Cognitive Challenge Tests such as the Loewenstein-Acevedo Scales for Semantic Interference and Learning (LASSI-L) present participants with two 15-item word lists that share semantically related targets. Identical semantic cues used during both encoding and retrieval trigger proactive semantic interference (PSI) effects. PSI occurs on the LASSI-L while trying to recall items from the second list. It represents a failure to learn new information in the face of recently learned competing information (1). It has been posited that PSI or failure to recover from PSI (fPSI), despite repeated presentations of the semantically competing word list, reflects a cognitive impairment that goes beyond a mere retrieval issue. This includes problems with source memory, inhibition, and, in the case of intrusion errors, the failure to self-monitor. These processes are thought to involve disruptions between medial temporal and prefrontal circuits (1).

PSI using the LASSI-L has been useful to detect prodromal AD (2) and has been associated with degeneration in AD prone regions (3, 4), which are predictive of disease progression (5). PSI has also been identified as an early cognitive deficit associated in other neurodegenerative conditions such as pre-manifest Huntington's disease (6, 7) and multiple sclerosis (8). Deficits in frPSI have been consistently observed in individuals with amnestic mild cognitive impairment (MCI) across diverse, multicultural cohorts in the United States and internationally (1, 4, 9–12). SIEs have consistently been observed in persons with underlying Alzheimer's disease (AD) pathology (1, 5, 12) and have shown a great deal of specificity to AD-related cognitive impairment. These errors primarily involve incorrectly recalling target items during PSI and frPSI trials of the LASSI-L (13). Individuals with amnestic MCI (aMCI) suspected to have an AD etiology, based on positive amyloid beta (Aβ) positron emission tomography (PET), demonstrate significantly more SIEs than aMCI older adults who are Aβ PET negative, despite both groups showing a clinically progressive disease course (5, 14, 15).

Blood tests for diagnosis of AD have been validated and found to have accuracy comparable to amyloid PET and cerebrospinal fluid (CSF) tests. This advancement offers exciting possibilities for larger-scale studies of AD-related cognitive phenotyping (16). One of the most promising blood-based biomarkers, plasma p-tau₂₁₇, consistently demonstrates high sensitivity and specificity for Aβ PET positivity and tau PET results (17, 18). However, without AD-specific and culturally validated cognitive assessments, blood-based biomarkers may have limited applicability in primary care or community health settings.

Given the high concordance between plasma p-tau₂₁₇ with neuropathological, CSF, and Aβ and tau PET biomarkers of AD, the objective of this study was to investigate the relationship between p-tau₂₁₇ and AD-related cognitive markers such as frPSI and SIEs on the LASSI-L. A limitation of previous studies using the LASSI-L has been the failure to control for maximum learning capacity which may influence the occurrence of frPSI and SIEs. In this study, we addressed this previous limitation by using maximum learning capacity on the first target list of the LASSI-L. It was hypothesized that individuals with aMCI and higher p-tau₂₁₇ levels, would exhibit an increased number of SIEs and a higher percentage of SIE impairments, based on established cut-off values.

As depicted in Table 1, participants classified as aMCI p-tau₂₁₇+ did not differ from those who were aMCI p-tau₂₁₇- with regards to years education attained, sex, or Hispanic/Latino ethnicity. Participants who were aMCI p-tau₂₁₇+ were older [F (1,107) = 4.10; p = 0.045], had lower MMSE scores [F (1,107) = 6.98; p = 0.009], and a higher percentage of Aβ positive PET scans (87 vs. 13%) [X² (df = 1) = 44.01; p < 0.001]. In addition, those who were p-tau₂₁₇+ were more often carriers of the ApoE ε4 allele (54.2 vs. 32.0%) [X² (df = 1) = 4.57; p = 0.03]. Factors such as being Hispanic/Latino or language of evaluation had no effect on obtained results.

