). This approach necessitates screening older adults to identify asymptomatic individuals with the highest risk of rapid clinical progression. Previous studies have consistently demonstrated that cognitively normal (CN) individuals with high Aβ burden are at higher risk of future cognitive decline than CN individuals with low Aβ levels (Donohue et al., Reference Donohue, Sperling, Petersen, Sun, Weiner and Aisen2017; Hanseeuw et al., Reference Hanseeuw, Malotaux, Dricot, Quenon, Sznajer, Cerman, Woodard, Buckley, Farrar, Ivanoiu and Lhommel2021). However, not all CN individuals with elevated Aβ burden experience similar patterns of cognitive decline. A recent study demonstrated that CN individuals with both abnormal global Aβ deposition and tau burden in the medial temporal lobe (Aβ + Tau + MTL), as evidenced using positron emission tomography (PET) imaging, had a six-fold increased risk (risk = 50%) of short-term conversion (within 3–5 years) to symptomatic AD, compared to those CN individuals with only high Aβ burden (Aβ + Tau−, risk = 8%) (Ossenkoppele et al., Reference Ossenkoppele, Binette, Groot, Smith, Strandberg, Palmqvist, Stomrud, Tideman, Ohlsson, Jögi, Johnson, Sperling, Dore, Masters, Rowe, Visser, van Berckel, van der Flier, Baker and Hansson2022). Therefore, identifying Aβ + Tau + CN individuals could enrich clinical trials with persons at risk for rapid cognitive decline (Ossenkoppele et al., Reference Ossenkoppele, Binette, Groot, Smith, Strandberg, Palmqvist, Stomrud, Tideman, Ohlsson, Jögi, Johnson, Sperling, Dore, Masters, Rowe, Visser, van Berckel, van der Flier, Baker and Hansson2022). However, PET imaging techniques can hardly be used as a general screening tools due to their cost, invasiveness, and limited availability. Thus, cognitive tests are promising alternatives for this purpose (Parra et al., Reference Parra, Abrahams, Logie and Della Sala2010) as they might be cost-effective, non-invasive, rapid, and widely accessible. Despite this potential, the cognitive endpoints that best reflect AD pathology during the preclinical phase are still unknown.
Standard cognitive measures, including episodic memory tests such as the Free and Cued Selective Reminding Test (FCSRT, Grober et al., Reference Grober, Buschke, Crystal, Bang and Dresner1988), were designed to distinguish Mild Cognitively Impaired (MCI) or dementia patients from CN individuals (Lemos et al., Reference Lemos, Simões, Santiago and Santana2015). These metrics have proven effective in identifying MCI patients progressing to dementia (Grande et al., Reference Grande, Vanacore, Vetrano, Cova, Rizzuto, Mayer, Maggiore, Ghiretti, Cucumo, Mariani, Cappa and Pomati2018; Grober et al., Reference Grober, Wang, Kitner-Triolo, Lipton, Kawas and Resnick2022). However, standard cognitive tests lack sensitivity to detect subtle cognitive changes during the preclinical stage of AD (Johnson et al., Reference Johnson, Schultz, Betensky, Becker, Sepulcre, Rentz, Mormino, Chhatwal, Amariglio, Papp, Marshall, Albers, Mauro, Pepin, Alverio, Judge, Philiossaint, Shoup, Yokell and Sperling2016; Parra et al., Reference Parra, Calia, Pattan and Della Sala2022; Rentz et al., Reference Rentz, Parra Rodriguez, Amariglio, Stern, Sperling and Ferris2013). To address this limitation, cognitive composite scores that aggregate standard measures, such as the Preclinical Alzheimer Cognitive Composite (PACC, Donohue et al., Reference Donohue, Sperling, Salmon, Rentz, Raman, Thomas, Weiner and Aisen2014), have been proposed to increase the sensitivity of these measures to detect early preclinical cognitive decline. However, cross-sectional studies involving CN individuals are inconsistent, and those that evidenced differences between Aβ+ and Aβ- individuals often show only small effect sizes (AUC between 0.580 and 0.630) (Papp et al., Reference Papp, Rofael, Veroff, Donohue, Wang, Randolph, Grober, Brashear, Novak, Ernstrom, Raman, Aisen, Sperling, Romano and Henley2022). Furthermore, due to its multidomain properties, the PACC likely lacks the specificity to detect early tau aggregation, beyond the effects of amyloidosis at the preclinical stage (McKay et al., Reference McKay, Millar, Nicosia, Aschenbrenner, Gordon, Benzinger, Cruchaga, Schindler, Morris and Hassenstab2024). This issue is critical because tau pathology significantly impacts brain function and cognition (Hanseeuw et al., Reference Hanseeuw, Jacobs, Schultz, Buckley, Farrell, Guehl, Becker, Properzi, Sanchez, Quiroz, Vannini, Sepulcre, Yang, Chhatwal, Gatchel, Marshall, Amariglio, Papp, Rentz and Johnson2023; Ossenkoppele et al., Reference Ossenkoppele, Schonhaut, Schöll, Lockhart, Ayakta, Baker, O’Neil, Janabi, Lazaris, Cantwell, Vogel, Santos, Miller, Bettcher, Vossel, Kramer, Gorno-Tempini, Miller, Jagust and Rabinovici2016). Autopsies studied by Braak and Braak outlined the staging progression of tau pathology ranging from stage 0 to VI. Tau first accumulates in the transentorhinal cortex (tEC, comprising the medial portion of the perirhinal and anterolateral entorhinal cortices, alEC; Braak stage I) before extending to the MTL (Braak stage II) and subsequently to the rest of the temporal lobe (Braak stage III-IV) (Braak & Braak Reference Braak and Braak1991). This typical spatial spreading of tau pathology has been primarily confirmed through tau-PET studies (Pascoal et al., Reference Pascoal, Benedet, Tudorascu, Therriault, Mathotaarachchi, Savard, Lussier, Tissot, Chamoun, Kang, Stevenson, Massarweh, Guiot, Soucy, Gauthier and Rosa-Neto2021). There is a need to validate cognitive tasks that may engage the tEC (Bastin & Delhaye Reference Bastin and Delhaye2023) to identify who will progress to symptomatic AD more effectively. The tEC is involved in context-free memory functions, such as familiarity-based recognition (Parra et al., Reference Parra, Gazes, Habeck and Stern2024). It is also necessary to create, store, and retrieve unique representations of objects, allowing one to distinguish between very similar entities by integrating all their characteristics (Bastin & Delhaye Reference Bastin and Delhaye2023; Bussey & Saksida Reference Bussey and Saksida2002). This ability can be assessed by the Visual-Short Term Memory Binding Test (VSTMBT, Parra et al., Reference Parra, Abrahams, Logie and Della Sala2010), which evaluates conjunctive binding skills by requiring the retention of integrated features (e.g., shapes and colors) within object representation in short-term memory. This ability enables us to integrate the different features that we perceive from an object (color, shape, size, texture, smell) into a unique and unified memory representation, allowing us to later recognize and differentiate these complete entities from similar items (e.g., such as when we memorize the car of a friend by integrating its size, shape, color, license plate, and brand into a unified memory representation). The VSTMBT has been shown to distinguish MCI (Cecchini et al., Reference Cecchini, Foss, Tumas, Patrocinio, Chiari-Correia, Novaretti, Brozinga, Bahia, de Souza, Guimarães, Caramelli, Lima-Silva, Cassimiro, Brucki, Nitrini, Sala, Parra and Yassuda2020) and patients with subjective cognitive impairment (SCI) (Koppara et al., Reference Koppara, Frommann, Polcher, Parra, Maier, Jessen, Klockgether and Wagner2015; Valdés Hernández et al., Reference Valdés Hernández, Clark, Wang, Guazzo, Calia, Pattan, Starr, Sala and Parra2020) from CN individuals. Deficits in the VSTMBT were also observed in CN carriers of a presenilin 1 mutation leading to familial AD, compared with CN non-carriers (Parra et al., Reference Parra, Abrahams, Logie, Méndez, Lopera and Della Sala2010; Parra et al., Reference Parra, Calia, García, Olazarán-Rodríguez, Hernandez-Tamames, Alvarez-Linera, Della Sala and Fernandez Guinea2019). Furthermore, performance on this task correlated with the Aβ burden in this population (Norton et al., Reference Norton, Parra, Sperling, Baena, Guzman-Velez, Jin, Andrea, Khang, Schultz, Rentz, Pardilla-Delgado, K.Johnson, Reiman, Lopera and Quiroz2020) and in prodromal sporadic AD patients (Cecchini et al., Reference Cecchini, Yassuda, Squarzoni, Coutinho, de Paula Faria, de Souza Duran, da Costa, de Gobbi Porto, Nitrini, Forlenza, Brucki, Buchpiguel, Parra and Busatto2021). Moreover, Parra and colleagues (2024) found that CN participants with binding difficulties had higher Aβ deposits in the lateral-occipital cortex, fusiform gyrus, and EC than those with a low binding cost. In addition, VSTMBT scores were shown to correlate with tEC atrophy (Valdés Hernández et al., Reference Valdés Hernández, Clark, Wang, Guazzo, Calia, Pattan, Starr, Sala and Parra2020). While the VSTMBT appears promising for detecting cognitive correlates of early AD pathology, it has been chiefly used by the team that developed the test, and the sensitivity to identify individuals with preclinical AD individuals is not yet established. Moreover, the relationship between VTSMBT performance and EC tau burden (ECtau) has rarely been investigated in preclinical AD.
In the current study, we investigated VSTMBT performance in a sample of CN individuals with both Aβ and tau-PET imaging. We also included a group of patients with prodromal AD (Aβ+ MCI) as a positive control group assessing individuals further along in the AD spectrum. Our primary objectives were to evaluate the associations of VSTMBT performance with (1) the Aβ status in CN individuals and (2) with tau aggregation in the MTL (EC) and the ITC. Additionally, we compared the sensitivity and specificity of the VSTMBT and the PACC5 for identifying CN individuals with abnormal Aβ or tau burden.
Methods
Participants
One hundred and three individuals (aged 52–86) participated in this study, including 76 cognitively normal (CN) individuals and 27 patients with MCI (Mini-Mental State Examination (MMSE) ≥23 (Ruchinskas & Curyto Reference Ruchinskas and Curyto2003)). MCI patients were recruited at the Memory Clinic of the Cliniques Universitaires Saint-Luc in Brussels, Belgium. CN volunteers were enlisted via other clinical studies through mailbox announcements and advertisements in the hospital’s vicinity. Volunteers were selected from this pool to participate in the current study. Volunteer’ selection was enriched for carriers of the Apolipoprotein E ϵ4 allele (APOE ϵ4) to match the frequency of APOE ϵ4 carriership observed in patients. Recruitment and examinations were conducted between June 2019 and May 2024. Exclusion criteria were non-AD neurodegenerative pathologies, focal brain lesions, major depression or psychiatric diseases, and alcohol or drug abuse.
Informed consent was obtained from all participants, adhering to the principles of the Declaration of Helsinki. The Ethical Committee of UCLouvain approved the study (#UCL-2016-121, Date: 13/05/2019; Eudra-CT number: 2018-0034/73-94).
All participants underwent the VSTMBT (Parra et al., Reference Parra, Abrahams, Logie, Méndez, Lopera and Della Sala2010), a standard neuropsychological assessment to determine their cognitive status, a 3D-T1 brain Magnetic Resonance Imaging (MRI), a [18F]-MK-6240 Tau-PET scan, and an examination of the Aβ status either PET or cerebrospinal fluid (CSF).
