Participants

Thirty-two older adults (OAs) aged between 60 and 74 years (M 69.2 years, SD 3.4) were recruited from the local community (56% females). Of these participants, two (females) were excluded from the analyses because of incomplete test data and poor performance (> 3 SD from mean scores of YAs) on the behavioral task. All other participants reported a healthy lifestyle, and no history of neurological or psychiatric illness. Twenty-three healthy young adults aged between 21 and 27 years (M 23.1 years, SD 1.8) served as the control group (91% females). The data from three young adults (2 females, 1 male) were excluded from the analyses due to excessive EEG artifacts (≥ ±150 μV). The final sample comprised data of 30 older (M 69.1 years, 1 left-handed) and 20 younger (M 23.0 years, 2 left-handed) adults, respectively. All participants in the older (9 bilinguals) and younger (10 bilinguals) groups were native New Zealand English speakers. All participants had self-reported normal or corrected to normal vision with glasses or contact lenses. Participants in both groups had 12 or more years of education. None were receiving medication (anti-depressant/anxiety) deemed likely to affect EEG. All participants were compensated per hour for their participation in the study. Although none of the OAs were screened for mild cognitive impairment, it is important to note that all were living independently and none of the participants in the OA group performed poorer than three SD from the mean-scores of the younger participants on any of the behavioral measures described below.

Audiometry

Pure-tone air conduction thresholds were obtained for all participants at octave frequencies from 250 to 8000 Hz (GSI-61, Headphones: TDH39) (see Fig 2). The mean pure-tone threshold (0.250 to 8 kHz) averaged between ears (L/R) for OAs was 20.3 dB HL (SD = 7.3, Range: 10.1–41.3 dB HL). Eighteen OA participants had thresholds ≥ 20 dB HL at or beyond 4 kHz indicating some (mild-severe) high frequency hearing loss [21]. Three participants in this group were frequent hearing aid users for age-related hearing loss (~ 2.5 years of consistent use), but did not wear their hearing aids during testing. These participants were also included as previous studies have shown that hearing aid experience did not alter the ERP components significantly [51, 54]. In addition, the EEG findings (topographical analyses) were replicated with the three participants excluded and hence we chose to include them here so as to not lose statistical power. Stimulus-spectrum audibility was adjusted for OAs with hearing loss (see EEG: Signal audibility). YA listeners had thresholds (0.250 to 8 kHz) better than 15 dB (Mean 5.3, SD = 3.4, Range: 4.3–6.3 dB HL). In both groups, inter-aural hearing asymmetry was not greater than 15 dB at any of the frequencies. All participants had normal tympanograms (GSI TympStar), showing no evidence of middle ear abnormality.

Temporal processing

Sensitivity to monaural TFS cues was assessed using the TFS1 test [55]. The test was installed on a laptop and stimuli were presented via Sennheiser HD 25 1-ii headphones in a double-walled sound-attenuating booth. This test has previously been shown to correlate with speech recognition thresholds in noise and with age [4, 21]. The test utilizes an adaptive two-interval forced-choice decision task, with trial-trial feedback. Participants discriminate a harmonic complex tone (A) from an inharmonic complex tone (B) in which the harmonics are shifted upwards by the same amount in Hz (ΔF). Both ‘A’ and ‘B’ tone complexes had an envelope repetition rate with equal fundamental frequency (F0), but differed in their TFS. The test F0 was 200 Hz and the corresponding center frequency was 2200 Hz. The lowest frequency in the passband was 1800 Hz. All other parameters were set to the test default values. This ensured that the harmonics in the passbands were mostly unresolved, consequently producing a weaker pitch percept which might be conveyed solely based on TFS information [56]. The ΔF was the manipulated variable, and it was initially set to 0.5F0. The ΔF varied from trial to trial according to a 2-down 1-up procedure, to estimate the value of ΔF producing 70.7% correct responses [57]. The overall level of the stimulus was set to 30 dB SL, based on absolute-threshold measurements for pure tones at the filter center frequencies. All participants completed a practice run to ensure they understood the task. Following the practice run, one run was completed for each participant in the better ear, which was determined based on the participant’s audiometric thresholds. Measurements were repeated if the software indicated high variability (SD 0.2) in performance. Test results have been shown to be stable for levels of 30 dB SL and higher [55] and comparisons between the left and right ears show no significant differences for adults with symmetrical hearing [21].

Visual selective attention

Oberfeld and Kloeckner-Nowotny [9] showed evidence for both visual (flanker task) and auditory (backward-masked intensity discrimination) measures of selective attention to explain the same proportion of variance in speech-identification performance. Hence, we used a visual flanker task to measure selective attentional abilities that were independent of the individual’s auditory abilities. The short version of the visual attention network test (ANT; [58]) was administered using the Inquisit 4 software [59]. Distance between the computer screen and the participant was ~ 60 cm. The ANT is designed to measure efficiency of the orienting and executive control networks through the combination of cued (no cue, center cue, spatial cue) and flanker (congruent, incongruent) conditions. While orienting involves the ability to selectively attend and focus on the sensory information, executive control involves the ability to ignore distractors. The short form of the ANT was selected as assessment of the alerting network was not the focus of this study and this helped reduce participants’ cognitive fatigue. During the ANT task, participants are visually presented with a row of arrows and a fixation cross. Participants used a mouse button press to respond to the pointing-direction (i.e., left or right) of a central target arrow while ignoring the surrounding flanking arrows. Participants were instructed to respond as quickly and accurately as possible. The efficiency of the attention network was determined by calculating mean reaction times (RTs, msec) for correct trials as a function of cue and flanker conditions for each participant. Participants first completed a practice block (12 trials) with feedback and then one experimental block (120 trials) with no feedback. Participants were able to use glasses or contact lenses while performing this task.

