While recent studies suggest an important role of higher order olfactory brain areas for basic olfactory performance, the extent to which cortical and peripheral neural markers account for separate portions of the variability in olfactory perceptual acuity is still unclear. We addressed this question by correlating voxel-based morphometry data from 90 healthy adults with olfactory performance measures. Supplementing this approach with region of interest (ROI) analyses of functionally defined olfactory cortical regions and olfactory bulb volume, we sought to disentangle the relative contribution of central and peripheral areas to behavioral variability. Whole-brain analyses revealed a significant positive correlation of gray matter volume and olfactory function scores in the right orbital sulcus. This effect was confirmed by the ROI analyses, which further indicated a significant association of the olfactory score with olfactory bulb volume. Moreover, a functional dissociation was observed, with central and peripheral mechanisms explaining different aspects of the observed behavioral variance in the olfactory subscores. In line with previous clinical studies, these data thus suggest an important role of regional gray matter volume in the right orbitofrontal cortex and olfactory bulb volume for olfactory performance in healthy individuals.
Basic research on the sense of smell has progressed substantially during the last 30 years, particularly increasing our understanding of early olfactory processing at the level of the olfactory receptor and olfactory bulb (Wilson and Mainen 2006; Lundstrom et al. 2011). In spite of these significant advances, the field at large has been surprisingly unsuccessful in identifying biological markers, be it receptor activity or individual gene expression, that explain more than a fraction of interindividual variance of olfactory perceptual performance (Menashe et al. 2007; Lundstrom et al. 2012). In other words, what distinguishes an individual with a highly differentiated sense of smell from a person with lower perceptual acuity is still largely unknown. A likely reason for this discrepancy is that human olfactory performance is determined by complex interactions between early sensory processing in the periphery, and higher order cortical integration of perceptual and cognitive input (Mainland et al. 2002; Laska et al. 2005). While it has been speculated that conceptually different perceptual skills, such as odor threshold or identification performance, would vary in the extent to which they recruit peripheral and central brain structures, addressing this question directly in human subjects has posed significant challenges. Only recent methodological advances, such as high resolution functional magnetic resonance imaging (fMRI) (Howard et al. 2009) and coordinate-based meta-analyses of neuroimaging data (Albrecht et al. 2010; Veldhuizen et al. 2011), are now opening up possibilities to apply neuroimaging procedures previously used to explore interindividual differences in other sensory modalities to olfactory research.
A common approach to explore the neural bases of behavioral performance is the linkage of psychometric assessment outcomes to quantitative measures of neural tissue integrity. Using voxel-wise gray matter volume estimates as a single measure of anatomical variability, voxel-based morphometry (VBM) has provided unique insight into systematic structural changes that explain increased auditory and visual perceptual acuity or complex motor skills (Maguire et al. 2000; Foster and Zatorre 2010; Bezzola et al. 2011). These findings suggest that associations between structural anatomy and behavioral performance can provide information about the regions of functional specialization in the human brain even within a healthy population, which are difficult to extract on the basis of functional measures alone.
In olfactory perception research, correlations with performance measures have to date only been conducted separately for central and peripheral volumetric data. Olfactory bulb volume was shown to be associated with odor identification, but not with other measures of olfactory performance, such as discrimination and detection threshold (Buschhuter et al. 2008). Significant relationships have also been suggested between areas in cortical regions associated with olfactory functioning: in particular, the volume of core olfactory areas such as the piriform cortex, orbitofrontal cortex (OFC), and the insular cortex were correlated with olfactory functions as assessed by the Threshold, Discrimination, and Identification (TDI score), a composite measure of threshold, identification, and discrimination scores (Frasnelli et al. 2010). However, as neither of these studies has addressed the relationship between peripheral and central components of the olfactory system within the same sample, it is still unclear whether these measures explain common or different sources of variance in behavioral performance.