Visual short-term memory binding test (VSTMBT)
Participants were assessed using a 16-inch laptop running an E-prime script (v.3.0, Psychological Software Tools, Pittsburgh, PA). Responses were collected using the E-prime Chronos device which records reaction times with millisecond accuracy.
The trial design for each condition of the VSTMBT is shown in Figure 1. The task was based on a change detection paradigm. Each trial began with a fixation cross that remained on the screen for 500 ms, followed by the study display presented for 2000ms. The study display presented two or three shapes the participant had to remember. After a 900 ms unfilled retention interval, the test display appeared, and participants had to determine whether the shapes presented in the test display were identical to those given in the study display, regardless of their position and orientation changes. Shapes were similar in 50% of the trials. Participants were instructed to press the green button on the E-prime Chronos device as fast as possible to indicate whether the shapes were identical and the red button if they were different.
Illustration of the conditions of the Visual-Short-Term Memory Binding Test (VSTMBT). The VSTMBT had four conditions: two with two items; and two with three items. In shape only conditions (SOC), participants had to determine whether the shapes presented on the test display were identical to the ones given earlier on the study display. In the shape color binding conditions (SCBC), participants had to judge whether the shapes and their respective colors on the test display were the same as on the study display. The orientation and position of the shapes on the screen were irrelevant in each condition. To respond, participants used the E-Prime Chronos device. If they judged the sets of shapes as identical, they were asked to press the green button and the red one otherwise.
There were four conditions, varying the number of shapes (two or three) and their color (white for the “Shape-Only condition” (SOC) or colored). The colored conditions required binding in memory of the shapes with their respective color (“Shape-Color Binding condition” (SCBC)), whereas the SOC assesses visual short-term memory for a single feature. The shapes presented within a trial were selected among a sample of eight non-verbalizable polygons combined with eight colors in the SCBC. The combination of shapes and colors was predefined and consistent across participants. No shape or color was repeated within a given array. All participants started with a two-item condition, followed by the same three-item condition. Half of the participants started with the SOC and the other half with the SCBC. Each condition included one practice trial followed by 16 test trials presented randomly. The accuracy score (AS) was calculated as the percentage of correct answers for each condition and the entire task (total AS). We also calculated the binding cost (Forno et al., Reference Forno, Parra, Thumala, Villagra, Cerda, Zitko, Ibañez, Lillo and Slachevsky2023), which offers insights into the cognitive resources required to maintain integrated information (SCBC) compared to those needed for storing individual features (SOC). The results about binding cost can be found in supplementary materials (S2, S3, S4).
A perceptual binding task was used as a screening test to rule out perceptual binding deficits of color and shape (see Parra et al., Reference Parra, Abrahams, Logie and Della Sala2010 for more details). The data for the perception condition are not presented, but all the participants in this study obtained a performance above the 80% cut-off (Forno et al., Reference Forno, Parra, Thumala, Villagra, Cerda, Zitko, Ibañez, Lillo and Slachevsky2023).
Due to difficulties in correctly distinguishing certain contrasts observed during a pre-test phase with healthy older individuals (n = 2), the color of some shapes and the background (changed from gray to black) were adapted from the original task. For a full description of the modified colors, see figure S1. The instructions were also translated in French, and the response modality was adjusted from a keyboard to the Eprime Chronos device to enable millisecond reaction times to be recorded. Despite these modifications, the performance of our sample is broadly comparable to that obtained in other studies (see for instance (Forno et al., Reference Forno, Parra, Thumala, Villagra, Cerda, Zitko, Ibañez, Lillo and Slachevsky2023; Parra et al., Reference Parra, Abrahams, Logie, Méndez, Lopera and Della Sala2010; Valdés Hernández et al., Reference Valdés Hernández, Clark, Wang, Guazzo, Calia, Pattan, Starr, Sala and Parra2020).
The VSTMBT was performed within a year relative to the tau-PET scan and the standard cognitive assessment.
Neuropsychological assessment
The neuropsychological testing evaluated four cognitive domains: (1) verbal episodic memory (FCSRT, French version (Van der Linden et al., Reference Van der Linden, Coyette, Poitrenaud, Kalafat, Calicis, Wyns and Adam2004)), (2) language (Lexis Naming Test, Category and Letter Fluency Test for animals and letter’ P,’ (de Partz de Courtray et al., Reference de Partz de Courtray, Bilocq, De Wilde, Seron and Pillon2001)), (3) executive functions (Trail Making Test part A and B (Reitan, Reference Reitan1955) and Luria’s Graphic Sequences (Weiner et al., Reference Weiner, Hynan, Rossetti and Falkowski2011)), and (4) visuospatial functions (Clock Drawing Test (Rouleau et al., Reference Rouleau, Salmon, Butters, Kennedy and McGuire1992) and Praxis part of the CERAD battery (Morris et al., Reference Morris, Mohs, Rogers, Fillenbaum and Heyman1988)). Z-scores were computed for each cognitive domain based on three measures within each domain and averaged to create a global cognitive Z-score (see (Ivanoiu et al., Reference Ivanoiu, Dricot, Gilis, Grandin, Lhommel, Quenon and Hanseeuw2015) for more information about cognitive testing). A cognitive domain was considered impaired if the Z-score fell below −1.5 standard deviations of the mean of an independent sample composed of 32 CN individuals who remained cognitively stable over an eight-year period. Patients were considered to have MCI if at least one z-score was below this cut-off and as being CN otherwise.
