Working memory filtering at encoding and maintenance in healthy ageing, Alzheimer's and Parkinson's disease

Abstract

The differential impact on working memory (WM) performance of distractors presented at encoding or during maintenance was investigated in Alzheimer's Disease (AD), Parkinson's Disease (PD) patients, elderly (EHC) and young healthy controls (YHC), (n = 28 per group). Participants reported the orientation of an arrow from a set of either two or three items, with a distractor present either at encoding or at maintenance. MRI data with hippocampal volumes was also acquired. Mean absolute error and mixture model metrics i.e., memory precision, target detection, misbinding (swapping the features of an object with another probed item) and guessing were computed. EHC and PD patients showed good filtering abilities both at encoding and maintenance. However, AD patients exhibited significant filtering deficits specifically when the distractor appeared during maintenance. In healthy ageing there was a prominent decline in WM memory precision, whilst in AD lower target detection and higher guessing were the main sources of error. Conversely, PD was associated only with higher guessing rates. Hippocampal volume was significantly correlated with filtering during maintenance - but not at encoding. These findings demonstrate how healthy ageing and neurodegenerative diseases exhibit distinct patterns of WM impairment, including when filtering irrelevant material either at encoding and maintenance.

Keywords: Ageing; Alzheimer’s disease; Filtering; Hippocampus; Parkinson’s disease.

Fig. 1

Filtering at Encoding and Maintenance Task design. Participants were presented with either two (SS2) or three (SS3E) arrows at encoding and were instructed to remember the orientations of all arrows (SS2 and SS3E), or to ignore one of them in the Filter at Encoding (FE) condition. After a blank interval (2000 msec), one of the target arrows (probe) re-appeared in black in its original location but in a random orientation, which the participants had to rotate to its remembered orientation. In the Filter at Maintenance (FM) and Set Size three at Maintenance (SS3M) the first two arrows appeared simultaneously and had to be remembered, and after a 500 msec delay during the maintenance period, a third arrow appeared for 1000 msec, which either had to be ignored (FM), or remembered (SS3M). After 500 msec the probe was displayed and had to be rotated to its original orientation.

Fig. 2

Mean absolute error (MAE) and mixture model metrics in young (YHC) and elderly healthy controls (EHC). Performance of participant groups are shown labeled with different colors, EHC in coral red, YHC in turquoise. Set Size 2 (SS2), Filtering at Encoding (FE), Filtering at Maintenance (FM), Set Size three at Encoding (SS3E), Set Size three at Maintenance (SS3M). 2 A: MAE performance across the five different conditions. On the Y-axis MAE in degrees. Shaded areas represent confidence intervals (CI). 2B: Mixture Model metrics across conditions in YHC and EHC. Panel a: Precision as concentration parameter κ, Panel b: Target probability, panel c: Misbinding probability, panel d: Guessing probability.

Fig. 3

Mean absolute error (MAE) and mixture model metrics in EHC and AD. Performance across the five different conditions: Set Size 2 (SS2), Filtering at Encoding (FE), Filtering at Maintenance (FM), Set Size three at Encoding (SS3E), Set Size three at Maintenance (SS3M). The two groups are shown as EHC in coral red, AD in green. 3 A: On the Y-axis MAE in degrees. Note in particular the differential effect of Filtering at Maintenance (FM) compared to Filtering at Encoding (FE) within the AD group compared to within the EHC group. Further, in AD, FM performance (when a to-be-ignored distractor was presented during maintenance) was equivalent to that Set Size three at Maintenance (SS3M, when the new item had to be retained). 3B: a | Precision as concentration parameter κ. b | Target probability. c | Misbinding probability. d | Guessing probability.

Fig. 4

Mean absolute error (MAE) and mixture model metrics in EHC and PD. Performance across the five different conditions: Set Size 2 (SS2), Filtering at Encoding (FE), Filtering at Maintenance (FM), Set Size three at Encoding (SS3E), Set Size three at Maintenance (SS3M). Different participant groups are labeled with different colors, i.e. EHC in coral red, PD in violet. 4 A: On the Y-axis, MAE in degrees for the two groups. 4B: a | Precision as concentration parameter κ. b | Target probability. c | Misbinding probability. d | Guessing probability.

Fig. 5

Mixture model parameters correlations with hippocampal volumes. a | Correlation between Filtering rate at Maintenance and whole hippocampal volume (WHV). b |Correlation between Mixture Model metrics and whole hippocampal volume. On the X-axis the volume in mm³ of the whole hippocampus. In b, the Y-axis represent from left to right respectively MAE in degrees, probabilities of Target detection, Misbinding and random Guessing.

Fig. 6

Filtering and MAE stratified by WHV and diagnosis. FM on the X-axis, MAE on the Y-axis, hippocampal volumes (WHV) on the Z-axis. Subjects with higher WHV (top of the image) have lower MAE (back of the image), lower Filtering rate at Maintenance scores (on the left) and are predominantly represented by elderly healthy subjects (in coral red), and PD patients (in violet). On the other end of the spectrum, AD patients (in green) have higher Filtering rate at Maintenance scores (on the right), higher MAE (front of the image), and have lower hippocampal volumes (bottom of the image).