The main goal of the present study was to dissociate the contributions of peripheral and central volumetric measures to variability in olfactory functioning within the same sample. Using the VBM methodology, a reliable tool for the assessment of olfactory loss-related changes in cortical structure (Bitter, Bruderle et al. 2010; Bitter, Gudziol et al. 2010), we first aimed to identify robust regional linear associations of cortical gray and white matter volume with olfactory functioning on a whole-brain level. As a next step, we used functional meta-analysis data to restrict our analyses to key regions of the cortical olfactory network and explored separate and overlapping contributions of the resulting gray matter values and volumetric ROI information of the olfactory bulb from the same subject group.
A total of 103 right-handed, healthy subjects (55 females) with a mean age of 37 years (SD ±17.3) participated in the study. Exclusion criteria were neurological, psychiatric, or otorhinolaryngological diseases known to be associated with olfactory dysfunctions. All subjects underwent structural MRI scanning and assessment of their olfactory functions by means of the Sniffin' Sticks test battery (see below). The acquired structural MRI images were screened for gross-anatomical anomalies, leading to the exclusion of 3 subjects. Another 3 subjects were excluded for not meeting the criteria of normosmia on the Sniffin’ Sticks according to normative data (Hummel et al. 2007). Finally, we assessed the homogeneity of our sample using the covariance visualization tool from the VBM8 toolbox (http://dbm.neuro.uni-jena.de), implemented in SPM8 (www.fil.ion.ucl.ac.uk/SPM), and excluded subjects with a covariance of 2 standard deviations below and above the mean. The final statistical sample included 90 subjects (49 females) with a mean age of 35 years (SD ±15.9). All measures were approved both by the University of Pennsylvania Institutional Review Board, where the data were analyzed, and by the Ethical Review Board of the Technical University of Dresden Medical School, the study site. All participants provided written informed consent prior to entering the study.
Individual olfactory performance was assessed adjunct to the MRI acquisition by measures of absolute odor sensitivity (threshold), cued olfactory identification, and olfactory quality discrimination using the validated Sniffin’ Sticks testing set (Kobal et al. 1996; Hummel et al. 1997).
Absolute Odor Sensitivity (Threshold)
Absolute sensitivity was assessed for phenylethyl alcohol (PEA), an odorant often used in measures of absolute sensitivity due to its low level of trigeminal irritation (Wysocki et al. 2003), using a standardized detection threshold. The detection threshold consisted of a 7 reversals, 3-alternative, forced-choice ascending staircase procedure for PEA in a 16-step binary dilution series starting from a 4% v/v concentration. Starting with the weakest concentration, the experimenter presented triplets of 2 blank pens and 1 containing the odorant in the presently tested concentration. The blindfolded subject was then asked to indicate the odor-containing pen. Reversal of the staircase was initiated when the target was either correctly identified on 2 consecutive trails or whenever it was incorrectly identified. The arithmetic mean of the 4 last reversal points was calculated as the individual's threshold score. The possible range for the threshold measure was 1–16 with higher scores indicating a greater sensitivity to PEA.
Odor Quality Discrimination
Several forms of odor discrimination tasks exist where the 2 most frequently employed are odor quality and odor intensity discrimination (see among others Lundstrom et al. (2008)). To better work as a conceptual bridge between detection threshold and odor identification performance (see below), as well as due to being the dominant discrimination task in the literature, odor quality rather than odor intensity performance was assessed. To this end, 16 individual triplets of pens, consisting of 2 pens with identical odorants and 1 with an odorant of different quality (target), were presented to the blindfolded subject. Their task was to distinguish the pen containing the odorant different in quality from the other 2 in a forced-choice task. The possible range for the odor quality discrimination measure was 0–16 with higher scores indicating better performance.
Cued Odor Identification
Olfactory identification performance was assessed with a 4 alternative forced-choice cued identification task. Sixteen odorants were presented one by one to the unmasked subject, who was asked to pick the odor label which best described the quality of the smelled odorant from a cue card listing 4 alternative odor labels. The possible range for the odor identification measure was 0–16 with higher scores indicating a better performance.