EEG: Data acquisition and stimulus presentation

The EEG recording was performed in an electrically shielded sound-attenuating booth. During the EEG recording session, participants were seated comfortably and instructed to fixate their eye-gaze on a stationary-cross displayed on a computer monitor. EEG data were acquired from 36 channels: Fz, FP1, FP2, F3, F4, F7, F8, FCz, FC1, FC2, FC5, FC6, FT7, FT8, Cz, C3, C4, T7, T8, CPz, CP1, CP2, CP5, CP6, TP7, TP8, Pz, P3, P4, P7, P8, Oz, O1, O2, M1 and M2, using a Compumedics Neuroscan Quick-Cap with sintered Ag/AgCl electrodes. Scalp electrode montage was based on the international 10/10 system. Vertical eye blinks were recorded with electrodes above and below the right eye and a ground electrode was placed anteriorly on the midline, between FPz and Fz electrodes. A midline electrode placed between the Cz and CPz electrodes served as the online reference channel. Impedances were kept below 5 kΩ using traditional skin aberration method. In a few instances (e.g. high impedances despite skin prep, or thick hair), we adopted the SurePrep method as an alternative to obtain low skin potential values [60]. Signals were amplified via a SynAmps RT system, filtered online with a 400 Hz low-pass, and digitized at a sampling rate of 1000 Hz. The configuration and number of EEG channels were based on previous studies that have decomposed spatio-temporal components associated with auditory target detection [61, 62]. This 36-channel array also allowed us to speed the electrode placement process, prevent bridging between channels, and record good quality data. As spatial resolution was based on a 36 -channel array with a 10–10 montage, localization inaccuracies are possible. A recent study, using a simulated EEG source, showed that localization error distance with 32-channel recording mostly varied between 1.45 and 3.38 mm [63]. While high-density recording channels are preferable, studies have equally suggested the importance of recording good quality data (low noise) and report that low-density EEG recordings (<32 channels) can also reveal valuable information of source activation especially in the presence of focal neural activity [64].

Source activation differences between conditions in the current study were estimated using robust statistical procedures, which included corrections for multiple comparison. In addition, significant activations were validated with additional (between-groups) comparisons, as well as with reference to corroborating findings within existing fMRI literature. Resting-state EEG data and frequency following responses were recorded in some participants but not considered for this analysis. These measurements were done after completion of the experimental procedures described here. Stimulus presentation was controlled in the Gentask module of the Compumedics Neuroscan Stim2 system. To bypass delays associated with the stimulus presentation computer and accurately measure trigger timing at the onset of the stimulus, a Cedrus StimTracker was used to generate trigger pulses at stimulus onset. All stimuli were presented binaurally via mu-metal shielded ER-3C insert earphones (Etymotic Research).

EEG: Stimuli and task

Stimuli used in this study consisted of spoken digits presented in competing background noise. Each monosyllabic digit (“One”, “Two” & “Ten”) was uttered by a New Zealand female speaker (mean F0 = 210 Hz) and recorded using an AKG HC 577 L omnidirectional headset microphone with pre-amp and Adobe Audition (AA) sound editing software (sampling rate, 44.1 kHz). From the original recording, digits were shortened to match in duration (370 ms) and were ramped on and off using a cosine function (4%). Inclusion of monosyllabic digits ensured that we could record artifact-free ERP components with adequate number of trials, without increasing the length of the recording session. The background two-talker babble was created using recordings of two female talkers reading semantically anomalous sentences [65]. The twenty sentences recorded from both speakers were concatenated and mixed into a single file. Any loud and quiet sections in the audio file were edited using the AA compression function. Both speech and noise files were normalized to the same RMS level (-22 dB full scale). The two-talker babble was played continuously for the duration of each block/condition to avoid the speech babble being time-locked with the onset (evoking ERP activity) of the target digit presentation.

Procedures for acoustically degrading the digits and noise stimuli have been presented in detail previously (see [40, 66] for details). Briefly, both speech and noise stimuli were first decomposed into 16 channels using a gammatone filterbank, with center frequencies ranging from 0.08 to 10 kHz. Then the TFS information was altered by replacing it with an envelope-modulated sinusoidal tone above a cut-off value. In the current study, we used a cut-off value of 0.11 kHz, such that TFS information for all the channels below this value was preserved and at or above 0.11 kHz, it was degraded. The alteration of the natural cues in the stimuli made listening in the degraded condition less intelligible and cognitively demanding compared to the clear condition where the signals were unaltered from the original recordings.

