Poorer Speech Reception Threshold in Noise Is Associated With Lower Brain Volume in Auditory and Cognitive Processing Regions.

Purpose Hearing loss is associated with changes in brain volume in regions supporting auditory and cognitive processing. The purpose of this study was to determine whether there is a systematic association between hearing ability and brain volume in cross-sectional data from a large nonclinical cohort of middle-aged adults available from the UK Biobank Resource ( http://www.ukbiobank.ac.uk ). Method We performed a set of regression analyses to determine the association between speech reception threshold in noise (SRTn) and global brain volume as well as predefined regions of interest (ROIs) based on T1-weighted structural images, controlling for hearing-related comorbidities and cognition as well as demographic factors. In a 2nd set of analyses, we additionally controlled for hearing aid (HA) use. We predicted statistically significant associations globally and in ROIs including auditory and cognitive processing regions, possibly modulated by HA use. Results Whole-brain gray matter volume was significantly lower for individuals with poorer SRTn. Furthermore, the volume of 9 predicted ROIs including both auditory and cognitive processing regions was lower for individuals with poorer SRTn. The greatest percentage difference (-0.57%) in ROI volume relating to a 1 SD worsening of SRTn was found in the left superior temporal gyrus. HA use did not substantially modulate the pattern of association between brain volume and SRTn. Conclusions In a large middle-aged nonclinical population, poorer hearing ability is associated with lower brain volume globally as well as in cortical and subcortical regions involved in auditory and cognitive processing, but there was no conclusive evidence that this effect is moderated by HA use. This pattern of results supports the notion that poor hearing leads to reduced volume in brain regions recruited during speech understanding under challenging conditions. These findings should be tested in future longitudinal, experimental studies. Supplemental Material https://doi.org/10.23641/asha.7949357.


Introduction
Differences in sensory input to the brain cause changes in its functional and structural organization (Merabet & Pascual-Leone, 2010). It is well-established that congenital deafness drives cross-modal plasticity in auditory processing regions participants with normal hearing (mean age 48). Whole brain analysis showed that the hearing impairment group had lower grey matter volume in regions associated with cognitive control including the right anterior cingulate as well as the medial frontal gyrus bilaterally. ROI analysis showed that the hearing impairment group had lower grey matter volume in STG bilaterally than the group with normal hearing. The authors' interpretation was that hearing loss causes grey matter reduction in both auditory and higher cognitive processing regions. Thus, generally, the literature demonstates that acquired hearing loss is associated with smaller brain volume in both auditory and cognitive processing regions, and suggests that hearing loss probably causes neural atrophy in these regions.
However, one study, Boyen, Langers, de Kleine & van Dijk (2013) showed greater volume in STG and middle temporal gyrus (MTG) associated with hearing loss. Like the other cross-sectional studies reviewed here, this study was based on T1weighted magnetic resonance (MR) images. Specifically, they were collected from 31 participants with hearing impairment and tinnitus (mean age 56 years), 16 participants with hearing impairment but no tinnitus (mean age 63 year) and 24 participants with normal hearing (mean age 58 years). Whole brain analysis showed that both hearing impairment groups compared to controls had larger grey matter volume in STG. The authors' interpretation of this finding was that brain volume increase associated with hearing impairment may be related to the role of the STG in semantic memory. In particular, they suggested that because individuals with hearing impairment miss part of the speech signal they may rely more heavily on semantic memory to maintain normal communication.
This explanation is in line with findings in the literature showing that cognitive training leads to increases in the volume of brain structures involved in cognitive processing (for a review see Lövdén, Wenger, Mårtensson, Lindenberger & Bäckman, 2013).
The most common treatment for hearing impairment is hearing aids (HA) which amplify weak sounds, making them more audibile. It seems reasonable to assume that this may lead to enhanced neural representation of speech and consequently preservation of neural integrity in PAC and other auditory processing regions.
Indeed, it has been shown that HA intervention leads to enhanced cortical In the present study, we investigated the relationship between hearing ability and brain volume in a much larger set of data from a non-clinical cohort of middle-aged adults available from the UK Biobank Resource. The hearing measure in this data set is speech reception threshold in noise (SRTn) indexed by the Digit Triplets Test . We predicted that poorer SRTn would be associated with lower brain volume globally as well as in PAC and cognitive processing regions outside STG involved in the processing of speech in noise. We also predicted an association in STG although we did not specify its direction owing to contradictory results reported in the literature. We also expected that these predicted associations would be modulated by HA use.