Threshold and discrimination scores were acquired for each nostril separately. The arithmetic mean of both nostrils for each test was used for all reported analyses. Additionally, the sum score of the mean threshold, mean discrimination, and Identification score was used as a global estimate of olfactory functioning.
MRI Image Acquisition
Individual T1-weighted MRI images were acquired on a 1.5 T Sonata Vision whole-body scanner (Siemens Medical Systems, Erlangen, Germany), using an 8-channel IPAT head coil and a Siemens turboflash sequence. This sequence provides axial scans with 224 slices covering the whole brain, an isotropic spatial resolution of 1 mm3, a repetition time (TR) of 2130 ms, and echo time (TE) of 3.93 ms. Two-fold oversampling was performed in the read direction to prevent aliasing. The volumetric measures of the olfactory bulb were obtained from a focused acquisition paradigm covering the anterior and middle segments of the base of the skull of using T2-weighted fast echo-planar imaging images (voxel size 2 mm3, no gap). These images were acquired in the coronal plane to minimize the impact of partial volume effects resulting from the flat transverse shape of the bulb (Yousry et al. 2000). Manual segmentation of the olfactory bulbs was performed using the AMIRA 3D visualization and modeling system (Visage Imaging, Carlsbad, United States of America). The sudden change in diameter at the beginning of the olfactory tract was used as the proximal demarcation of the olfactory bulb (Yousem et al. 1998). Volumes were obtained by planimetric manual contouring (surface in mm2), and subsequent addition of all surfaces, which were multiplied by the slice thickness to obtain the volume in cubic millimeters. Previous studies using this approach have emphasized its high reliability and accuracy (Yousem et al. 1998), and applications within our own research group (Mueller, Abolmaali et al. 2005; Mueller, Rodewald et al. 2005; Buschhuter et al. 2008) have consistently yielded high intraclass correlations both within and between observers.
MRI data analysis was performed using SPM8, implemented in Matlab 2008a (Mathworks, Sherborn, MA, United States of America). Images were first bias-corrected and partitioned into different tissue classes using the unified segmentation approach (Ashburner and Friston 2005). Deformations for image alignment were then estimated by iteratively registering the resulting images with their mean using the DARTEL algorithm (Ashburner 2007). The resulting template and flow fields were used to create spatially normalized and Jacobian-scaled gray and white matter images in MNI space with a voxel size of 1.5 × 1.5 × 1.5 mm. As our interest was to compare tissue volumes at different levels of olfactory functioning, we chose to use modulated images which preserve the total amount of signal from each region. Finally, a Gaussian smoothing kernel of 8 FWHM was applied to the images.
Statistical analysis of the data was performed within the framework of the general linear model. Dependencies between each voxel and the total intracranial volume of the subject were modeled as globals, that is the preprocessed voxel values were scaled to be proportional to the fraction of brain volume accounted for by that particular gray matter voxel. We first conducted a whole-brain analysis to identify regions where regional gray matter volume was associated with overall olfactory performance. Here, we entered the composite TDI score as the variable of interest into a whole-brain multiple regression analysis, with age included as a covariate. Voxels with gray matter values of <0.2 (absolute threshold masking) were excluded from the analyses in order to avoid edge effects between the different tissue types. Whole-brain regression analyses were also conducted for the olfactory threshold, discrimination, and identification scores separately. The same procedure was then repeated for the white matter data. For all analyses, we applied an uncorrected peak-level threshold of P < 0.001 and a cluster-level threshold of P < 0.05, using nonstationary cluster correction implemented in SPM8 to account for the nonisotropic smoothness of the VBM data. We chose to emphasize a stringent cluster-level correction (clusters of a size which unlikely occurred by chance) over a more conservative peak-level correction, assuming that any differences in gray matter volume attributable to olfactory functioning within a group of healthy controls would be rather subtle, but continuously observed throughout functionally homogeneous regions. In order to quantify individual contributions of the three assessed subaspects of olfactory functioning and their relationship with olfactory bulb volume, we extracted average voxel gray matter values from ROIs in the bilateral orbitofrontal and bilateral piriform cortex.
