18 Sep 2020

As Aβ accumulates in a person’s brain during the long preclinical stage of Alzheimer’s disease, deficits in learning emerge prior to impairments in episodic memory, according to a study published in Neurology on September 4. Cognitively normal people who tested positive for brain amyloid learned fewer Chinese characters over a six-day period than their amyloid-negative peers. Their learning deficit was more pronounced than any form of memory impairment, and it correlated with smaller hippocampi. The researchers, led by Yen Ying Lim at the Florey Institute and Paul Maruff of Cogstate, Ltd., both in Melbourne, Australia, suggest that learning tests could be used to more sensitively detect emerging cognitive deficits in the earliest stages of the disease.

People with brain amyloid learned Chinese characters more slowly than people without.
This deficit was greater than any memory impairment.
It correlated with less hippocampal volume.

In public parlance, memory loss has become almost synonymous with AD. Alas, during the disease’s preclinical stage—which scientists are trying to target for intervention— memory loss creeps up slowly and varies from person to person (Baker et al., 2017; Feb 2018 news; Aug 2018 news). Hence trials enrolling people at those stages struggle to assess clinical efficacy of the drug under investigation. New tests are sorely needed.

In the Australian Imaging, Biomarker, and Lifestyle study, Lim, Maruff, and colleagues had previously reported that cognitively normal, Aβ-positive people performed as well on tests of episodic memory as their Aβ-negative peers (Baker et al., 2019). Curiously, though, after several years of repeated testing, Aβ-positives had not improved, while Aβ-negatives benefitted from practice. Studies of other cohorts also reported this lack of a practice effect in preclinical AD (Hassenstab et al., 2015).

Suspecting that a learning deficit was at play here, the scientists devised a means to tease it out within days, rather than years. Called Online Repeated Cognitive Assessment, aka ORCA, and taken on a home computer, their test taught participants the meaning of 50 Chinese characters (see image below). In daily half-hour sessions, participants viewed two training blocks of 200 Chinese characters paired with a spoken English noun. For each pair, they were asked to guess if the translation was correct or incorrect, but received no feedback on their answers. The ratio of correct to incorrect pairings presented to the participants increased from block to block. On day 1, participants saw two correct pairings and two incorrect pairings. Incorrect pairings were not repeated to avoid off-target learning. On day 2, the correct-to-incorrect pairing ratio doubled, on day three it tripled, etc., so that by day six, there were 24 correct and only two incorrect pairings. In this way, the volunteers increasingly connected each character with its proper meaning. No one in the study previously knew Chinese characters, Maruff said, so at the start of the first day they were blindly guessing, and by the end of the first session, they had started to learn. ORCA was modeled after a previously developed learning test that had used a “nonsense language,” instead of Chinese characters, Maruff said (Breitenstein and Knecht, 2002). The benefit of using Chinese characters instead is that participants felt they were learning something meaningful.

Familiar Words, Novel Characters. Chinese characters paired with spoken English words were presented to participants, who were asked to guess whether each translation was correct. They received no feedback, but each day the number of correct pairings increased, such that volunteers began to recognize correct translations by experience. [Courtesy of Lim et al., Neurology, 2020.]

At AAIC in 2018, co-author Jenalle Baker had presented preliminary findings of this test from a cohort of 50 cognitively normal participants, including 20 whose amyloid-PET scans were positive, and 30 whose scans were negative. She reported that while both groups learned to associate characters with spoken words to some extent, the amyloid-positive volunteers made more mistakes already by the second session and the gap widened over the six-day period (Aug 2018 conference news).

The new study tested 80 cognitively normal participants—42 amyloid-negative, 38 -positive—and included other cognitive and neuroimaging measures. As had been seen in the previous cohort, amyloid-negative participants learned the meaning of Chinese characters faster, with accuracy differences between the groups emerging on day one and growing in each session. The two groups’ average rate of learning over the entire six days differed by more than two standard deviations; as gauged by a Cohen’s d value, this came to an effect size of 2.2.