For each block, participants performed a number comparison task (adapted from [40, 66]) while they listened to the stream of spoken-digits presented in a two-talker co-located speech masker (see Fig 2). The stimulus train consisted of three single-digit numbers (“One”, “Two” & “Ten”) presented in a pseudo-randomized sequence. Participants were required to indicate if the target number (“Two”) was smaller or bigger than the previous number (“One” or “Ten”) in the stimulus train (with a two-alternative, smaller/bigger keypress). Participants made their responses with a response pad (Cedrus) placed on their lap, using the left thumb for “smaller” and the right thumb for “bigger”. Participants were instructed to listen for the target number and respond (smaller/bigger) as quickly and accurately as possible. Cues or feedbacks were not provided during this task. The task was performed under two listening conditions: clear, where the digits were easily discriminable from the noise (~100% correct recognition) and acoustically degraded, where the digits were less discriminable from the background noise (71% correct recognition), respectively. Participants completed this task after a session of familiarization using oral instructions and practice trials with feedback.

The rationale for using this task in comparison to a simple detection task was so that it reflected processes (auditory and cognitive) involved when listening to speech in complex environments. Only three digits were included in this closed-set number comparison task so as to reduce the memory load and assess the ability to direct auditory selective attention. During piloting younger participants found the task very difficult with the addition of more numbers and required many more practice trials. The task was developed with the intention that it could be used in future studies of other populations such as children with listening difficulties and adults with mild cognitive impairment. The addition of varying background babble during the active listening task ensured that the recognition process was challenging, and also representative of a challenging everyday listening task, within the constraints of a controlled lab environment. The task difficulty is reflected in the relatively slow reaction times obtained while participants performed the task during EEG measurements (see Fig 4). Reaction times to oddball stimuli (e.g. [67]) and n-back tasks (e.g. [68]) are often reported to be faster (~500 msec).

To record early and later stages of neural processing during the active listening task, the ERPs were elicited in the context of the three-stimulus paradigm [69]. Of the three digits, digits “one” and “ten” formed the non-targets, which occurred frequently (70% probability) and the digit “two” formed the target, which occurred infrequently (15% probability). An unexpected (novel), infrequent (15%) white noise stimulus (370 ms) was also presented, randomly interspersed among non-target stimuli (see Fig 1). The novel white noise stimulus was included to investigate neural indices (e.g. P3a ERP) of distraction/novelty processing and age-related decline of inhibitory processes during the selective listening task [70]. Therefore, this paradigm allowed us to study the effects of salient, but task-irrelevant distractors (involuntary attention) while participants selectively focused on the task-relevant information i.e. number comparison (voluntary attention). The data for the novel stimulus will be discussed separately (paper in prep.). Targets did not occur consecutively or at the start of a block. For each listening condition, two blocks of 250 stimuli (non-targets = 175, targets = 37, novel = 37) were presented. The clear and degraded listening conditions were counter-balanced across participants to reduce potential test order effects. To reduce temporal predictability, stimuli were presented in a random order with an inter-trial interval sampled at a uniform distribution (2203.39 ms ±100). Before the experimental EEG recording participants were presented with a brief practice trial of the number comparison task. Total recording time was ~34 minutes (excluding breaks).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Example of task and sequence of events per trial, including time and auditory stimuli.

https://doi.org/10.1371/journal.pone.0273304.g001

EEG: Signal audibility and adjusting for task performance level

Both procedures described below were completed on the first day of testing. These adjustments helped reducing the confounding effects of inter-individual differences in stimulus-spectrum audibility and recognition performance within each group. Controlling for stimulus-spectrum audibility is especially important as it accounts for some of the variances observed in the neural response of OAs with and without hearing loss [51].

Hearing loss in some participants would reduce stimulus-spectrum audibility and hence was compensated for using custom scripts (MATLAB) based on the NAL-R prescriptive formula for hearing aid amplification (National Acoustic Laboratories-Revised, [71]). Similar spectral adjustment using the NAL-R formula has been used previously in age-related research studies [47, 72]. Audibility/spectrum adjustments were performed for 18 of the 30 participants in the OA group who had some hearing levels of 20 dB HL or more. Signals (digits and babble) were amplified according to the participants’ audiometric thresholds (250–6 kHz). The digits were always presented at 75 dB SPL (for both groups), and this was calibrated (RMS level) using a 2cc coupler for insert earphones (G.R.A.S 90AB calibration system). All participants reported the digits (in quiet) to be clear and audible, without any perceived discomfort following this adjustment. We did not observe any significant correlations between behavioral task performance, ERP amplitude (TFS1, SNR_71%, N1 amplitudes) and high frequency hearing thresholds in the OA group (p > 0.05). Validation of these efforts to reduce audibility effects was further provided by the behavioral performance in the clear condition where all participants (older & younger) performed the task with ~ 100 recognition.