We chose to base our region of interest (ROI) analyses on functional anatomical data, because the olfactory cortical structures are notoriously difficult to localize based on macroanatomy or extrapolations from animal data (Goncalves Pereira et al. 2005; Gottfried and Zald 2005). The increasing number of publications reporting findings from functional neuroimaging of olfactory processes opens up the possibility to achieve reliable estimates of delineation of the olfactory cortex using activation likelihood estimation (ALE; Turkeltaub et al. 2002; Eickhoff et al. 2009), a method we used in the present study. ALE is a meta-analytical technique that estimates the convergence of activation foci from many individual fMRI studies, thus providing a reliable quantitative estimate of the probability of activation on a whole-brain level. We defined core olfactory areas by means of ALE, based on reported activations of 40 individually published olfactory fMRI studies, as a priori (ROIs) for our volumetric analyses. ROIs were created by transforming the t-map from the meta-analysis into anatomical masks, including all active voxels in the specified regions above a t-value of 50% of that of the local peak voxel. In the orbitofrontal cortex, this procedure resulted in a cluster that extended into the anterior insula; to restrict the ROI to the prefrontal cortex, we thus multiplied the functional mask with an anatomical mask based on the frontal pole and orbitofrontal cortex regions defined by the Harvard–Oxford cortical structural atlas implemented in FSL (FMRIB's Software Library, www.fmrib.ox.ac.uk/fsl). The anterior insular cortex is thought to be functionally distinct from the OFC with respect to odor processing (for an overview, see Price (2008)). We therefore excluded the anterior insular cortex from the OFC ROI to maximize the functional homogeneity of our ROIs. The resulting masks are depicted in Figure 1. Mean gray matter values (scaled proportionally by total intracranial volume) for these ROIs and olfactory bulb volumes were averaged across the two hemispheres.
All off-line analyses were conducted in the R statistical computing environment (www.R-project.org). All included data sets were normally distributed; thus, Pearson's product moment correlation analyses between the volumetric measures from the three ROIs were conducted to investigate possible associations between them. They were then submitted to multiple linear regression analyses with the behavioral scores. Analyses were conducted separately to predict the TDI, threshold, discrimination, and identification scores. Given that age can affect olfactory functioning (Larsson et al. 2000; Boesveldt et al. 2011), we also calculated separate models with subject's age modeled as a covariate to estimate the proportion of variance explained by this factor. As a result of intercorrelations between our regressors, explained variance (R2) cannot easily be broken down into shares from individual regressors using a successive omission approach. We therefore used the method of “Relative Importance” proposed by Lindeman, Merenda, and Gold, as implemented in the R package “relaimpo” (Groemping 2006). This measure assesses the contributions of individual regressors to the model, taking into account both its direct effect (i.e., the direct correlation with the criterion) and also its effect when combined with the other model regressors. To visualize the results from the regression analysis, volumetric and behavioral measures were converted to z-scores. We then fit surfaces of the form z = f(x,y) to the data separately for each behavioral measure, interpolating between the specified data points. These surfaces were subsequently plotted as pseudo-color plots in Matlab 2008a.
The results of the behavioral assessment of olfactory functions are summarized in Table 1. In an additional exploration of possible sex differences, we conducted a MANOVA on the T, D, and I scores, in which the overall effect of sex failed to reach significance, F(3, 88) = 2.47, P = n.s. The same was true for a between-group t-test for the TDI score, t(88) = 1.34, P = n.s.