**Learning Gap.** People without amyloid (black line) learned the meaning of Chinese characters faster than did people with amyloid (red line). *[Courtesy of Lim et al., Neurology, 2020.]*

In contrast, the groups barely differed on their most recent scores on any test of episodic memory, as measured using single and composite scores from the California Verbal Learning Test, Second Edition (CVLT-II), and the Logical Memory (LM) subtest of the Wechsler Memory Scale–Revised (WMS-R). Even though memory scores in the amyloid-positive group declined more than in the amyloid-negative over the previous six years, the Cohen’s d effect size of 0.52 was four times smaller than that of the six-day learning deficit.

Among Aβ-positive participants only, those who learned the Chinese characters more slowly had smaller hippocampi and larger brain ventricles, suggesting less gray-matter volume.

Michael Weiner of the University of California, San Francisco, called the study highly innovative. He called the effect size of the learning deficit substantial. “It will be very interesting to determine the ability of such tests to detect longitudinal cognitive decline,” he wrote, noting that the results need to be replicated.

Nikki Stricker of the Mayo Clinic in Rochester, Minnesota, believes the study highlights the importance of learning deficits in preclinical Alzheimer’s disease and their potential in clinical testing. “The large effect size elicited over a short period of time … illustrates that a well-designed learning paradigm can outperform traditional measures of delayed recall, and brings to light the important role that sensitive cognitive measures play in the early detection of Alzheimer’s disease,” Stricker noted.

The findings mesh with studies led by Kevin Duff at the University of Utah in Salt Lake City, which found that weaker practice effects on memory tests taken over one week correlated with amyloid burden (Duff et al., 2014; Duff et al., 2017). “Taken together, these two sets of results indicate that more amyloid in the brain leads to a failure to benefit from experience with novel stimuli, even over days to weeks,” commented Duff. “Ultimately, deficits in learning, especially repeated exposure to novel information, may be one of the earliest harbingers of eventual Alzheimer’s disease.”

Ironically, scientists view practice effects as a fly in the ointment in clinical studies that track cognition. They have gone to great lengths to minimize them by changing the material between tests. Unexpected practice effects have shown up in the DIAN trial of solanezumab and gantenerumab and the Banner/Novartis trial of umibecestat (Apr 2020 conference news; Aug 2020 conference news). “The findings from Lim et al. suggest that learning rates could instead be considered to screen for early cognitive impairment or even future cognitive decline,” wrote Oskar Hansson and David Berron of Lund University, Sweden.

Maruff told Alzforum that the diminished ability to acquire new information on the basis of experience echoes complaints from patients. “They notice something is wrong when they try to learn something new, yet score normally on memory tests,” Maruff said. While ORCA has a memory dimension, the researchers skewed it toward acquisition of new information by including large amounts of it—too much to learn in a single sitting.

This six-day test is not ready for use in clinical trials, Maruff said. The researchers are exploring different ways to implement it. It could be employed at different times throughout a study, each time using new Chinese characters, or some other form of novel information. Alternatively, the researchers could spread out the single-day sessions—one day spaced every two weeks, for example—then compare learning curves of different groups throughout the trial. Cogstate sells computer-based cognitive assessment tools for clinical trial and research applications.

“Tools such as the ORCA-LLT have the advantage of being scalable and simple to administer, making them good candidate tools for population-level screening and longitudinal monitoring,” commented Michael Gold of AbbVie in Chicago. “One can only hope to see rapid independent replication of this paper in a larger cohort and data looking at the effects of APOE and tau on learning efficiency.”

Michelle Farrell of Massachusetts General Hospital in Charlestown wrote that further pinpointing when the learning deficit arises in preclinical AD, and whether it is driven by amyloid, tau, or both pathologies, will help researchers decide when and if they should shift away from traditional measures of memory recall to learning tests instead.