CN Volunteers also completed the Digit-Symbol Coding test (WAIS-IV, (Wechlser, 2011)) and the Logical Memory Test (WMS-III, (Wechsler, Reference Wechsler2001)), which are classically used to calculate the PACC5 (Papp et al., Reference Papp, Rentz, Orlovsky, Sperling and Mormino2017). In the current study, the PACC5 was computed as the average of five Z-scores calculated based on the MMSE (0-30) (Folstein et al., Reference Folstein, Folstein and McHugh1975), the Logical Memory Delayed Recall Story A (0-25) (WMS-III, (Wechsler, Reference Wechsler2001)), the Digit-Symbol Coding Test (0-135, (Wechlser, 2011)), the sum of free and total recall from the French-version of the FCSRT (0-96) (Van der Linden et al., Reference Van der Linden, Coyette, Poitrenaud, Kalafat, Calicis, Wyns and Adam2004) and categorical fluency (animals, 2 minutes, (de Partz de Courtray et al., Reference de Partz de Courtray, Bilocq, De Wilde, Seron and Pillon2001)). The Z-scores were calculated by referring to the performance obtained by all the CN individuals (n = 76). The PACC5 was not available for MCI patients.
MRI
We acquired three-dimensional (3D) T1-weighted and T2-weighted sequences for each participant at the Saint-Luc University Hospital (UCLouvain, Belgium) using a 3T head scanner (Signa™ Premier, General Electric Company, USA). The T1 MRI was segmented with Freesurfer (version 7.2), using the T2 sequence to improve the segmentation. MRI enabled us to determine EC and ITC volume and quantify WMH. WMH lesion load (mm3) was extracted from T1-weighted MRIs. WMH were identified by using spatial intensity gradients across tissue classes (Fischl et al., Reference Fischl, Salat, van der Kouwe, Makris, Ségonne, Quinn and Dale2004).
Tau [18F]-MK-6240 PET
[18F]-MK-6240 (Lantheus Inc.) is an investigational drug studied as a second-generation cerebral tau tangles imaging agent. Radiosynthesis was performed at KULeuven and shipped to our clinic. Ninety minutes after intravenous administration of [18F]-MK-6240 (target activity = 185 ± 5 MBq), a 30-minute dynamic acquisition was performed on a Philips Vereos digital PET-CT. Images were reconstructed using manufacturer’s standard reconstruction algorithm (including attenuation, scatter, decay correction, and time-of-flight information). Point spread function and 1 mm reslicing were also computed using the manufacturer’s algorithm to obtain a better resolution recovery.
For all participants, tau-PET images were co-registered with a 3D-T1 MRI using the PetSurfer pipeline, a set of tools within FreeSurfer for end-to-end integrated MRI-PET analysis (Greve et al., Reference Greve, Salat, Bowen, Izquierdo-Garcia, Schultz, Catana, Becker, Svarer, Knudsen, Sperling and Johnson2016). Standardized Uptake Value ratio (SUVr) values were extracted for all regions from the Desikan-Killiany Atlas (Desikan et al., Reference Desikan, Ségonne, Fischl, Quinn, Dickerson, Blacker, Buckner, Dale, Maguire, Hyman, Albert and Killiany2006) using cerebellum gray matter as a reference region. The Braak 5 region was calculated following previously defined regions-of-interest (Schöll et al., Reference Schöll, Lockhart, Schonhaut, O’Neil, Janabi, Ossenkoppele, Baker, Vogel, Faria, Schwimmer, Rabinovici and Jagust2016). The interest of including a Braak 5 ROI was to show that our participants did not have extensive tauopathy. Our regions of interest were the EC and the ITC. We selected the EC because our hypothesis was to test whether VSTMBT performance could reflect early tauopathy. The ITC was selected because it is a region adjacent to the EC that is affected later in the course of the disease. By including this region in our models, we were able to demonstrate the specificity of the EC on VSTMBT performance.
Individuals were classified as Tau+ when the Braak stage was rated superior to 0 on visual read by the nuclear medicine physician.
Amyloid status
The Aβ status was determined either by lumbar puncture (n = 17) or Aβ PET-scan with [18F]-Flutemetamol (VizamylTM, GE Healthcare, n = 68) or Pittsburgh compound B (PIB) (n = 18). The Aβ-PET burden was calculated and expressed in the Centiloid scale (Hanseeuw et al., Reference Hanseeuw, Malotaux, Dricot, Quenon, Sznajer, Cerman, Woodard, Buckley, Farrar, Ivanoiu and Lhommel2021; Klunk et al., Reference Klunk, Koeppe, Price, Benzinger, Devous, Jagust, Johnson, Mathis, Minhas, Pontecorvo, Rowe, Skovronsky and Mintun2015). In CSF, measurements of Aβ42 were conducted using Lumipulse automated assays. Participants were considered amyloid positive (Aβ+) if at least one of the following criteria was met: Centiloid >20 (Amadoru et al., Reference Amadoru, Doré, McLean, Hinton, Shepherd, Halliday, Leyton, Yates, Hodges, Masters, Villemagne and Rowe2020) or CSF Aβ42 < 437pg/ml (Bayart et al., Reference Bayart, Hanseeuw, Ivanoiu and van Pesch2019). Fourteen participants with borderline Aβ42 CSF levels (between 437pg/ml and 650pg/ml) also underwent PIB-PET, which proved positive in all these participants.
Clinical classification
Based on the neuropsychological examination and the Aβ status, three clinical groups were defined as follows: (1) amyloid-negative cognitively normal individuals (Aβ − CN, n = 54); (2) amyloid-positive cognitively normal individuals (Aβ + CN, considered as having preclinical AD, n = 22); (3) amyloid-positive MCI participants (Aβ + MCI, considered as having prodromal AD, n = 18). As this research focused on AD early diagnosis, Aβ − MCI participants (n = 9) were excluded from the analyses.
Among the 94 participants, one Aβ + MCI participant had no available tau-PET data because he moved during the acquisition. Moreover, another Aβ + MCI individual only had data for the SOC because of color perception deficits.
Statistical analyses
The accuracy score (AS) for the SOC was calculated by averaging performance on the two-items SOC and the three-items SOC. Similarly, the AS score for the SCBC was computed by averaging performance on the two and three-items SCBC. The total AS score was obtained by averaging these four conditions.