Gray matter volume was significantly associated with TDI scores in an area of the right orbitofrontal cortex (Fig. 2A), centered on the orbital sulcus, and extending into the intermediate and lateral orbital gyri (cluster size = 463 voxels, peak voxel (x, y, z) [35 38 −3], t = 4.29, PPeak < 0.001, PCluster < 0.02). This association remained significant when the subjects' sex was added to the analysis as a covariate. Statistical tendencies for correlations in the same area were equally observed for the discrimination and threshold scores, but they failed to reach our predefined cluster threshold of P < 0.05 (181 voxels for discrimination, 13 voxels for threshold). Overall, no significant associations with any olfactory subscores were found in our whole-brain analyses.
|TDI score||33.72 (3.50)||25.63||42.13|
|PEA threshold||7.67 (2.08)||3.25||12.13|
|Olfactory discrimination||12.38 (1.74)||5||15.5|
|Olfactory identification||13.67 (1.42)||11||16|
|TDI score||33.72 (3.50)||25.63||42.13|
|PEA threshold||7.67 (2.08)||3.25||12.13|
|Olfactory discrimination||12.38 (1.74)||5||15.5|
|Olfactory identification||13.67 (1.42)||11||16|
Note that the TDI score is a summated score of threshold, discrimination, and identification performance.
Similar whole-brain analyses revealed no association of white matter volume with the overall TDI score. However, we observed a significant effect of olfactory threshold in a cluster located within the right superior longitudinal fasciculus in a region adjacent to the anterior insula, frontal and parietal operculum, as well as the primary motor cortex (cluster size = 1004 voxels, peak voxel = [53 −13 21], t = 4.47, PPeak< 0.001, PCluster< 0.003), see Figure 2B.
Comparisons between the extracted gray matter volume measures from the OFC, piriform cortex, and olfactory bulb ROIs demonstrated significant correlations between the OFC and the piriform cortex, r(90) = 0.28, P < 0.01, and a significant negative correlation between bulb volume and piriform cortex volume, r(90) = −0.27, P < 0.01. There was no significant correlation between olfactory bulb and OFC volume. Despite these correlations, an analysis of multicollinearity showed that the variance inflation factor was low (all <1.2, with scores >5 considered an indicator for the presence of high multicollinearities), thus indicating that these correlations did not artificially inflate the overall variance explained by the model.
The multiple regression analysis of the ROI data with the TDI score revealed a significant model fit, F(3,86) = 10.85, P < 0.001, adjusted R2= 0.25. This effect was mainly mediated by a significant relationship between OFC volume and TDI score, t(86) = 3.74, P < 0.001, as well as a significant relationship between olfactory bulb volume and TDI score, t(86) = 3.59, P< 0.001. For the subscores, a functional division emerged: threshold scores and olfactory discrimination ability were predicted by OFC volume (threshold: t(86) = 2.85, P< 0.005, discrimination: t(86) = 2.83, P< 0.006), while olfactory bulb volume and piriform cortex volume did not explain significant proportions of the variance. On the other hand, for the olfactory identification scores, we found that olfactory bulb volume, t(86) = 6.85, P< 0.001, but not any of the cortical measures, predicted behavioral performance (see Table 2). The structural–behavioral relationships are illustrated by means of pseudo-color plots (Fig. 3).
|Behavioral measure||F||Adj. R2||Relative importance||T||P|
|Mean discrimination score||4.38||0.10||0.006*|
|Odor identification score||16.76||0.35||<0.001*|
|Behavioral measure||F||Adj. R2||Relative importance||T||P|
|Mean discrimination score||4.38||0.10||0.006*|
|Odor identification score||16.76||0.35||<0.001*|
Average gray matter values, scaled by total intracranial volume, were extracted from ROIs in the piriform and orbitofrontal cortex. Olfactory bulb volumes were extracted from T2-weighted images by manual segmentation.
*Variable with a significant contribution to the model. Note that relative importance metrics were normalized to sum up to 1.
Finally, since age as a factor has been associated with a decline in olfactory functions, we assessed whether participants' age significantly influenced any of the above-mentioned structural–behavioral associations. To this end, we performed separate age-corrected models for the general TDI score, as well as for each subcomponent. Age did not explain a significant proportion of the variance [TDI P(age) = 0.47; threshold P(age) = 0.54; discrimination P(age) = 0.10; identification P(age) = 0.19].
