Investigating the Association Between Verbal Forgetting and Pathological Markers of Alzheimer’s and Lewy Body Diseases

INTRODUCTION

Alzheimer’s disease (AD) and dementia with Lewy bodies (DLB) are the two most common forms of neurodegenerative dementia [1]. Although less prominent in DLB, both AD and DLB patients experience persistent memory impairments during the early stages of the disorders [2, 3] making it difficult to differentiate between these disorders. An incorrect diagnosis of AD has both clinical and economic consequences, as patients may have to pay for inappropriate treatments and medical services until a correct diagnosis is given [4, 5]. These costs are not only incurred by patients and their families; indeed, misdiagnoses of AD also costs millions of dollars to society [5]. In research protocols, incorrect diagnoses may result in incorrect conclusions about study findings. As such, an accurate, low-cost measure is needed to correctly differentiate AD from DLB.

Such a measure was recently proposed by Bussè and colleagues [6], who examined whether specific measures of the Rey Auditory Verbal Learning test (RAVLT) [7] could be used to differentiate between these dementias. They found that DLB patients had higher performance in verbal learning (VL) than AD patients, and a lower frequency of verbal forgetting (VF) and reduced recency effect. Statistically, the groups could be distinguished based on percentage of verbal forgetting (VF%), which is a retention index of words recalled after a 20–min delay, adjusted for short-term memory performance. These results suggest that the RAVLT may potentially be used as a simple, easily accessible, low-cost instrument to discriminate these dementias. However, a major limitation to Bussè and colleagues’ study was that the diagnoses of DLB and AD were based solely on the clinical progression of symptoms at a three-year follow-up and were not corroborated with autopsy or biomarker confirmation. This is problematic because misdiagnosis of DLB as AD based on clinical presentation is frequent [8] with sensitivity estimates <33% [9]. Indeed, up to half of individuals who are carefully screened and selected for AD in cohort studies [10] and clinical trials [11] are later confirmed to have DLB pathology at autopsy. It is unknown whether RAVLT measures may be helpful in screening for DLB pathology in individuals presenting phenotypically as AD.

Histopathological confirmation at autopsy is the only definitive diagnostic method [12] and serves as the gold standard for determining the accuracy of clinical diagnoses of AD [13] and DLB [14]. The pathological hallmarks that define AD are amyloid plaques and neurofibrillary tangles (NFTs) [15] and the defining pathology in DLB is the presence of protein aggregates known as Lewy bodies [16]. In premortem studies, in vivo biomarkers have become an essential component of classifying dementias and are the second-best method of confirming clinical diagnoses, after autopsy. Biomarkers for corroborating a diagnosis of AD include reduced levels of amyloid-β (Aβ) and elevated levels of total-tau (t-tau) and phosphorylated-tau (p-tau) in cerebrospinal fluid (CSF), which have been associated with plaques in the hippocampus and neocortex [17] and NFT pathology in several neocortical regions [18], respectively. A third biomarker for AD is medial temporal lobe neurodegeneration seen on structural neuroimaging [19] which is strongly correlated with percent area of amyloid plaques and NFTs but not Lewy body pathology [20]. Relative preservation of medial temporal lobe structures, on the other hand, is a supportive biomarker for the diagnosis of DLB [3] with possible atrophy of basal ganglia structures (e.g., the putamen) [21–23]. Reduced CSF levels of α-synuclein have also been proposed as a biomarker for DLB [24] as this may reflect the buildup of Lewy body pathology [25]. Lastly, occipital hypometabolism is an additional supportive biomarker for DLB [3].