All analyses were computed using SPSS 28.0.1.1 Statistics for Windows (Armonk, NY: IBM Corp.) with two-tailed p-values reported. Data were tested for normality using the Shapiro-Wilk test (data not shown).
To investigate whether VSTMBT conditions could distinguish Aβ+CN from Aβ−CN individuals, we first compared VSTMBT performance across clinical groups using parametric or non-parametric ANOVA’s, with post hoc multiple comparisons tests adjusted using Bonferroni correction. Second, we performed multiple linear regression models to test whether ECtau or ITCtau explains VSTMBT performance in the entire sample, and specifically with CN participants (Aβ − CN and Aβ + CN). All regression models were adjusted for age, sex, and education and time between VSTMBT administration and tau-PET. We also controlled for cognitive status to ensure that MCI patients did not drive the analysis. Additionally, amyloid burden (as measured by PET and expressed in Centiloids) and white matter hypointensities (WMH) were introduced as covariates in separate models. Participants who only underwent a lumbar puncture were therefore excluded from this analysis (n = 10). Third, mixed-effect models were used to determine whether the effect of regional tauopathy on VSTMBT performance varied by conditions. Linear mixed-effects regression was conducted with the clinical group as the between-subject factor and condition as the within-subject factor (SOC vs. SCBC and two items vs. three items) to evaluate if the associations between regional tau burden and performance varied according to the groups and conditions. Finally, we computed ROC curve analyses to compare the sensitivity and the specificity of the VSTMBT conditions, the FCSRT, and the PACC5 to discriminate Aβ+CN from Aβ-CN and Tau+CN from Tau-CN.
Results
Characteristics of participants
The 94 participants included 54 Aβ − CN (57%), 22 Aβ + CN (23%), and 18 Aβ + MCI (19%). Groups did not significantly differ in terms of age (p = .074, η2 = .054), education (p = .086, η2 = .052), sex (p = .919, V = .068) or percentage of APOE ϵ4 carriers (p = .140, see Table 1). Regarding cognitive measures, Aβ + CN had similar MMSE and global cognitive composite scores to those of Aβ − CN (p = .223, η2 = .04 and p = .122, η2 = .04, respectively), but Aβ + CN performed significantly worse than Aβ-CN on the PACC5 (p = .020, η2 = .082). As expected, Aβ + MCI performed substantially worse on all cognitive measures than both CN groups (all p < .05). Aβ+ CN had a higher tau burden than Aβ − CN in the EC (p < .001, η2 = .19), but not in the ITC (p = .237, η2 = .04). Aβ + MCI had higher ECtau and ITCtau (both p < .001, η2 = .33 and η2 = .24, respectively), compared to Aβ-CN and higher ITCtau than Aβ+CN (p = .023, η2 = .15), but non-significantly higher ECtau (p = .169, η2 = .08). All groups had less tau-pet signal in Braak 5 regions than in EC and ITC.
Demographic characteristics and biomarkers values
SD = Standard deviation; CN = cognitively normal; MCI = mild cognitive impairment; Aβ = amyloid-β; CSF = cerebrospinal fluid; SUVr = Standard Uptake Value ratio; * significantly different from Aβ-CN; $ significantly different from Aβ + CN. Seven MCI patients and six CN individuals with borderline amyloid CSF results underwent Aβ-PET. In these cases, the Aβ classification was based on PET data.
Group differences in visual short-term memory binding performances
AS was comparable across the groups in the shape-only condition with two items (two-items SOC) (p = .219). In contrast, there was a group difference in the AS when three items were used (three-items SOC) or in the shape-color binding conditions (SCBC) with either two or three items (all p’s < .001). Post-Hoc tests revealed that Aβ+CN participants performed worse than Aβ-CN individuals in the three-items SOC (p = .028), in the two-items SCBC (p = .001), and in the three-items SCBC (p = .038) conditions.
When averaging performance across the two- and three-items conditions, Aβ+CN individuals did not significantly differ from Aβ − CN participants in the SOC (p = .114), but they did in the SCBC (p < .001; see Figure 2). Aβ+MCI had a lower performance than Aβ-CN in both SOC (p = .001) and SCBC (p< .001). Averaging AS across all four conditions revealed a significant difference in total AS by Aβ status in CN (p< 0.001). Reaction times did not distinguish groups in any condition. Moreover, there was no effect on the order of the conditions presented to the participants (SOC versus SCBC) (p = .822).
Boxplots of Visual-Short-Term Memory Binding Test performances in each group. SOC = shape-only condition; SCBC = shape-color binding condition; CN = cognitively normal; MCI = mild cognitive impairment; Aβ = amyloid-β. Each participant is therefore represented twice on this graph: once for his performance in the SOC, and once for his performance in the SCBC. Aβ + CN did not differ from Aβ − CN participants in SOC (p = .114), but they did in SCBC (p < .001) and in total score (p < .001), which is the average between the SOC and SCBC scores.
Association between performance and demographics
The total AS was significantly predicted by age, with older age associated with lower performance (β= −.297, SE = .097, partial η2 = .094, p = .003) but not by education nor sex (β= .271, SE = .241, partial η2 = .014, p = .265 and β = 1.498, SE = 1.615, partial η2 = .010, p = .356 respectively). An interaction was found between condition (SOC versus SCBC) and age (F(1,91) = 5.917, partial η2 = .061, p = .017). Specifically, age was related to AS in SCBC (β= −.454, SE = .131, partial η2 = .117, p < .001) but not in SOC (β= −.154, SE = .095, partial η2 = .028, p = .106).
When only considering CN participants, the association between age and performance was not significantly different across conditions (F(1,74) = 1.717, partial η2 = .023, p = .194).