For each condition (clear & degraded), participants in the two groups were tested at the same performance level specific for that condition, ~100% recognition for clear and 71% for degraded speech. Performance scores were matched across all participants to ensure that results were not influenced by differences in task difficulty. This was achieved by first measuring the signal-to-noise ratio (SNR) that yielded a 71% correct score, i.e. SNR_71%, in the degraded condition and then using this level to obtain ceiling-level performance (~100% correct) in the clear condition (SNR_degraded + 2 dB). Specifically, each trial started at an ‘easy’ 8–10 dB SNR (10 dB SNR for OA), and noise level was increased following two correct responses and decreased following each incorrect response (1 dB step size). Participants completed 30 trials in total. The arithmetic mean of the last six reversals was calculated to estimate the SNR yielding a 71% correct score [57]. Two threshold criteria, ~100% recognition for clear and 71% for degraded speech, were calculated for the participants in both groups. The SNR to achieve 71% in the degraded condition was also used as an individual outcome measure of speech understanding in adverse listening conditions (see Fig 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Behavioral measures.

a. Pure-tone audiograms averaged between ears (L/R) for the Older and Younger participants. b,c. Performance on the attention network test (ANT): (c) Orienting network, (d) Executive control network. d. Sensitivity to TFS cues, using center frequency = 2.2 kHz in both groups. Poor performance on this task indicated poor temporal processing abilities. e. SNR_71% in the two groups measured in the digits-in-noise recognition task for the degraded listening condition. Poor performance on this task indicates poor visual sectional abilities. * p < 0.05 ** p < 0.01.

https://doi.org/10.1371/journal.pone.0273304.g002

EEG data pre- processing and ERP analysis

Offline processing of the EEG data was done using the CURRY 8 Neuroimaging Suite (Compumedics Neuroscan). Continuous EEG data files for each block were first concatenated into one file and then down-sampled to 250 Hz. Data were filtered (0.1–30 Hz) using a 2^nd order Butterworth IIR filter with zero-phase lag. Next, a threshold artefact rejection was applied (±150 μV) to remove any noisy segments. Independent component analysis (ICA), as implemented in EEGLAB (RUNICA, [73]), was used to remove ocular artifacts. The total number of removed artifactual ICA components per participant was on average 1.9 (Range 0–3) components for the EEG data. After the pre-processing, data were re-referenced to the average of all the channels (common average reference, CAR). Datasets were then segmented into 1100 ms epochs, starting 200 ms before each stimulus-onset. Data were baseline corrected relative to the 200 ms time window preceding stimulus-onset. Only epochs having the SNR thresholds (0.5–1.5 as implemented in CURRY) were included for averaging. In addition, only trials with correct responses to targets and without false positive responses to non-targets were included in the final averaging. Total trials included for averaging were similar across groups, listening conditions, and time-windows that were analyzed. For young and older adults, respectively, average numbers of accepted trials were: Clear: non-target trials: M 301.5 and M 227.2; target trials: M 55.8 and M 44.4; Degraded: non-target trials: M 292.5 and M 239.7; target trials: M 51.2 and M 38.2).

ERP topographical analyses

Differences in topographies and underlying neural processes between the listening conditions were statistically tested using a permutation‐based nonparametric topographic analysis of variance (TANOVA, [74]). TANOVA identifies latencies of significantly different map topographies that are independent of the ERP magnitude. TANOVA is a global, unbiased statistical method that is not dependent on the choice of reference electrode or on predefined electrodes of interest or time-points to test differences in conditions. TANOVAs were performed using CURRY 8, which allows for within/across group comparisons to determine significant differences between conditions, latencies and, source analysis results. The analysis was performed on the non-target and target epochs separately with listening condition (clear & degraded) as a within-subject factor. This approach enabled us to investigate modulations of early and/or later stages of auditory processing for this acoustic manipulation of the stimuli. The multiple-comparison-corrected significance level was calculated based on the alpha level (0.05), the sampling rate of the data, and the low pass filter frequency [75]. This yielded a new alpha level value of 0.012 and randomization value of 4087, which further reduced false positive results. Only effects lasting for 20 msec or longer were considered significant. This approach has been shown to control for multiple comparisons and reduce false positives in mean global field power analyses [52, 76].

Mean Global Field Power (MGFP)

The MGFP was calculated as the standard deviation of the amplitude data across all the EEG channels, reflecting synchronous activity at each time-point [77]. Similar to TANOVA, MGFP analysis helps summarize global brain activity and prevents the need to apply multiple testing correction on a subset of a priori electrode-sites that results in insufficient power. MGFP values were computed for each participant and each condition separately. Nonparametric randomization statistics for within-subject comparisons were performed using the Statistical Toolbox for Electrical Neuroimaging (STEN) developed by Jean-François Knebel (http://www.unil.ch/line/home/menuinst/about-the-line/software–analysis-tools.html). Statistical significance was obtained using 4000 permutations and an effect was considered reliable only for intervals where >20 ms contiguous significance was achieved (p ≤ 0.01).