As it relates to LASSI-L performance, participants who were aMCI p-tau₂₁₇+ had lower initial learning scores (Cued A2) [F (1,107) = 8.71; p = 0.004] and lower recall on the trial that measures frPSI (Cued B2 recall) [F (1,107) = 11.11; p = 0.001]. Participants who were aMCI p-tau₂₁₇+ also had more SIEs on trials that measure PSI and frPSI (Cued B1 intrusions) [F (1,107) = 11.92; p < 0.001, d = 0.67]; and Cued B2 intrusions [F (1,107) = 17.10; p < 0.001, d = 0.82]. Cohen's d was calculated for the LASSI-L Cued B2 recall (d = 0.64); Cued B1 intrusions (d = 0.67) and Cued B2 intrusions (d = 0.82) and yielded moderate to large effect sizes.

Results were generally consistent after statistical adjustments for age, MMSE, and maximum learning scores (Cued A2) using a series of ANCOVA models evaluating SIEs on Cued B1 and Cued B2 (Table 2). Even after this covariate adjustment, aMCI p-tau₂₁₇+ participants continued to evidence more SIEs on Cued B2 (p = 0.006) and Cued B1 (p = 0.035), albeit to a lesser extent. It should be noted in Table 2, mean values adjusted for covariates with their standard errors of measurement are depicted. While ANOVA and ANCOVA models are mathematically equivalent and specialized cases of linear regression, we conducted separate regression analyses that focused on biological predictors of SIEs independent of diagnostic status. Since sex as a biological variable was not found to be related to SIEs, we only included the three biological variables that were available to us (age, p-tau₂₁₇ positivity, and ApoE4 positivity) as predictors of SIEs in the entire sample. When all variables were entered simultaneously in a linear regression model, each was statistically significant (Table 3). We also employed a stepwise regression approach which ordered predictors in the model resulting in a Total R of 0.489 (R² = 0.240; Table 4).

Individuals with clinically diagnosed MCI and underlying Alzheimer's disease pathology exhibit more semantic intrusion errors (SIEs) than those with MCI of similar severity due to other etiologies, such as multiple cerebral infarctions, neurological disorders, or psychiatric conditions (40, 41). SIEs and false recognition among aMCI patients may be related to AD pathology, as measured by Aβ deposition on PET or low CSF levels of Aβ (1-42) in aMCI (42, 43). Thomas et al. (44) found that intrusion errors on the Rey Auditory Verbal Learning Test at baseline could predict progression to MCI among cognitively unimpaired individuals studied longitudinally in the Alzheimer's Disease Neuroimaging Initiative (ADNI). Similarly, Crocco et al. (31) found that SIEs on the LASSI-L could predict those persons with aMCI who would progress to dementia. Loewenstein et al. (10) and Kitaigorodsky et al. (5) found that SIEs on a cognitive challenge test that measured proactive semantic interference (PSI) and the failure to recover from proactive semantic interference (frPSI) could differentiate between older adults with aMCI who were Aβ PET positive, from persons diagnosed with aMCI who were Aβ PET negative, as well as cognitively unimpaired individuals who were Aβ PET negative.

One of the unique aspects of the LASSI-L assessment paradigm is that it was designed to elicit semantic intrusion errors because the test is structured to maximize learning of two semantically competing word lists. While detection of SIEs is partially possible on other list learning test such as the California Verbal Learning Test, these tests may have a few items that overlap semantically between lists, but do not use a 1:1 semantic competing list, thus, they will fail to capture SIEs. Methods have been developed by Thomas et al. (44) that use neuropsychological process scores that have related word list intrusion errors in the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset where she found that these errors improved earlier identification of persons with subtle cognitive decline.