Methods
The present study is based on cross-sectional data available from the UK Biobank Resource, a prospective study including lifestyle, physical and cognitive measures for over 500 000 participants who were aged 40-69 years when they were recruited

T1 data acquisition
According to UK Biobank Brain Imaging Documentation (Smith, Almagro & Miller, 2017), the T1 data were collected using a dedicated standard Siemens Skyra 3T running VD13A SP4, with a standard Siemens 32-channel RF receive head coil. The field-of-view was automatically determined based on Siemens' auto-align software or failing that, set by the radiographer. The T1 structurals were acquired using straight sagittal orientation and a 3D MPRAGE sequence. T1 scanning lasted 5 minutes and was part of a longer series of scans lasting 35 minutes. Resolution is 1x1x1 mm and field-of-view is a 208x256x256 matrix. Standard Siemens on-scanner conversion of complex multi-coil data was carried out for the T1 data.
2.3 T1 data processing T1 data processing took place as follows (Smith, Almagro & Miller, 2017). The full field of view was cut down to reduce the amount of non-brain tissue and gradient distortion correction was applied in conjunction with a standard-space T1 template.
This was achieved using Brain Extraction Tool (Smith, 2002)  The data were then nonlinearly warped to MNI152 space. As Alfaro-Almagro et al., (2018) have pointed out, this is a critical processing step in the pipeline, and because T1 images in UK Biobank had brighter internal carotid arteries than those in the MNI152 template a custom reference brain mask was applied to exclude this part of the image when estimating the transformation. A standard-space brain mask was then back-transformed into the space of the T1 and applied to the T1 image to generate a brain-extracted T1. Next, tissue-type segmentation was applied to generate a fully bias-field-corrected version of the brain-extracted T1. The external surface of the skull was estimated from the T1, and used to normalise brain tissue volumes for head size.
A total of 139 regional grey matter volumes were generated by summing the grey matter partial volume estimates within 139 regions of interest (ROIs) defined in MNI152 space using the Harvard-Oxford cortical and subcortical atlases Mean SRTn among the participants in the present study was -6 dB, SD = 1.6 dB. A higher SRTn indicates poorer hearing and normal hearing is considered to be represented by SRTn < -5.5 dB (Dawes, Fortnum et al., 2014). Thus, 61% of the participants in the present study had an SRTn in the normal range.

Cognitive ability
The cognitive abilities such as reasoning and memory that are engaged to achieve Further, hearing impairment is associated with increased risk of incident dementia, ie. a pathological reduction in cognitive ability (Deal et al., 2017), which in turn is related to reduced brain volume (Henneman et al., 2009). Thus, both SRTn and brain volume are likely to be related to cognitive ability, making it important to control for cognitive ability when investigating the association between SRTn and brain volume. The UK Biobank includes data from a test of Verbal Reasoning (VR) which taps into the cognitive functions that support speech processing under challenging conditions. We have previously shown an association between worse VR score and worse SRTn (Keidser, Rudner, Seeto, Hygge & Rönnberg, 2016).
The VR data were collected during a separate visit to a regional assessment center using a self-administered program running on a touch screen computer. VR was measured using 13 multiple choice questions of which as many as possible were to be answered in two minutes. Examples of questions are: 1) "Bud is to flower as child is to?" with the answer to be selected from: Grow, Develop, Improve, Adult and Old; and 2) "If sixty is more than half of seventy-five, multiply twenty-three by three. If not subtract 15 from eighty-five. Is the answer?" with the answer to be selected from: 68, 69, 70, 71, 72 and with the additional options of "Do not know" and "Prefer not to answer" in both cases. These questions tap into semantic LTM and at the same time require logical reasoning. For more details see Keidser et al. (2016). The mean VR score of participants in the present study was 6.9, SD = 2.1.

Vascular and vision problems
Poor hearing is associated with vascular and vision problems. Vascular problems are assumed to lead to poor hearing due to a reduced blood supply to the cochlea, which results in a disruption of the chemical balance of the inner ear that affects the electrical activity of the hair cells.