ERP source reconstruction

Information about the neural generators contributing to differences between conditions were estimated using the current density reconstruction (CDR) algorithm in CURRY 8 (Compumedics Neuroscan). First, an in-built head model based on the boundary element head method (BEM) was computed for each participant [78]. CDR was done using the standardized Low Resolution Brain Electromagnetic Tomography (sLORETA) method [79] to obtain current density images for the conditions compared. We only looked at current distribution that comprised the cortical grey matter and hippocampus in the MNI-referenced (Montreal Neurological Institute) brain. sLORETA estimates the underlying sources under the assumption that neighboring cortical regions should have synchronized and highly correlated neural activity. We used sLORETA because it has previously been shown to estimate (blurred) sources with minimal error when comparing it to fMRI-activations using an auditory oddball paradigm [80]. To test for significant differences in the source activation between conditions, CDR statistical nonparametric mapping (SnPM) was used [75]. This is a permutation or randomization test on the three-dimensional sLORETA images. As CDR SnPm extracts locations of significant source activation differences at every time-point, a correction for multiple testing was applied based on the spectral properties of the EEG data as described above. The resulting alpha level (0.012) and randomization value (4087) helped further reduce localization errors. Results of source localization analyses described in the next section include only that of the later time-window, as our initial analyses did not reveal any significant differences in source activation across conditions for either groups in the earlier time-windows of non-target stimulus processing.

Effects of aging on behavioral measures

Fig 1 summarizes the results of all behavioral measures that were compared between the two groups. As the data were not normally distributed (Shapiro–Wilk Test, p < 0.05), non-parametric statistics (Mann–Whitney U-test) were applied for the analyses. The effect sizes were also computed using the following formula r = Z / √N [81]. For the TFS1 test, the OA group (Mean 24.6 Hz, SD 13.3) showed significantly poorer temporal processing abilities than YAs (Mean 17.8 Hz, SD 9.4), however, the effect sizes were small (U = 399.5, p = .049, r = 0.27). The OAs also exhibited significantly less efficient attention functions of orienting (U = 161.0, p = 0.006, r = 0.38) and executive control (U = 132.0, p = 0.001, r = 0.46) compared to the YAs, suggesting poorer selective attentional abilities (see Fig 2). The limitations imposed by poor temporal processing and selective attentional abilities in the OA group were also reflected in their performance on the digits-in-noise task. Noticeably, the OAs required higher SNRs (M 2.3 dB, SD 1.8) to achieve 71% correct performance compared to YAs (M -1.4 dB, SD 1.6). This difference was significant with a large effect size (U = 571.5, p < 0.001, r = 0.76). At the individual level, only five participants in the OA group had scores ≤ 0 dB SNR_71%.

To determine whether auditory (TFS1) and/or cognitive (ANT) abilities were predictive of their SNR_71% scores, a stepwise multilinear regression analysis was conducted separately for the older and younger adults [21]. To reduce the risks of overfitting and multiple testing errors associated with regression analyses, we performed a Bonferroni adjustment for the analyses (α = 0.025). Since the ANT scores for the orienting and executive control functions, measured on the same scale (msec), were correlated (r = -0.384, p = 0.006), the regression analysis was performed using the average of these two independent variables. For the regression analyses, the composite ANT and TFS1 test scores served as the predictors and the SNR_71% scores served as the dependent variable. For the OA group, the analysis indicated that SNR_71% scores were best predicted by a model only with the TFS1 scores (R² = 0.17, F_{(1, 28)} = 6.02, p _(corrected) = 0.021). A scatter plot showing this association is shown in Fig 3. For the YA group, however, none of the variables significantly predicted the speech recognition scores (p = 0.048). Full model results for each group are shown in Tables 1 and 2 included in the S1 File. It is important to note that 18 participants in the OA group had age-related hearing loss and stimulus-audibility in these participants was individually compensated for using the NAL-R threshold correction. However, to ensure that addition of OAs with hearing loss did not affect the results of this analysis, we conducted a hierarchical regression analysis to investigate how much of the variance in the SNR_71% scores was accounted for by the hearing thresholds (250–8 kHz, averaged between ears) in OAs (see Table 3 in S1 File). The analysis revealed that, after removing the hearing thresholds, predictors (TFS1, ANT) in Model 2 accounted for significant variation (R² = 0.19, F_{(2, 27)} = 3.35, p = 0.05). Hearing thresholds accounted for just 0.6% of variance in the SNR_71% scores. The most important predictor of SNR_71% scores was TFS1 scores, which explained 18% of the variance. This indicates that hearing loss may not have contributed to the entirety of the variation in the SNR_71% scores observed here.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Scatter plots showing relationship between performance on the TFS1 task and the SNR_71% in older and younger participants.

https://doi.org/10.1371/journal.pone.0273304.g003

Early stages of auditory processing (non-target stimulus processing)

Grand average ERP waveforms and corresponding scalp topographies (averaged across listening conditions) for the older and younger adults are shown in Fig 4. Overall, non-target stimulus processing elicited a conventional P1-N1-P2 complex with the P1 component occurring at ~ 65 msec (Range 53–90 msec), N1 at ~202 msec (Range 165–296 msec), and P2 component at ~ 490 msec (Range 375–540 msec) (based on visual inspection of grand averaged waveforms). These latency intervals for the P1-N1 components are consistent with previous reports on ERPs recorded to speech stimuli in noise (e.g. [82]). The P2 component, however, appears later and temporally overlaps with the P3 component. This can be attributed to the paradigm/task used here, because participants had to make an internal decision (smaller/bigger) about the correct response, each time they heard the non-target number (“1” or “10”). Each response was thus prepared and withheld until the onset of the target number (“2”) and hence it is plausible that P3 overlaps with the P2 component during non-target stimulus processing. Time windows for TANOVA and MGFP analyses encompassed a broad latency range between 0 and 600 msec, which included the latencies of the early P1-N1-P2 complex based on the grand averaged waveform (see Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Grand-averaged mean global field power (MGFP) waveforms for the target and non-target stimulus processing for the older and younger participants.