Association between total accuracy score and regional tau burden
The total AS was significantly associated with ECtau and ITCtau across the entire sample (β= −4.028, SE = 1.082, partial η2 = .137, p < .001 and β = −3.366, SE = 1.236, partial η2 = .079, p = .008, respectively). The delay between VSTMBT administration and tau-PET has no effect on performance (β= .001, SE = .001, partial η2 = .001, p = .741). An interaction effect was observed between ECtau and condition (F(1,87) = 12.29, partial η2 = .124, p < .001, see Figure 3), indicating that the association between the ECtau and the AS was more robust in the SCBC (β = −6.594, SE = 1.457, partial η2 = .191, p < .001) than in the SOC (β= −1.497, SE = 1.121, partial η2 = .020, p = .185). There was no interaction between ECtau and the number of items (F(1,86) = .005, partial η2 = .000, p = .946) nor between ECtau, the condition (SOC versus SCBC), and the number of items (F(1,87) = 1.021, partial η2 = .012, p = .315). Finally, no interaction was found between ITCtau and any conditions (all p’s > .05).
Association between entorhinal tau PET signal and accuracy score in each condition of the Visual-Short-Term Memory Binding Test. SOC = shape-only condition (dotted line); SCBC = shape-color binding condition (plain line); CN = cognitively normal; MCI = mild cognitive impairment; Aβ = amyloid-β. Each participant is represented twice on this graph: once for SOC, and once for SCBC performance. This graph highlights a stronger relationship between AS and entorhinal tau PET signal in the SCBC (plain line) than in the SOC (dotted line).
Association between binding performance and regional tau burden
We then focused on binding performance, as this condition was more strongly associated with ECtau than the non-binding SOC subtask.
Over the entire sample, the SCBC AS was associated with ECtau and ITCtau (β= −6.594, SE = 1.457, partial η2 = .191, p < .001 and β= −4.313, SE = 1.731, partial η2 = .067, p = .015 respectively). While ECtau remained a significant predictor after adjusting for cognitive status, ITCtau did not (Table 2, model 1), suggesting that MCI participants drove the effect of ITCtau on binding performance. In contrast, there was no interaction between ECtau and clinical group (F(2,92) = 1.464, partial η2 = .034, p = .237), indicating that a specific group did not drive the association between tau and binding performance.
Multiple regression models predicting the accuracy score in the shape-color binding condition and in shape-only condition
Models 3 and 4 only present the results in CN participants (Aβ-CN and Aβ+CN). Models 2 and 4 are adjusted for amyloid-PET. In CN individuals, the ECtau remains associated with binding (SCBC score) even after adjusting for amyloid burden. The ECtau and ITCtau did not contribute to explaining performance in SOC.
When analyzing CN only, ECtau was associated with SCBC AS (β= −10.561, SE = 2.54, partial η2 = .198, p < .001), while the ITCtau was not (β= −9.542, SE = 5.199, partial η2 = .041, p = .071; Table 2, model 3).
Association between binding performance and entorhinal tau after adjusting for amyloid load and white matter hypointensities
ECtau, amyloid PET signal and WMH were independently associated with SCBC AS when included in a model predicting binding performance in the entire sample (Table 2, model 2). In this model, the ECtau explained the highest percentage of variance. Similar results were observed when restricting the analysis to CN participants (Table 2, model 4). In this subgroup, ECtau only explained more than two times as much variance of binding performance (17%) compared to amyloid or WMH (7.8% and 8.2%, respectively). Similar results were obtained when considering two- or three-items SCBC, with slightly smaller effect sizes for ECtau in the three items SCBC, when considering the entire sample (β= −7.089, SE = 2.612, partial η2 = .095, p = .008 and β= −8.377, SE = 5.129, partial η2 = .050, p = .059, respectively) or CN only (β= −3.396, SE = 4.479, partial η2 = .116, p = .005 and β= −3.10, SE = 5.081, partial η2 = .107, p = .007, respectively).
ECtau burden remains significant after adjusting for the effect of ITCtau in the entire sample and in CN participants (β= −8.857, SE = 2.778, η2 = .125, p = .002 and β= −11.022, SE = 2.950, η2 = .181, p < .001, respectively).
Association between non-binding performance and regional tau burden
Neither ECtau nor ITCtau was associated with SOC performance in either model (Table 2, right panel). Non-binding (SOC) performance was thus only influenced by the cognitive status in Model 1 and age in Model 3 in CN participants.
ROC curves
We finally aimed to evaluate which measure among the FCSRT, the PACC5, and VSTMBT performance best distinguished Aβ+MCI from CN (Aβ + and Aβ−), Aβ + CN from Aβ− CN, and Tau + CN (n = 12) from Tau − CN (n = 64). As a reminder, the PACC5 was not available for MCI participants. For distinguishing Aβ+MCI individuals from CN participants (both Aβ+ and Aβ−), the FCSRT (AUC = 0.947, p < .001, best threshold = −1.28 Z-score, Sensitivity (Sn) = .76, Specificity (Sp) = .98) showed the greatest AUC, surpassing the SCBC (AUC = 0.772, p< 0.001, best threshold = 82.8%, Sn = .94, Sp = .49), which is consistent with the use of this test to highlight episodic memory impairment, the more frequent deficit observed in symptomatic AD patients.
To distinguish Aβ + CN from Aβ − CN, the highest AUC was found for the AS in the VSTMBT two-items SCBC (AUC = 0.773, p < .001, best threshold = 90.5%, Sn = .77, Sp = .65; Figure 4a). In contrast, the PACC5, with a threshold of -0.155 Z-score, had a lower AUC (AUC = 0.697, p = .007, Sn = .72, Sp = .68).