The strong connection between right orbitofrontal cortex volume and general olfactory performance observed in this study highlights the important role of this structure in the formation of the olfactory percept. While many reports have previously linked OFC damage with substantial olfactory deficits, we can for the first time demonstrate the reverse: within a group of healthy subjects, larger gray matter volume in the central orbital sulcus contributes significantly to higher acuity of olfactory perception. As previously described for the visual and auditory modalities, these results demonstrate that VBM equally constitutes a useful tool for the identification of structural changes associated with higher perceptual acuity in olfaction. In addition, our ROI analysis implicates that variations in OFC volume contribute unique portions of explained variance to differences in behavioral performance, which are separable from the behavioral variance accounted for by olfactory bulb volume. These differential contributions of our pre-defined ROIs to the investigated behavioral subtasks provide important insight into functional division within the human olfactory network.
The orbitofrontal cortex receives major projections from the piriform cortex (Carmichael and Price 1994; Gottfried and Zald 2005), and constitutes as such arguably the most important olfactory target area within heteromodal association cortex. Our whole-brain analyses localized the primary site of association of gray matter volume with the TDI score in an area centered on the transverse orbital sulcus. This area has in previous neuroimaging studies and our ALE meta-analysis been localized as the site of higher order olfactory processing (Zatorre et al. 1992; Gottfried and Zald 2005). Moreover, the meta-analysis-based ROI analyses demonstrated that gray matter volume differences in this area accounted for the majority of the variance explained by our regression model in the discrimination scores, and a smaller extent of the variance explained by threshold scores.
Various authors have over the past few years explored the contributions of the OFC to the formation of the olfactory percept. While earlier lesion studies suggested an essential role for all aspects of olfactory processing (Eichenbaum et al. 1980; Potter and Butters 1980; Jones-Gotman and Zatorre 1988; Zatorre and Jones-Gotman 1991), more recent studies have pointed out that some “hard-wired” complex olfactory tasks, such as distinctions between the smells of predator and kin, can in fact be accomplished without an intact orbitofrontal cortex (Gottfried 2007; Li et al. 2010).
Considering commonalities between olfactory functional neuroimaging studies that report OFC involvement, it is now thought that its main role is to maintain flexible odor representations for the adjustment of behavior, subserving such diverse tasks as perceptual decision-making and confidence judgment (Kepecs et al. 2008; Bowman et al. 2009), but also perceptual learning, valence judgment, and aversive conditioning (Rolls et al. 2003; Gottfried 2007), as well as the integration of information from other sensory modalities (Gottfried and Dolan 2003; Price 2008). Recently, studies have been able to directly identify flexible odor-evoked spatial response patterns in the OFC as a potential neural mechanism behind the formation of these adaptable odor representations (Gottfried and Zelano 2011; Zelano et al. 2011). Both odor quality discrimination and odor detection tasks depend strongly on the ability to form transient odor representations (Hedner et al. 2010) as they demand ongoing perceptual decision-making, response criteria assessment, and at times integration of multisensory information (Jadauji et al. 2012). The association of these tasks with OFC gray matter volume reported here, thus ties in well with the suggested functions of this brain area, and overlaps with functional activations previously observed during these tasks (Savic et al. 2000).
Our study further demonstrates that odor identification, the subtask which depends on discrete representations of odor quality, is strongly associated with olfactory bulb volume, predicting a whole 35% of the explained variance. This connection is intriguing because it occurs at the first synapse of the olfactory system, where neuronal convergence produces maps that provide distinguishable spatial and temporal patterns for different odorous stimuli (Rubin and Katz 1999; Xu et al. 2003; Murthy 2011). Here, the ability to separate the highly overlapping spatial activation patterns of different odorants in the bulb into discrete representations of quality has been suggested to form the basis of learned odor recognition abilities (Sahay et al. 2011). Rodent studies have indicated that such improvements in pattern separation ability may be achieved through the ongoing formation of local contrast-enhancing inhibitory microcircuits. Diverse odor environments stimulate neurogenesis of these interneurons during early development (Rochefort et al. 2002; Arenkiel et al. 2011) and possibly all through adulthood (for a discussion, see Mandairon et al. 2011). Furthermore, these neuronal changes are related to improved behavioral performance, thus directly implying a relationship between neuronal growth and perceptual acuity.