Statistical analysis
The data were analysed using two sets of regression models for whole brain and each of the 154 predefined brain regions. In model sets 1 and 2, the dependent variable was the normalized volume (in mm 3 ) of one of the predefined brain regions and the independent variables included: age, age 2 , gender, TDI, VR, vascular problems, vision problems and SRTn. In model set 1, these were the only independent variables, see appendix A. In model set 2, HA use was also entered as an independent variable, see appendix B.
Examination of intercorrelations among independent variables revealed the expected association between SRTn and age (r = .26, p < .001) and SRTn and HA use (r = .18, p < .001) as well as between age and HA use (r = .18, p < .001). There were also associations between age and vascular problems (r = .23, p < .001) and age and vision problems (r = .25, p < .001). Further, there was an association between vascular problems and gender (r = .13, p < .001) indicating that vascular problems were more prevalent in men than women. None of the other intercorrelations had a coefficient exceeding .1.
In all models, gender, vascular and vision problems as well as HA use were entered as dichotomous variables while the remaining variables were continuous. Preliminary analyses conducted on the full set of data included squared terms for all independent variables as linearity was not assumed. The only squared term that significantly improved model fit across many regions was the age-squared term. Thus, the agesquared term was retained in the reported regression models, while the squared terms relating to the other variables were excluded to simplify the models.
We predicted an association between SRTn and brain volume in 48 regions. We found a significant association in nine of those regions with model 1 (see Table 1) and in all but one (right insular cortex) of these nine regions in model 2. All significant associations indicated that poorer SRTn was related to smaller regional brain volume. No significant associations were found in any of the other defined regions with either model sets 1 or 2.

Auditory processing regions
In accordance with our prediction, model set 1 showed that greater SRTn (i.e. poorer hearing) was associated with smaller grey matter volume in auditory processing regions including the left STG within both anterior and posterior divisions (see Figure   1) as well as right STG within the posterior division, see Table 1  When HA use was controlled for in model set 2, associations between SRTn and grey matter volume in these three regions all remained significant, and in addition, we found that HA use was associated with smaller grey matter volume in the left thalamus (p = .0084, volume change = -42.41 mm3 or 1.22 % of mean volume 3476 mm3).
Contrary to previous studies, we did not find any significant association between regional brain volume and SRTn in primary auditory cortex (PAC, Eckert et al., 2012;Peelle et al., 2011) or the subcortical auditory pathway, see Appendix A.

Cognitive processing regions
We predicted that greater SRTn (poorer hearing) was related to lower brain volume in a number of regions known to be engaged during comprehension of degraded speech. In particular, we predicted lower brain volume in frontal, superior parietal and medial temporal regions.

Frontal regions
Lower grey matter volume in the right insular cortex, right middle frontal gyrus, the left precentral gyrus, right frontal medial cortex and the left paracingulate gyrus was associated with poorer SRTn with model set 1. In model set 2 (controlling for HA use), the association between right insular cortex volume and SRTn was no longer significant while the effects of age and gender remained significant.

Superior parietal regions
There were no significant associations between SRTn and the volume of superior parietal regions.

Medial temporal region
Poorer SRTn was associated with lower regional brain volume in the right hippocampus with both models. There was no significant effect of SRTn in the left hippocampus but it is worth noting that here the effect of vascular disorder was significant along with the effects of age and gender. This region was defined according to both extraction methods, a significant association was found in the right hemisphere region only with the FIRST extraction method which does not distinguish tissue type. There was no significant association with SRTn in this region in either hemisphere defined according to the atlas based extraction method that isolates grey matter. However, there were significant effects of age and gender on hippocampus grey matter volume bilaterally as well as a significant effect of vascular disorder in the left hemisphere. Importantly, there was an effect of VR on hippocampus grey matter volume bilaterally such that poorer VR was associated with lower volume. Together these findings suggest that the association with SRTn in the right hippocampus relates to white rather than grey matter. This is in contrast to the other regional associations as well as the whole brain association with SRTn.
The change in right hippocampal volume of 19.7 mm 3 per SD SRTn with model 1 can be compared to the difference in annual rate of decrease in right hippocampal volume of 0.01 cm 3 (10 mm 3 ) relating to hearing impairment reported by Lin et al.  Other regions included in the medial temporal region are the anterior and posterior divisions of the parahippocampal gyrus. There was no significant association with SRTn in any of these regions but all of them showed effects of age and gender, the left anterior region showed a significant effect of vascular disorder, while the right anterior and the left posterior regions showed significant effects of VR.