The corresponding scalp maps for the major ERP components; N1, P2 (Non-target) & P3 (Target), are displayed with positive values in red and negative values in blue. The dashed line shows the mean reaction time (averaged across conditions) for each group.

https://doi.org/10.1371/journal.pone.0273304.g004

Topographical analyses.

TANOVA analysis comparing listening conditions was applied to the ERPs time-locked to the onset of the non-target stimulus from 0 to 600 msec. In the OA group, TANOVA revealed significant topographical dissimilarities in the latency ranges of 140.0 to 176.0 msec, 220.0 to 292.0 msec, and 316.0 to 368.0 msec, between the clear and degraded conditions (p < 0.001). From Fig 5 it is evident that differences in the listening conditions were most evident in the earlier, N1 latency range for the OAs. N1 (interval) topography showed largest negativity in the fronto-central electrodes with maximum activity over the left hemisphere. The topographies were stronger for the clear compared to the degraded listening condition. TANOVA in the YA group however, revealed significant differences in the topographies in the later intervals following signal onset between 312.0 to 324.0 msec, 332.0 to 384.0 msec, and 468.0 to 484.0 msec, (p < 0.001). Differences in the topographies between the listening conditions fell within the P2 range for the YAs. The P2 (interval) topography showed largest positivity in the centro-parietal electrodes with activity localized over the left hemisphere. The topographies were stronger for the clear compared to the degraded listening condition.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Topographical comparison of the non-target stimulus processing (epoch = 0–600 msec) for the older and younger participants: The top panel shows the grand-averaged mean global field power (MGFP) waveforms for the clear and degraded listening conditions in each group.

Timeframes in the shaded region indicates periods of statistically significant differences (p < 0.01) between listening conditions. The bottom panel shows the corresponding topographical maps that are linked to the shaded timeframes of the MGFP for each listening condition.

https://doi.org/10.1371/journal.pone.0273304.g005

Later stages of auditory processing (target stimulus processing)

Time windows for TANOVA and MGFP analyses encompassed a broad range between 100 and 800 msec, including the late P3 component based on the grand averaged waveform (see Fig 6).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Topographical comparison of the target stimulus processing (epoch = 100–800 msec) for the older and younger participants: Top panel shows the grand-averaged mean global field power (MGFP) waveforms for the clear and degraded listening conditions in each group.

Timeframes in the shaded region indicates periods of statistically significant differences (p < 0.01) between listening conditions. Bottom panel shows the corresponding topographical maps that are linked to the shaded timeframes of the MGFP for each listening condition.

https://doi.org/10.1371/journal.pone.0273304.g006

Source localization analyses.

In general, source activation averaged across both listening conditions showed distributed activation patterns in the frontal, temporal, parietal, and limbic networks for both groups (> baseline [-200–0 msec], p < 0.01). Differences in brain activation were only computed for the time interval corresponding to significant effects of listening condition in the TANOVA analyses (distribution of scalp activity). Hence, statistical analyses of source estimations were computed only for the OA group (clear vs. degraded) in the time intervals between 350.0 and 764.0 msec. This interval showed significant differences in scalp distribution and comprised maximum P3 activity (see Fig 6). Results indicated significant differences in the ERP source activity in the frontal cortex (p < 0.01) of the left hemisphere with peak activity (degraded > clear) in the regions of middle frontal gyrus (MFG; Brodmann area (BA) 8, 9) followed by precentral gyrus (BA 6) (see Fig 7). To validate this source difference between conditions, additional sLORETA analysis for the P3 interval was computed between the two groups (YA vs. OA) for the degraded condition. The test for statistical significance (CDR SnPm) and multiple testing adjustments followed the same procedure as describer earlier. As expected, result of this analysis also showed an activation pattern of the prefrontal cortex (left MFG; BA 8, 9) during degraded listening in the OA group (p < 0.01, see Fig 7).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7.

Top panel: a,b. Displays results of the current density reconstruction statistical nonparametric mapping (CDR SnPM) testing. The input data for the CDR SnPM were the standardized Low Resolution Brain Electromagnetic Tomography (sLORETA) source images for the conditions compared. Source analyses revealed that Older participants recruited the left prefrontal cortex during degraded listening. Bottom panel: c. Scatter plot describing a significant positive correlation between scores on the digits-in-noise recognition task (SNR_71%) and the left middle-frontal-gyrus activation in the Older participants (r = 0.42, p = 0.01). d. Displays differences in P3 amplitudes (Pz minus Fz electrodes) across younger (n = 20), good (n = 15) and poor (n = 15) older participants.