The findings also raise the question of how amyloid deposition interferes with brain networks involved in learning, Hansson and Berron added. “This is especially true for the posterior-medial recollection network, which is particularly prone to early deposition of Aβ plaques,” they wrote. Colin Masters of the Florey Institute, a co-author on the paper, agreed that in addition to accumulating amyloid early in AD, the posterior parietal cortex/precuneus is critical for learning (Brodt et al., 2018). Aβ-inflicted synaptic damage in that region could lead to learning defects, and also disrupt connections with the medial/temporal lobe, ultimately manifesting in more overt memory impairments as preclinical AD progresses, he wrote.—Jessica Shugart

Comments

- Jason Hassenstab
  Washington University
- Posted: 18 Sep 2020
Learning, practice effects, learning to learn, statistical learning, associate learning, retest, performance gains—there are many ways to describe the underlying construct in the exciting novel task called the Online Repeatable Cognitive Assessment-Language Learning Task (ORCA-LLT) used here by Lim et al.

The task itself has broad appeal: Participants are taught to recognize Chinese characters over several days simply by asking them to decide whether a character matches an English word. Critically, no feedback is given during the test. The participants learn to associate the correct English word simply by observing the correct matches much more often than the incorrect matches.

This touches upon statistical learning approaches from cognitive psychology, but the appeal in this particular task is that unlike most cognitive testing done in clinical studies, the participants gain something from participation—and something that could be interesting and even useful to them in their everyday lives, such as learning a second language. As the field of Alzheimer’s disease clinical trials searches for cognitive outcome measures, we must face the reality that there are significant shortcomings with our current methods of lengthy, one-shot assessments that often repurpose conventional neuropsychological and mental status measures that were designed to measure gross deficits and were not designed to be sensitive to the earliest brain changes that occur in the asymptomatic stages before dementia onset. Current methods also suffer from extremely low reliability. Day-to-day fluctuations in fatigue, mood, attentiveness, short-term illnesses, not to mention test anxiety, can produce high variability in test performance. In clinical trials, this variability can have drastic impacts on statistical power.

ORCA-LLT captures learning rates, which have been shown to be more indicative of the earliest AD-related changes than recall deficits, and does so with extremely high reliability due to its format of long exposures to test stimuli (experienced in the participant’s familiar environment) over several consecutive days. The results of the study suggest that this method of assessment has exceptional promise. Cognitively normal but amyloid-positive older adults demonstrated extremely large effect-size impairments in learning rates compared to amyloid-negative older adults. To contrast between long-term and short-term learning, the study compared their learning rates on ORCA-LLT to learning rates across repeated AIBL study visits at 18-month intervals and to meta-analytic studies from the extant literature on preclinical AD. The ORCA-LLT had four times the effect size observed in long-term learning. In addition to amyloid, the test was also significantly related to hippocampal volume, which is often used as an indicator of neurodegeneration in AD studies.

There are of course challenges with this approach. Like any remote study that is unsupervised, adherence can be a significant problem. This method of assessment requires participants to independently log onto a website for up to 30 minutes a day for six consecutive days and to endure a potentially very frustrating experience since the paradigm, by design, provides no feedback. On day one, the task appears to be totally random and some participants that need a sense of mastery may react strongly to the unstructured nature of the task so day one dropout is a valid concern. Handling missing data is another challenge with this task design. If all participants are not given the same opportunity to learn the stimuli over the same period, how does one model these data to be comparable to the other participants? Although not addressed in the publication, the authors are indeed aware of these concerns.

There are so many possibilities to use this paradigm with different stimuli (learning abstract art and artists’ names, flora and fauna identification, faces and names, etc.), different set lengths (perhaps 20 characters over three days is enough?), and over different time periods (can learning rates be produced in one day that are just as sensitive?). Although the concept of this task is not entirely new, the clever design, the clinical meaningfulness of the task, and the readily available technology in the current age opens the door for learning rates as an assessment modality that could have game-changing impacts for observational studies and clinical trials.