In the absence of diagnostic confirmation from autopsy or biomarker data, no definitive conclusions can be made regarding the usefulness of the cognitive markers proposed by Bussè and colleagues to distinguish true DLB from AD, independent of clinical presentation. The purpose of the current study was to determine if the percentage of verbal forgetting (VF%) measure proposed by Bussè et al. [6] is accurate in differentiating DLB from AD pathology, quantified using in vivo biomarkers and postmortem pathology. Because our aim was to directly test the claims put forth by Bussè and colleagues (i.e., that using VF% as an initial classification measure should robustly distinguish AD from DLB, in the absence of any other markers or clinical indicators), we elected to first classify cognitively impaired participants based on VF%, then to verify whether this classification aligned with more established markers of disease pathology (CSF levels of Aβ, tau, and α-synuclein, as well as integrity of medial temporal lobe structures). Of note, this is how RAVLT scores would be used clinically if a practitioner were to follow Bussè and colleagues’ recommendations and apply VF scores as cognitive markers of AD and DLB. Based on the results of the original study, it was hypothesized that participants who were classified as probable AD based on their VF% scores (AD_VF%) would have higher CSF levels of p-tau and t-tau, lower levels of CSF Aβ, and greater hippocampal atrophy compared to participants classified as probable DLB based on their VF% scores (DLB_VF%). AD_VF% classification was also expected to be associated with significant plaques and tangles at autopsy. In addition, it was hypothesized that DLB_VF% participants would have lower CSF levels of α-synuclein, atrophy of the putamen with relative hippocampal sparing, and greater occipital hypometabolism compared to AD_VF% participants. DLB_VF% classification was also expected to be associated with significant Lewy bodies at autopsy. A sample of healthy controls was included as a benchmark for normal age-related change in CSF protein levels and brain structures.

METHODS

The data used in this study were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (http://adni.loni.usc.edu/), a large-scale data sharing effort across North America to study the progression of AD. ADNI is led by Principal Investigator Michael W. Weiner, MD and was launched in 2003 as a public-private partnership. The primary goal of ADNI has been to determine whether the progression of mild cognitive impairment (MCI) and AD can be measured by combing serial magnetic resonance (MR) imaging, positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessments. Up-to-date information can be found at http://www.adni-info.org.

RESULTS

CSF biomarkers

All biomarker data are summarized in Table 2. Baseline levels of CSF Aβ (χ² (2)  = 150.134, p < 0.001, r = 0.51; Fig. 1A), t-tau (χ² (2) = 128.788, p < 0.001, r = 0.48; Fig. 1B), p-tau (χ² (2) = 85.732, p < 0.001, r = 0.39; Fig. 1C), and α-synuclein (χ² (2) = 9.400, p = 0.009, r = 0.21; Fig. 1D) differed between groups. However, post-hoc analyses revealed that AD_VF% and DLB_VF% participants differed from each other only on levels of t-tau (p = 0.021, r = 0.18) and marginally on levels of Aβ (p = 0.056, r = 0.16). They did not differ from each other on levels of p-tau (p = 0.180, r = 0.13) or α-synuclein (p = 1.000, r = 0.01). Control participants had significantly higher levels of Aβ (p < 0.001, r = 0.52) and α-synuclein (p = 0.027, r = 0.20) than AD_VF% participants, and lower levels of t-tau (p < 0.001, r = 0.49) and p-tau (p < 0.001, r = 0.39). Control participants also had significantly higher levels of Aβ (p < 0.001, r = 0.33) and lower levels of t-tau (p < 0.001, r = 0.28) and p-tau (p < 0.001, r = 0.24) than DLB_VF% participants, as well as marginally higher CSF levels of α-synuclein (p = 0.055, r = 0.20).

OPEN IN VIEWER

Fig.1

Mean CSF Aβ (A), CSF t-tau (B), CSF p-tau (C), CSF α-synuclein (D), putamen volume (E), hippocampal volume (F), and occipital glucose metabolism (G) of normal controls and patients diagnosed as AD or DLB based on their VF% scores. Error bars represent the standard error of the mean (SEM).