Longitudinal studies suggest that SIES may have predictive utility for determining transitions from PreMCI to MCI, and MCI to dementia as well as longitudinal changes in CDR sum of boxes. Zheng et al. (45) investigated the predictors of SIEs in a longitudinal path analysis involving 216 participants from the 1Florida ADRC. This analysis accounted for covariates such as sex, education, language of evaluation, and overall cognitive impairment. The results revealed that both Aβ PET positivity and reduced brain volumes in regions vulnerable to AD were directly associated with SIEs. Additionally, the APOE ε4 status influenced SIEs indirectly by affecting amyloid positivity. Age also had an indirect effect on SIEs through its relationship with Aβ positivity and brain volumes. Notably, global mental status did not significantly impact these findings. In a study conducted by Crocco et al. (31) found that SIEs are predictive of progression of cognitive impairment in persons diagnosed with aMCI or PreMCI (not cognitively normal but not meeting full criteria for MCI). Additional longitudinal studies are ongoing to evaluate whether elevated SIE rates predict future progression to biologically confirmed AD/ADRD.

This study is among the first to link an AD-relevant cognitive marker (SIEs on LASSI-L's PSI and frPSI measures) to the promising plasma-based AD pathology marker, p-tau₂₁₇. This association persisted after adjusting for initial learning capacity. It is likely that these deficits likely extend beyond source memory and involve deficits in self-monitoring (1). Semantic inhibition and self-monitoring difficulties may significantly contribute to semantic intrusions. Chasles et al. (46) found that impaired deep semantic processing deficits, beyond source memory and general inhibition, exacerbated semantic interference on the LASSI-L compared to a phonemic processing test.

Previous studies using path analytic models have elegantly shown that among those with both Aβ PET positivity and neurodegeneration in AD prone regions, ApoE ε4 genotype has an indirect effect on SIEs through Aβ PET positivity (45). These previous studies were limited in that they did not examine SIEs in relation to initial learning capacity [measured by maximum recall score (Cued A2) after two list A trials]. The strength of initial learning appears to attenuate Cued B1 SIEs but does not have nearly the effect on the attenuation of SIEs that are elicited on Cued B2, even though two consecutive learning trials of list B had occurred at that point. It might be argued that initial strength of encoding and recall of the original list A targets might have created the opportunity for greater proactive semantic interference when semantically competing list B items were presented using identical category retrieval cues (fruits, musical instruments, and articles of clothing). However, in this study, the degree of initial learning of list A targets had less of an effect on SIEs on Cued B2 among aMCI p-tau₂₁₇+ vs. aMCI p-tau₂₁₇-. This indicates that SIEs generated during a cognitive challenge measured by the frPSI most likely represented a continuous breakdown in self-monitoring and response inhibition in the face of semantically competing information, despite continued opportunities to learn. We postulate that the LASSI-L elicits a far greater number of semantic intrusions than other list learning tasks given that both the A list and B list contain 15 targets that have identical semantic cues used during both acquisition and retrieval. Presenting identical semantic cues on list B Cued B1 and Cued B2 retrieval trials likely enhances the semantic interference effect (10, 47) that occurs during activation of both semantic and episodic memory circuits that have been conceptualized as interdependent (48) and affected during the earliest stages of prodromal AD.

The mechanisms underlying frPSI are being more actively explored. Recently, Valles-Salgado et al. (49) reported that low correct responses on tests sensitive to frPSI on the LASSI-L are associated with ¹⁸F fluorodeoxyglucose (FDG) PET abnormalities in AD-relevant regions, including the medial temporal lobes, precuneus, and posterior cingulate. Rolls (50) postulated that there are connections between the limbic regions and cortical systems, with the posterior cingulate cortex being involved in memory due to its connections with the hippocampal memory system. With regards to SIEs, Sánchez et al. (12) found that the only LASSI-L measures that distinguished between middle-aged asymptomatic children of persons diagnosed with late-onset AD from controls were SIEs on Cued B2 of the LASSI-L. Moreover, the number of such intrusion errors were related to corticolimbic disconnection on fMRI. Taken together, these findings have led to the conclusion that SIEs, particularly as they occur during tests of frPSI, reflect deficits in self-monitoring and the inability to inhibit previously learned responses (1).