ROC curves comparing the different tests in cognitively normal individuals. CN = cognitively normal; MCI = mild cognitive impairment; Aβ = amyloid-β; AS = accuracy score; SOC = shape only condition; SCBC = shape-color binding condition; PACC5 = Preclinical Alzheimer Cognitive Composite; AUC = area under the curve; Sn = sensitivity; Sp = specificity. ROC curves data table evaluating the cognitive metrics to distinguish Aβ − CN vs Aβ + CN (left side) and tau − CN vs tau + CN (right side). The presence of amyloid is established based on the CL >20 threshold, and the presence of tau is established by a Braak stage >0. The best AUC curve for each comparison is in bold.
To distinguish CN with a positive tau-PET visual read (Braak >0) from CN with a negative tau-PET (Braak 0), the AS in the two-items SCBC was the most sensitive measure (AUC = 0.841, p = <.001, best threshold = 80%, Sn = .92, Sp = .61). In contrast, all other metrics, including PACC5 had lower AUC (AUC = .654, p = .020, best threshold = −.33 Z-score, Sn = .61, Sp = .73; Figure 4b).
Discussion
Detecting early changes in cognitive performances in individuals with preclinical AD is essential for early diagnosis, prevention and clinical trials. The present study investigated whether VSTMBT could identify early AD, including Aβ pathology and ECtau. We found that Aβ + CN individuals had lower VSTMBT SCBC performance than Aβ− CN and that binding performance was associated with ECtau. Moreover, binding performance distinguished CN individuals with Aβ or tau better than the PACC5.
Our observations are consistent with previous studies demonstrating reduced binding performance in individuals with autosomal-dominant AD mutations (E280A-PSEN1) (Parra et al., Reference Parra, Abrahams, Logie and Della Sala2010; Reference Parra, Calia, García, Olazarán-Rodríguez, Hernandez-Tamames, Alvarez-Linera, Della Sala and Fernandez Guinea2019), and in sporadic SCI and MCI patients (Cecchini et al., Reference Cecchini, Foss, Tumas, Patrocinio, Chiari-Correia, Novaretti, Brozinga, Bahia, de Souza, Guimarães, Caramelli, Lima-Silva, Cassimiro, Brucki, Nitrini, Sala, Parra and Yassuda2020, Reference Cecchini, Yassuda, Squarzoni, Coutinho, de Paula Faria, de Souza Duran, da Costa, de Gobbi Porto, Nitrini, Forlenza, Brucki, Buchpiguel, Parra and Busatto2021, Reference Cecchini, Parra, Brazzelli, Logie and Della Sala2023; Koppara et al., Reference Koppara, Frommann, Polcher, Parra, Maier, Jessen, Klockgether and Wagner2015; Valdés Hernández et al., Reference Valdés Hernández, Clark, Wang, Guazzo, Calia, Pattan, Starr, Sala and Parra2020). Thanks to our large dataset of well-characterized CN individuals, our study adds to these previous findings by showing a decline in SCBC performance during the preclinical stage of sporadic AD, where most cognitive tests remain in the normal range. Our study highlights the superiority of SCBC over SOC (i.e., binding of multiple features over single-feature memory) to distinguish preclinical AD from Aβ-CN. In line with Koppara and colleagues’ findings in SCI participants (2015), the SOC condition lacks the sensitivity to differentiate preclinical AD from CN individuals. As our results showed that this condition is less sensitive than the SCBC alone, it seems clinically unnecessary to administer the SOC. This condition mainly served in this work as a comparison for SCBC and to calculate the binding cost, which did not demonstrate increased sensitivity compared to the SCBC alone (for more details see supplementary S2, S3, S4).
Second, we observed that the ECtau was associated with VSTMBT SCBC performance, but not in the SOC. Only one study has previously investigated the link between VSTMBT performance and regional tau deposits in familial AD (Norton et al., Reference Norton, Parra, Sperling, Baena, Guzman-Velez, Jin, Andrea, Khang, Schultz, Rentz, Pardilla-Delgado, K.Johnson, Reiman, Lopera and Quiroz2020). Unlike our findings, their results indicated a stronger association between ECtau and performance in SOC than in SCBC. This discrepancy could be due to differences in the studied populations, the task conditions, and the tau-PET radiotracer used. Norton and colleagues included CN participants who were significantly younger, less educated, and had familial AD, which is more frequently associated with an atypical distribution of tau pathology (with more extensive tau burden in the parietal lobe and relatively lower tau burden in the MTL (Gordon et al., Reference Gordon, Blazey, Christensen, Dincer, Flores, Keefe, Chen, Su, McDade, Wang, Li, Hassenstab, Aschenbrenner, Hornbeck, Jack, Ances, Berman, Brosch, Galasko and Gauthier2019)). Most importantly, their subjects showed comparable tauopathy between the EC and ITC, suggesting that their participants already had more advanced tauopathy than our sample. This argument, combined with the fact that they only used the most challenging condition (3 items), leads us to believe that the SCBC was already too complex, and that SOC performance led to fewer ceiling effects than in our study. Finally, Norton and colleagues used [18F]-Flortaucipir whereas we used [18F]-MK-6240 as a tau-PET radiotracer, which could explain the differences in our results as [18F]-MK-6240 has a more extensive dynamic range in SUVr, improving early ECtau detection (Gogola et al., Reference Gogola, Minhas, Villemagne, Cohen, Mountz, Pascoal, Laymon, Mason, Ikonomovic, Mathis, Snitz, Lopez, Klunk and Lopresti2022).
ECtau better explained binding performance than Aβ, WMH, or age. Previous studies have shown that participants with more significant difficulties in the SCBC compared to the SOC (“high binding cost”) were more likely to have Aβ pathology (Parra et al., Reference Parra, Gazes, Habeck and Stern2024) and/or WMH (Parra et al., Reference Parra, Saarimäki, Bastin, Londoño, Pettit, Lopera, Della Sala and Abrahams2015). This finding suggests that Aβ deposits and/or WMH might impair binding capacities, although their study did not test. Our analyses revealed that ECtau accounted for more variance in SCBC than Aβ or WMH. Moreover, ECtau remained significant after adjusting for ITCtau, suggesting that SCBC is specifically sensitive to the ECtau and not to more widespread tauopathy, making it possible to diagnose the disease at the preclinical stage as soon as tauopathy appears in this region.