Discussion
The present study is the first, to our knowledge, to show an association between smaller regional brain volume and functional hearing in a non-clinical cohort of middle-aged adults. Building on data from 8701 particiants in the UK Biobank, It shows that lower grey matter volume in both auditory processing regions in temporal cortex and cognitive processing regions in frontal cortex, as well as lower hippocampal volume are associated with poorer ability to recognize speech in noise. showed that white matter volume was associated with functional hearing (but not pure tone thresholds) in the older segment of the sample (age range 70 -100 years) but that in the younger segment of the sample, with an age-range (51 -69 years) more comparable to the middle-aged participants in the present study, the opposite pattern was found, i.e. an association between white matter volume and pure tone thresholds (but not functional hearing). The lack of an association between functional hearing and regional brain volume in middle-aged adults in the study by Rigters et al.
(2018) may be due to the substantially lower number of participants in the corresponding age group compared to the present study.

Whole brain
Results of the present study showed that poorer SRTn was associated with lower grey (but not white) matter volume globally. Because grey matter consists largely of synaptically dense neuropil while white matter consists largely of myelinated axons, this finding suggests that the lower volume is due to fewer synapses rather than fewer axons. The size of the effect was comparable to the reported increment in annual rate of change in whole brain volume associated with hearing impairment as measured with pure tone thresholds (Lin et al., 2014) and apparent despite significant effects of age, gender, socioeconomic status and vascular disorder. It did not change substantially when HA use was controlled for.

Auditory processing regions
Results showed that poorer SRTn was associated with smaller grey matter volume in Thus, the association between hearing ability and STG volume in the present study suggests that poorer hearing ability in a non-clinical population is associated with smaller grey matter volume in STG and that this decrement is not compensated by corresponding long-term engagement of semantic long-term memory. This notion should be tested in a longitudinal study. We propose that a bidirectional mechanism is likely to be in force. In one direction, mental agility keeps the episodic LTM encoding mechanism (supported by the Both these factors are controlled for in our analyses. However, efficacy of hearing rehabilitation is also affected by e.g. type of HA, how long the HA has been worn relative to when the hearing loss was first detected, and whether HA are fitted on one side or both; these data were not available from the UK Biobank and should be carefully controlled in future studies. HA use was significantly associated with smaller volume in one of the predicted regions, namely the left thalamus. This region is extensive and multifunctional. It forms part of the subcortical auditory pathway but also plays a key role in cognition Halassa, 2017). Importantly, we found a significant effect of VR in this region in addition to significant effects of age and gender. Although we found no hearing-related volume reduction in this region in the present study, such an association has been demonstrated in animal models (Basta, Tzschentke, Ernst, 2005), and Peelle et al., (2011) observed reduced neural activation during language processing relating to hearing loss. One interpretation of our findings is that HA use modulates the cognitive function of the thalamus during comprehension of degraded speech, leading to volume reduction. This should be investigated in further studies.

Limitations
The great advantage of the current study is that it includes data from a large cohort, allowing us to include only participants for whom complete data sets were available.
However, it also has a number of related limitations. In particular, the authors of the present study had no control over many aspects of the design including the choice of cognitive and audiometric tests or preprocessing of the imaging data. As regards cognitive and audiometric testing, availabilitity of standard tests of specific cognitive functions such as working memory and executive function as well as pure tone audiometry whould have allowed more analytical analysis of the imaging data. On the other hand, we believe that the VR task provides a good functional measure of everyday cognition in the present study while the DTT provides a well-established measure of functional hearing ability, lending the results of our study good validity.
ROIs were determined automatically based on established atlases in the present study and were thus not designed to test hypotheses specific to the study. However, the automaticity of the preprocessing pipeline does ensure good reliability.
Unfortunately, data on volume of cerebrospinal fluid within ROIs was not available to confirm the specificity of SRTn effects (c.f. Eckert et al., 2012).
Further, the information available about hearing rehabilitation was minimal.
Participants reported current HA use most of the time but there was no information on whether they used one or two aids or how long they had been using them. Also, among those who reported HA use, many performed well on the DTT, suggesting that their functional hearing ability was good. This makes it hard to interpret results relating to HA use.