This study enhances our understanding of cognitive breakdowns and their links to promising plasma biomarkers of AD. We found that Cued B2 SIEs were predicted by a multivariate combination of ApoE ε4, p-tau₂₁₇ positivity, and age, with explained variance in Cued B2 SIEs surpassing that previously reported for p-tau181 (1).

Strengths of the current investigation include a comprehensively evaluated and broadly representative sample of older adults with plasma p-tau₂₁₇ levels and ApoE ε4 genotyping, enabling comparison on tests measuring important cognitive markers relevant to prodromal AD namely PSI, frPSI, and SIEs. Additionally, there were no significant demographic differences between the p-tau₂₁₇+ and p-tau₂₁₇- groups on important variables such as years of education, sex, or Hispanic/Latino ethnicity. Another strength was statistical adjustment for baseline differences with regards to age and MMSE scores in aMCI participants across the p-tau₂₁₇+ and p-tau₂₁₇- groups.

A potential limitation of the study is the insufficient number of cognitively normal and preMCI participants (cognitively impaired-not MCI) who are p-tau₂₁₇ positive, limiting meaningful statistical comparisons with the current aMCI sample. Although ApoE ε4, p-tau₂₁₇ levels, and age were predictive of Cued B2 SIEs, a significant portion of the variability in SIEs remained unexplained by the regression models, which represents a potential limitation of the current study. This unexplained variance may stem, in part, from differences in cognitive reserve in the presence of brain pathology. Furthermore, beyond the scope of this investigation, emerging research suggests that factors such as neuroinflammation and co-pathologies may mediate or moderate the relationship between p-tau217 and SIEs. Future studies with larger sample sizes and longitudinal follow-up are needed to better understand the predictive value of plasma biomarkers and deficits in self-monitoring and semantic inhibitory control (as reflected by SIEs) in relation to clinical progression over time (51, 52). Given the brief administration time of the LASSI-L, SIEs could serve as a useful tool to identify individuals who may require more detailed assessments for clinical trial inclusion. As AD/ADRD research advances toward blood-based biomarker validation for clinical use, the interpretive value of these biomarkers will be enhanced if they align with AD-specific cognitive deficits, such as SIEs, which are easily measurable errors indicative of clinically relevant decline.

While the study sample is ethnically diverse, participants were recruited from a single research center in South Florida, which may limit applicability to more culturally or geographically heterogeneous populations. It is also important to note that although bilingual assessment was conducted in participants' preferred language (Spanish or English) by fluently speaking and culturally trained clinicians, subtle culturally based linguistic deviations may influence performance on verbal memory tasks, specifically in semantic intrusion errors, and differ in other bilingual or monolingual populations. Statistical modeling was standard for a study of this type but focused on individuals with amnestic MCI. As noted in the Discussion Section, there were only a modest number cognitively normal older adults, or individuals with non-amnestic profiles or those with PreMCI (Impaired-Not MCI) who were p-tau positive. As such, findings among those with aMCI may not generalize to these particular groups. Finally, because many individuals with aMCI in our sample also showed impairments in non-memory domains, statistical power was limited to target the unique effects of memory vs. non-memory dysfunction on semantic intrusion errors.

The clinical implications of these findings are significant. Plasma p-tau₂₁₇ (16) has emerged as the most sensitive and predictive biomarker of Alzheimer's disease. However, many individuals lack access to costly and time-intensive amyloid PET scans available only at major medical centers. Blood-based biomarkers offer an affordable and efficient means for early detection of amyloid and tau accumulation in the brain. Nevertheless, p-tau₂₁₇ alone, without the inclusion of a brief cognitive challenge test such as the LASSI-L or similar assessments in community-based settings, may not hold sufficient clinical relevance. This underscores the need for further research to explore the link between blood biomarkers with sensitive cognitive testing to improve the early identification of persons at risk for AD/ADRD.