OPEN IN VIEWER

Table 2

Biomarker data

	Group	N	Mean (SD)	Range	χ 2	df	p	p (ADVF% versus DLBVF%)
Aβ	Controls	353	201.53 (50.86)	79–314	150.134	2	<0.001	0.056
	ADVF%	133	139.00 (32.81)
	DLBVF%	84	156.52 (48.22)
t-tau	Controls	351	66.85 (31.13)	20.5–327	128.788	2	<0.001	0.021
	ADVF%	131	123.53 (60.04)
	DLBVF%	83	98.02 (47.06)
p-tau	Controls	353	32.04 (18.50)	6.9–173	85.732	2	<0.001	0.180
	ADVF%	133	51.68 (27.71)
	DLBVF%	84	44.42 (22.37)
α-synuclein	Controls	106	0.82 (1.41)	0.10–12.24	9.400	2	0.009	1.000
	ADVF%	65	0.82 (0.91)
	DLBVF%	39	0.81 (0.63)
Putamen Volume	Controls	206	0.47 (0.07)	0.29–0.65	3.348	2	0.188	0.828
	ADVF%	120	0.46 (0.07)
	DLBVF%	77	0.46 (0.06)
Hippocampus Volume	Controls	206	0.49 (0.09)	.26–.70	56.672	2	<0.001	0.355
	ADVF%	120	0.41 (0.08)
	DLBVF%	77	0.44 (0.09)
Occipital Glucose Metabolism	Controls	74	3.34 (0.20)	2.48–3.86	9.609	2	0.008	1.000
	ADVF%	49	3.20 (0.24)
	DLBVF%	31	3.21 (0.27)

DISCUSSION

This is the first study to use gold standard biomarker and autopsy data to determine if the VF% measure of the RAVLT is accurate in differentiating AD and DLB neuropathological markers. In contrast to Bussè and colleagues’ study [6], the results from the present study do not robustly support the hypothesis that VF% is accurate in differentiating DLB from AD pathology. These results do not appear to be confounded by age or years of education.

Participants classified as AD and DLB based on their VF% scores did not display biomarker levels consistent with our hypotheses; the groups did not differ on most biomarkers, with the exception of CSF t-tau which was more elevated in AD_VF% participants. One possible explanation for these results is that VF% classification results in categorization that is more closely tied to AD severity than to underlying pathology (i.e., VF% ≥75% representing worse severity). However, because the two patient groups did not differ with regards to age, this is unlikely. Additionally, the biomarker data displayed a wide range of high and low levels that should have been captured by the VF% measure had it been accurate in discriminating DLB and AD pathologies. Autopsy data showed that nearly half (46%) of the cognitively impaired participants had comorbid DLB and AD pathology, which was also not captured by the VF% measure.

Another explanation for these findings is that the participants in both patient groups may have had significant overlapping AD and DLB pathology, which may have obscured clear between-group differences in biomarker profiles; although most participants had no autopsy data, comorbid pathologies were common in those with available data, consistent with prior reports of comorbid AD and DLB pathologies at death [10, 11]. For instance, hippocampal atrophy in the DLB_VF% group may be due to concomitant AD pathology, as hippocampal atrophy has been strongly correlated with percent area of plaques and NFTs [20]. Similarly, elevated CSF p-tau and t-tau and reduced Aβ in the DLB_VF% group may also be due to comorbid AD pathology, with these biomarkers having been linked to greater number of plaques in the hippocampus and neocortex [17] and more severe NFT pathology in several neocortical regions [18]. Unfortunately, many participants were missing data for CSF protein values and structural brain volumes, potentially resulting in a lack of power to detect group effects, though the small effect sizes suggest minor existing differences.

Although CSF biomarkers are now a recommended biomarker to corroborate a diagnosis of AD [31], there has been some deliberation in the literature regarding their usefulness to distinguish between AD and DLB. Specifically, some studies have found that CSF Aβ levels do not differ between AD and DLB patients [32–34] suggesting that Aβ may not be a reliable biomarker to distinguish these diseases. That CSF Aβ did not robustly differentiate between AD and DLB may not be surprising, as both AD and DLB are known to display amyloid pathology [15, 35]. Similarly, there is conflicting evidence in the literature regarding whether CSF α-synuclein is a useful biomarker in differentiating AD from DLB, with some studies reporting lower levels of α-synuclein in DLB compared to AD [24, 36] and others reporting no difference [37]. It is thus possible that the results from our study reflect the inconstancy of CSF biomarkers to reliably associate with AD and DLB pathological processes. CSF t-tau may be more useful in differentiating AD and DLB pathologies, though effect sizes for t-tau remained small in the present study. However, even more robust biomarkers such as hippocampal atrophy and postmortem pathologies were not different between the patient groups, tentatively suggesting that VF% is not accurate in differentiating AD and DLB. Although these results are not consistent with Bussè et al.’s findings [6], they do corroborate previous findings that memory impairments are common amongst biomarker-diagnosed AD and DLB patients [38].