https://doi.org/10.1371/journal.pone.0273304.g007

Electrophysiological differences at earlier latencies

Despite adjusting for frequency-specific stimulus audibility, OAs showed differences in early sensory processing corresponding to the time-window of the N1 potential. This may be indicative of deficits in supra-threshold auditory processing rather than hearing acuity that is commonly associated with aging [87]. Aging effects observed in the current study for early (~140 ms) stages of neural processing are consistent with reports of age-related neural degeneration beginning at the level of the auditory nerve fibers (in animal studies, [88]) and reduced neural precision at the midbrain level [20, 29]. Primary afferent deficits observed in older experimental animal studies appear to involve high-threshold neuronal fibers, which determine performance on supra-threshold auditory tasks such as detection of signals in noise and speech coding, rather than sound detection. Early neural processing deficits such as this may account for some of the age-related effects on cortical ERPs and behavioral performance observed in the OAs of the current study. Age-related changes in the early evoked auditory potentials may also be mediated by age differences in the involvement of the PFC in attentional control [89]. Obleser and Kotz [90] suggested that age-effects on neural processing in the N1 time-window, as observed in the OAs in the current study for clear vs. degraded speech, may be reflective of increased ‘neural effort’ via resource-allocation required to decode degraded speech. Differences observed in the N1 time-window, can be interpreted as a change in cognitive strategy [53], as the listening demand increases (clear–degraded).

In contrast to OAs, YAs only showed topographical differences corresponding to the time-window of the P2 component between the two listening conditions. This might be linked to the role of the P2 time-window in allocation of attention to the salient stimulus during target detection [91, 92]. For instance, Getzmann et al. [92] showed P2-peak changes when participants focused their attention and when the target speaker was salient during a spatial listening task. The topographical differences in P2 time window observed here between clear and degraded speech for YAs but not OAs might therefore reflect the ability of YAs to selectively attend to task-relevant information (digits) and ignore the distractors (babble).

The time-wise MGFP analyses did not show any significant differences between the two conditions, differences were only evident between groups. This indicates that task difficulty and stimulus clarity did not modulate overall neural response strength, as indicated by MGFP. This finding was somewhat unexpected considering the perceptual difference between the clear and degraded conditions. Although visual inspection of the grand mean waveforms indicated larger differences at the fronto-central vertex electrodes (see also MGFP waveforms in Fig 5), these amplitude differences were not statistically significant at the MGFP level. It may also be that ERP amplitudes are not as sensitive to manipulations of the acoustic TFS in the stimuli as they are with envelope manipulations. The MGFP results were however consistent with the results of the topographical analyses which showed differences between OA and YA in the early time window. OAs had smaller overall neural response strength during the N1 time-window (with similar mean P2 amplitudes) compared to the younger participants (averaged between the two [clear and degraded] conditions, see Fig 4). This suggests that, as signal resolution (digits in noise) was reduced, at early stages of sensory processing (N1 time window) OAs allocated more of their cognitive resources (P2 time window) to process the unresolved input signal and to facilitate post-perceptual processing (P3 time window). YAs however, had a more accurate representation of the incoming bottom-up signal for both clear and degraded conditions, which enabled (early) top-down selection of salient features in the resolved signal before (later) post-perceptual processing had occurred [32]. This may be why the YAs outperformed the OAs in the number comparison task (SNR_71%), and why YA vs. OA differences were evident across a range of latency windows. These results highlight the possibility of earlier processing deficits in OAs affecting subsequent higher-level functions of stimulus selection and response execution. Previous studies using a-priori selected electrode/s have shown increased N1 amplitudes with increasing task difficulty/speech degradation (e.g. [90]), and increased (e.g. [93]) or no changes (e.g. [94]) in N1 amplitude with aging. Aside from reporting global neural activity (MGFP & TANOVA analyses) in the current study, differences across studies can arise from the choice of reference electrode [52] as well as the stimuli and task.

Previous studies have demonstrated compensatory allocation of resources at early stages of neural processing (as indexed by increased P2 amplitude) in adults with hearing loss (mild–moderate), compared to age-matched normal-hearing controls during passive listening [95, 96]. Roque et al. [97] showed aging effects only in the P1 time window (earlier latency values), using multichannel analysis techniques and after controlling for stimulus audibility effects. The authors attributed the earlier time-window effects to the nature of the task used in the study. Our results extend these findings to active listening, controlling for confounding effects of stimulus-spectrum audibility and performance levels when assessing changes associated with neuroplasticity in OAs with a range of supra-threshold auditory processing and selective attentional abilities [51]. Unlike the Roque et al. study, where ERPs were recorded during passive listening (words in quiet), our participants performed a number comparison task in the presence of a two-talker babble. Such active listening tasks, requiring focus and response, may highlight auditory processing effects on later ERP components/time-windows (>90 msec) [98–100]. Billings et al. [99] examined P1, N1, and P2 for passive vs. active paradigms and showed attention-related effects on N1-amplitude only when listening to tones presented in noise, not in quiet. Recent fMRI results show similar brain activation patterns in younger and older individuals during a word recognition in quiet task [101]. Tasks such as that used in the current study which used competing speech babble, degraded speech acoustics and, an active listening task are more likely to elicit ERP differences that reflect aging effects on both earlier and later auditory neural processing.