Consistent with our correlational analyses, we observed that the SCBC (and especially the two-items SCBC) had the highest AUC for distinguishing Tau + CN from Tau − CN, using the tau-PET visual read to define tau positivity (Braak stage >0). This result is noteworthy as Aβ + Tau + CN individuals are six times more likely to develop cognitive impairment within five years than Aβ + Tau- individuals (Ossenkoppele et al., Reference Ossenkoppele, Binette, Groot, Smith, Strandberg, Palmqvist, Stomrud, Tideman, Ohlsson, Jögi, Johnson, Sperling, Dore, Masters, Rowe, Visser, van Berckel, van der Flier, Baker and Hansson2022). In addition, VSTMBT SCBC, and especially the two-items condition had a higher AUC and sensitivity than PACC5 to distinguish Aβ+ from Aβ− CN individuals. Of note, the PACC5 was designed initially to separate Aβ− from Aβ+ CN individuals but requires a more extended administration time than the VSTMBT SCBC (30 vs. less than 5 minutes). Moreover, the VSTMBT is more straightforward to implement remotely than the PACC5, as recently demonstrated in a VSTMBT self-administration study (Butler et al., Reference Butler, Watermeyer, Matterson, Harper and Parra-Rodriguez2024).
Limitations
Although the shapes were selected to be infrequent and non-verbalizable polygons (to minimize the impact of verbal working memory on visual working memory), some participants reported relying on verbalization by assigning a label to specific shapes. Verbalization has been shown to facilitate memory (Nedergaard et al., Reference Nedergaard, Wallentin and Lupyan2023) as well as problem-solving ability (Baldo et al., Reference Baldo, Dronkers, Wilkins, Ludy, Raskin and Kim2005). We did not control for the strategies used, which may have led to overestimating participants’ performance using specific strategies.
Second, Aβ-MCI participants were excluded due to the small sample size. Thus, this study did not determine whether SCBC low performance is specific to AD. However, previous research has already demonstrated that performance in a color-object binding task, relatively similar to the VSTMBT, was not affected in other pathologies such as frontotemporal dementia, vascular dementia, Lewy body dementia and dementia associated with Parkinson’s disease (Della Sala et al., Reference Della Sala, Parra, Fabi, Luzzi and Abrahams2012; Kozlova et al., Reference Kozlova, Parra, Titova, Gantman and Sala2021).
Third, our sample is not representative of the general population, as it mainly consisted of white individuals with a high level of education and APOE ϵ4 carriers were overrepresented compared to the general population. Indeed, in the general population, only 23% of individuals carry at least one ϵ4 allele (Régy et al., Reference Régy, Dugravot, Sabia, Helmer, Tzourio, Hanseeuw, Singh-Manoux and Dumurgier2024)(versus 46 to 72% in our sample, see table 1). Further research should be conducted in more diverse populations.
Fourth, the colors currently used in this test make it unsuitable for colorblind patients. However, the task could easily be adapted with a color palette tailored to those with color vision deficiencies.
Fifth, participants were asked to respond by pressing a green or red button to record their answers and response time. Still, this procedure led to some errors (with some participants spontaneously reporting they aimed to press the other button during the task). We could not take the oral responses into account. However, we do not believe this limitation significantly impacts our findings, as it comprised less than 2% of the responses. Moreover, it has been shown that there was no difference in performance in this task compared to the same task with oral responses (Cecchini et al., Reference Cecchini, Parra, Brazzelli, Logie and Della Sala2023).
Finally, despite this task appearing robust across various administration procedures (for instance self-administration (Butler et al., Reference Butler, Watermeyer, Matterson, Harper and Parra-Rodriguez2024), flash cards (Della Sala et al., Reference Della Sala, Kozlova, Stamate and Parra2018) or computer tablets (Weir et al., Reference Weir, Paterson, Tieges, MacLullich, Parra-Rodriguez, Della Sala and Logie2014)) we still lack standardized cut-offs enabling to implement this task in clinical practice. Our study suggests the use of a 90.5% threshold for the two-items SCBC to detect both amyloid and tau pathologies in a CN population. Moreover, all the studies that have assessed the value of this task so far, including the current one, are cross-sectional. Future work should investigate how performance evolves over time, in association with biomarkers, and test whether it is predictive of future patient outcomes.
Perspectives
While the VSTMBT appears promising in identifying individuals at the preclinical stage of AD, further studies are needed to support the validation of this task as a screening tool (e.g., replication within larger preclinical samples, determining the specificity of this task to screen for a potential underlying AD pathology, standardization of the task design and administration procedures, assessing the test-retest reliability, development of normative data).
Moreover, longitudinal studies should be carried out to determine whether VSTMBT can predict clinical conversion and how VSTMBT performances evolve with disease progression.
Conclusion
In conclusion, our analyses show that conjunctive binding abilities, enabling us to integrate all the features of objects into unified representations, seems sensitive to early tau burden in preclinical AD. Assessing this ability seems therefore promising for large-scale early diagnosis of the disease.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1355617725000165.
Acknowledgements
We would like to extend our warmest thanks to the Belgian Alzheimer’s research foundation for its support.
The firm Lantheus Inc. supplied the [18F]MK6240 precursor for acquiring the PET images analyzed in this article. No other conflicts of interest are reported.
We also thank Jean Ferrier for contributing to data collection as part of his master’s degree.
Funding statement
L.H. was funded by the Belgian Fund for Scientific Research (FNRS), grant number FNRS40016560. B.H. was funded by the FNRS, grant number CCL40010417, the FRFS-WELBIO, grant number 40010035, and the SAO grand number 2022/0026.
Open access