View all comments by Jason Hassenstab

- David Berron
  German Center for Neurodegenerative Diseases (DZNE)
- Oskar Hansson
  Lund University
- Posted: 18 Sep 2020
Lim and colleagues use a language-learning test, with repeated learning sessions containing the same learning material, to investigate learning performance (instead of memory performance) in 80 older adults.

Following a supervised first learning session, participants continued with five more online sessions in the following days, where they learned associations of Chinese and English vocabulary. In addition, the authors determined amyloid-β status using positron emission tomography and they analyzed hippocampal and ventricular volume derived from MRI scans. Their main finding shows that Aβ-positive individuals show lower learning rates across six learning sessions on consecutive days. In addition, hippocampal volume was smaller, while ventricular volume was larger, for those individuals with smaller learning effects. Lim et al. conclude from their results that Aβ is primarily associated with a learning deficit and thus with a failure to benefit from earlier learning experience.

While we usually try to prevent or minimize learning and practice effects in repeated assessments of cognitive function by employing different sets of difficulty-matched parallel test versions, the findings from Lim et al. suggest that learning rates could instead be considered as a screen for early cognitive impairment, or even future cognitive decline, and thus add to a body of earlier papers on practice effects that have recently been reviewed (Jutten et al., 2020).

While these results strengthen the evidence for the usefulness of learning-related markers as measures of cognitive function in research studies and potentially in clinical trials, they also highlight our lack of understanding regarding the mnemonic and neural mechanisms. While the observed relationships with hippocampal atrophy suggest a medial temporal origin, future studies need to identify and dissect the wider functional networks involved in repeated learning and their local relationship with accumulation of Aβ and tau pathology. This is especially true for the posterior-medial recollection network, which is particularly prone to early deposition of Aβ plaques.

References:

Jutten RJ, Grandoit E, Foldi NS, Sikkes SA, Jones RN, Choi SE, Lamar ML, Louden DK, Rich J, Tommet D, Crane PK, Rabin LA. Lower practice effects as a marker of cognitive performance and dementia risk: A literature review. Alzheimers Dement (Amst). 2020;12(1):e12055. Epub 2020 Jul 9 PubMed.

View all comments by David Berron View all comments by Oskar Hansson

- Michael Gold
  AbbVie
- Posted: 18 Sep 2020
Once again, our friends from Australia and AIBL have conducted a very interesting experiment that could have significant potential for improving the diagnosis and prognosis of Alzheimer’s Disease (defined as abnormal Aβ deposition) as well as the development a new set of tools that could be used in clinical trials.

The gist of the experiment consisted of comparing the performance of a group of “cognitively normal” subjects who were Aβ+ with a group who were Aβ–, using a novel cognitive task (learning the association between a Chinese character and its meaning in English—the ORCA-LLT). The paper reports that Aβ+ subjects were not as efficient as Aβ– subjects when learning the new material over the six days of learning trials. In addition, the deficits seen in Aβ+ subjects correlated with increased ventricular volume and reduced hippocampal volumes, whereas performance on the task did not correlate with either measure in Aβ– subjects.

What does this mean? For one, we need to rethink how we define cognitively normal. Models of memory all include some form of acquisition, consolidation, and retrieval functions, and some traditional tests of declarative verbal memory (e.g., CVLT) include measures of learning efficiency. However, the brunt of focus on memory in AD patients (across the spectrum of severity) has been on measures of spontaneous or cued retrieval. The novelty of this paper is that it serves to remind us that the acquisition of novel information is impaired at very early stages of AD and that there are tools we can use now to measure learning efficiency.

Moreover, tools such as the ORCA-LLT have the advantage of being scalable and simple to administer, making them good candidates for population-level screening and longitudinal monitoring. One can only hope to see rapid independent replication of this paper in a larger cohort and data looking at the effects of APOE and tau on learning efficiency.