Electrophysiological differences at later latencies

Differences in neural processing in later time windows between OAs and YAs in the current study suggest that the effects of aging on early cognitive resource allocation subsequently had impacts on later time-windows of post-perceptual stimulus processing. Later time-windows beyond 400 ms include the P3b component that is often linked to processes involving cognitive processing and decision-making [67, 102]. Only the OAs showed topographical differences, with increased frontal distribution, as the listening demand increased. This is consistent with the suggestion that OAs tend to recruit more cognitive resources to address processing difficulties involving degraded bottom-up input signal [11, 103]. For instance, Wong et al. [103] showed that, when OAs listened to speech in the presence of noise, they exhibited reduced fMRI activation in the auditory cortex but increased activation in the cognitive regions of the prefrontal cortex and precuneus. However, despite this increased activation, OAs are still at a disadvantage relative to YAs as behavioral performance may be affected by reduced availability of cognitive resources [104]. This is consistent with the poorer scores obtained by the OAs for the behavioral tasks (see Fig 1) and the generally reduced ERP activity (MGFP, see Fig 4) compared to the YAs in the current study. Pergher et al. [68] similarly showed a decrease in P3 amplitude for OAs compared to YAs during a visual N-back task. They also found that P3 amplitude and alpha band activity reduced with increasing task difficulty in younger participants and alpha band activity increased in older participants due to mental fatigue. Whether the topographical differences observed in the OAs in the current study are related to fatigue, will need further investigation.

Prefrontal source activation during degraded listening

The increased frontal distribution and enhanced frontal P3 amplitudes observed for the degraded condition might be a strategy adopted by the OAs to match the level of cognitive control demanded by the task [105]. The increased frontal distribution is also supported by the results of the source estimation analyses that showed activation of the left MFG for the degraded condition. There is complementary evidence for this from an fMRI study which also showed activation in this region when participants listened to sentences with increasing levels of acoustic degradation [106]. Recent met-analyses of fMRI studies have also suggested the involvement of areas (fronto-parietal and insula regions) beyond the auditory cortex during effortful speech perception [107]. The meta-analyses also revealed that patterns of activation differed depending on the effortful listening condition (spectral degradation vs speech in noise). A combined ERP/fMRI study showed activation in the right MFG during visual-target-processing (P3b) for older compared to younger participants [67]; the difference in MFG laterality compared to the current study may be attributable to task differences such as the auditory and linguistic demands of the task used in the current study. Although literature (see meta-analyses [107] shows some variability in the exact location of activation, activity in the left prefrontal regions has been associated with resolving degraded speech input (e.g. [108]).

The current study demonstrated an association between left prefrontal cortex activation and poorer speech recognition abilities in OAs. This pattern is consistent with previous research that has shown age-related enhancement of frontal P3 ERPs associated with poorer performance on neuropsychological tests of frontal lobe function [109]. Similarly, Reuter et al. [49] showed that a subset of OAs with increased frontal P3 distribution exhibited poorer performance on a visual selective attention task. The authors suggested that such parietal vs. frontal ERP distribution helped differentiate subgroups of OAs with less efficient processing. Our study extends these findings using source analyses and shows exploratory evidence (moderate correlation) for inefficient processing with increasing listening demands through the recruitment of the left MFG in OAs with poorer scores on the listening task. It is important to note, however, that individual differences in performance and recruitment of the prefrontal cortex have been linked to brain maintenance (or inefficient processing) rather than compensatory processes associated with aging [110, 111]. In particular, Morcom and Henson [111] showed that increased frontal recruitment (via fMRI) was task dependent and that it did not necessarily facilitate (or improve) performance with increasing age. This finding challenges existing compensatory theories (e.g. PASA) that suggest a generalized recruitment of prefrontal cortex to maintain optimal performance in all older individuals.

The results observed here are also consistent with recent longitudinal studies where only participants with declining cognitive functions showed increasing prefrontal (including left MFG) activation (baseline < follow-up) compared to those with stable performance on behavioral tasks [112, 113]. The observed prefrontal activation in the current study, hence, cannot be generalized as a typical finding in the aging population. Rather, this occurred only during the most effortful listening condition (degraded) and in poorer performers who required higher SNRs to process degraded speech. In addition, we also showed supporting evidence for increasing posterior to anterior shift in poorer-performing OAs compared to high-performing OAs and YAs (see Fig 7(d)), thus, it is highly likely that our sample included a mixture of high-functioning older individuals as well as those with poorer cognitive function. Also, OAs in the poorer-performing group had poorer TFS1 and ANT-composite scores (M_TFS1 30.8, M_ANT 39.7) compared to the good-performing OAs (M_TFS1 18.4, M_ANT 42.2). Two previous studies that subdivided older participants based on their memory performance relative to IQ scores, showed similar evidence for more frontal activity (P3 time window) in poor performing OAs compared to high performers and younger adults [114, 115]. Hence, the frontal distribution in the P3 time-window, especially for the poor performing OAs in this study might be associated with less efficient recruitment of neural resources during task performance [109, 111]. Differences in network efficiency could be investigated using graph theoretical methods in future research (e.g. [48]), comparing OAs with good and poor performance to determine whether increased network efficiency in OAs results in task performance that is comparable to younger adults.