The neuropathology data were also inconsistent with the original hypothesis, though autopsy findings should be considered cautiously due to the substantial time lag between RAVLT administration and death (approximately 6 years). It is possible that initial associations between RAVLT scores and postmortem findings were washed out by the accumulation of comorbid pathology over the years following baseline cognitive assessment. We believe this is unlikely, as comorbidities were not systematically associated with longer time prior to autopsy in our sample (Table 3). Furthermore, prior work has found that individuals with cognitive impairment due to AD or DLB pathology can be differentiated based on cognitive performance during the MCI stage, several years before autopsy [39]. Rigorous time-lagged designs can help to untangle the progression and accumulation of disease in clinico-pathologic studies; however, the small sample size in the present study allowed only for a primarily descriptive exploration of these results. Of the 13 AD_VF% with available autopsy data, most participants had either pure AD (31%) or comorbid pathologies (54%), consistent with previous findings that patients with comorbid pathology tend to have a neuropsychological and clinical profile similar to AD [40]. On the other hand, those with DLB_VF% were roughly even split across the ‘pure AD’ (23%), ‘pure DLB’ (31%), and ‘comorbid’ (31%) postmortem classification groups. These results suggest that while a high VF% score is positively associated with the presence of AD pathology (pure or comorbid), the RAVLT is less useful in determining which patients have DLB pathology. Unfortunately, due to the very small number of participants with available autopsy data in this study and to the long clinico-pathologic lag, no strong conclusions can be drawn and results should be replicated independently in other work.

There were several methodological differences between the current study and Bussè and colleagues’ study [6] that may account for the conflicting results. While Bussè and colleagues examined whether VF% scores differ between clinically diagnosed AD and DLB, the present study examined whether VF% could distinguish AD and DLB pathology independent of the clinical diagnosis. Thus, our study was carried out under the premise that the clinical diagnosis of AD may be wrong, as misdiagnosis based on only clinical symptoms is common [8], and our primary question of interest was whether VF% could decipher patients who present clinically as AD but have a high likelihood of underlying Lewy body pathology. A second important difference is that Bussè and colleagues recruited a sample of local participants, which may reflect a better representation of AD and DLB patients than the ADNI database which is highly enriched for AD.

The major limitation to this study is that the ADNI database is specifically selected and enriched for AD. As a result, the number of pure DLB patients in the present study may have been limited. Because of this limitation, the results from the present study may not be generalizable to a broader population. A second important limitation was the considerable amount of missing data in ADNI datasets. In light of positive associations between VF% classification and some diagnostic markers (i.e., APOE ɛ4 status, histopathology data) as well as the almost medium effect sizes for the pairwise comparisons for t-tau and p-tau, it is possible that our hypothesis may have been supported had our study been sufficiently powered to detect positive associations with CSF and neuroimaging biomarkers. Finally, the high degree of overlapping pathology in the already small sample of participants with autopsy data makes it difficult to differentiate the role of these pathologies and their contribution to the clinical manifestations of each patient group, including VF%. In order to find significant differences in a population with a high degree of overlapping neuropathology, a large number of patients with postmortem data will need to be recruited for future studies.

In conclusion, the RAVLT VF% measure does not appear to reliably align with AD and DLB neuropathology. However, due to the selective nature of the ADNI database, future studies will need to be conducted with a less restrictive sample to draw firm conclusions regarding the usefulness of VF% in distinguishing AD from DLB.