View all comments by Michael Gold

- Kevin Duff
  University of Utah
- Posted: 18 Sep 2020
Lim et al. found that cognitively healthy older adults with Alzheimer’s pathology (greater amounts of amyloid) in their brains struggled to learn a series of Chinese characters, and their English-language equivalents, over six days compared to older adults without this brain pathology. This finding is consistent with our own work showing that one’s ability to learn to take memory tests during one week is related to the amount of that same Alzheimer’s pathology in the brain (Duff, 2014; Duff et al., 2017).

Taken together, these two sets of results indicate that more amyloid in the brain leads to a failure to benefit from experience with novel stimuli, even over days to weeks. What might be the implications of this information? Our ability to benefit from experience (whether learning Chinese/English pairs, taking memory tests, completing new tasks at work, or using a new remote control) is crucial for successful adaptation to daily life. Failure to benefit from that experience may have dire consequences.

Ultimately, deficits in learning, especially repeated exposure to novel information, may be one of the earliest harbingers of eventual Alzheimer’s disease. However, such information might also lead to interventions that bolster learning abilities and reduce risk of developing such conditions.

References:

Duff K. One-week practice effects in older adults: tools for assessing cognitive change. Clin Neuropsychol. 2014;28(5):714-25. Epub 2014 Jun 2 PubMed.

Duff K, Atkinson TJ, Suhrie KR, Dalley BC, Schaefer SY, Hammers DB. Short-term practice effects in mild cognitive impairment: Evaluating different methods of change. J Clin Exp Neuropsychol. 2017 May;39(4):396-407. Epub 2016 Sep 20 PubMed.

View all comments by Kevin Duff

- Nikki Stricker
  Mayo Clinic
- Posted: 18 Sep 2020
This is an elegant study that highlights the central nature of a learning deficit in preclinical Alzheimer’s disease. Although meta-analytic studies have suggested that measures of learning are as useful as measures of delayed recall for the prediction of conversion from MCI to dementia (Belleville et al., 2017) and for differentiating individuals with MCI or Alzheimer’s dementia from cognitive normal adults (Weissberger et al., 2017), there remains a predominant focus in the field on measures of delayed recall.

The large effect size elicited over a short period of time on the novel learning measure developed by Lim and colleagues illustrates that a well-designed learning paradigm can outperform traditional measures of delayed recall and brings to light the important role that sensitive cognitive measures play in the early detection of Alzheimer’s disease.

References:

Belleville S, Fouquet C, Hudon C, Zomahoun HT, Croteau J, Consortium for the Early Identification of Alzheimer’s disease-Quebec. Neuropsychological Measures that Predict Progression from Mild Cognitive Impairment to Alzheimer's type dementia in Older Adults: a Systematic Review and Meta-Analysis. Neuropsychol Rev. 2017 Dec;27(4):328-353. Epub 2017 Oct 10 PubMed.

Weissberger GH, Strong JV, Stefanidis KB, Summers MJ, Bondi MW, Stricker NH. Diagnostic Accuracy of Memory Measures in Alzheimer's Dementia and Mild Cognitive Impairment: a Systematic Review and Meta-Analysis. Neuropsychol Rev. 2017 Dec;27(4):354-388. Epub 2017 Sep 22 PubMed.

View all comments by Nikki Stricker

- Michelle Farrell
  Massachusetts General Hospital
- Posted: 18 Sep 2020
It is encouraging for secondary prevention trials seeking well-powered cognitive outcome measures that such a strong amyloid-related learning deficit is detectable over a short period in preclinical AD. Though not directly tested in the present study, it is particularly striking that the deficit in learning rate appears even at low-to-moderate Centiloid levels (i.e., 25-50 CL).

Future studies expounding when this learning deficit appears and whether amyloid or tau alone, or the combined effect of both pathologies, drive this learning deficit will be critical to determine whether preclinical AD researchers ought to shift away from traditional measures of memory recall to these measures of novel short-term learning.

View all comments by Michelle Farrell