While this link has not been directly studied in humans, the clinical literature supports an association of odor identification deficits with bulb volume. Relationships have frequently been reported in anosmia (Yousem et al. 1999; Rombaux et al. 2006a, 2006b), as well as various neurological and psychiatric illnesses that are associated with a loss of olfactory function, such as schizophrenia (Nguyen et al. 2011), depression (Negoias et al. 2010), and multiple sclerosis (Goektas et al. 2011). Our data provide compelling initial support for the idea that, akin to other mammals, larger olfactory bulbs in humans may be associated with better pattern separation skills for odors. However, longitudinal intervention studies will be needed to establish whether systematic olfactory discrimination training can modulate bulb volume in humans.
In light of the strong link between bulb size and odor identification, the complete absence of such a relationship with OFC volume initially seems counterintuitive given the cognitive nature of the task of matching the olfactory percept to a previously stored mental representation of a particular odorant. Previous functional studies that have investigated the cortical substrate of odor memory and identification (Kareken et al. 2003), however, provided little evidence of necessary prefrontal involvement. Most commonly, in the absence of recordings from human olfactory bulb, an important role of the amygdala and hippocampal complex as well as language areas has been implicated in the successful performance of these tasks (Jones-Gotman et al. 1997; Herz et al. 2004; Yeshurun et al. 2009). In this respect, the lack of an association of odor identification performance with neural integrity in our orbitofrontal ROI is consistent with previous literature. This may reflect the fact that in comparison to discrimination and threshold tasks, identification primarily consists of memory retrieval of a verbal label and depends less on the maintenance of parallel transient odor representations than other olfactory tasks.
The impact of white matter volume differences for olfactory functioning is still a neglected topic. While not of primary interest for the present study, it should be noted that the correlations between threshold scores and superior longitudinal fasciculus volume surrounding the insula and motor cortex were robust and showed overlap with the findings reported in the only study previously reporting white matter alterations in the context of olfactory functioning (Bitter, Bruderle et al. 2010). While the adjacent regions are part of the higher order olfactory processing network and have been reported in numerous functional neuroimaging studies (Savic et al. 2000; Kareken et al. 2003; Royet et al. 2003), their specific contributions to threshold processing have yet to be studied. Furthermore, while the association of volume and function has often been replicated in cortical gray matter, it is still debated whether white matter volume differences can be explained analogously (see Canu et al. 2010). Linking our findings to measures of white matter integrity as provided by diffusion tensor imaging will be an important next step in elucidating functional significance of white matter morphometry in the relevant areas. The combination of volumetric measures and DTI would not only help to establish the principal direction of potentially affected fiber tracts, but also determine whether increased fractional anisotropy and white matter volume occur in conjunction to improve network connectivity among olfactory association areas.
Several limitations of this study should be acknowledged. First of all, it should be considered that our functionally derived ROI approach might present an insufficient spatial resolution for isolation of smaller functional subregions. As such, the reasons for the absence of structural–functional associations in the piriform cortex ROI, an outcome which was unpredicted given this area's prominent role in the olfactory system (for an overview, see Lundstrom et al. 2011), can at this point only be speculated about. One possibility is that the modularization of the piriform cortex, with a separation into an anterior and a posterior portion, each with individual afferent and efferent projections and likely fulfilling distinct roles in the formation of an olfactory percept (Gottfried 2010), may obscure a potential linear relationship between volume and function. The complex correlation patterns observed between the volumetric measures extracted from our ROIs can be interpreted as support for this argument. While there was no significant relationship between the volumetric measures in the OFC and the olfactory bulb, piriform cortex volume was correlated with both measures, as would be expected on the basis of the extensive convergence of olfactory network connections taking place within this area (Carmichael and Price 1994). However, the observed correlations with the OFC and olfactory bulb took opposite directions, hence suggesting complex associations with higher and lower level areas of olfactory functioning. The causes for these opposing correlation patterns should be further explored by future studies, possibly using a more focused field of view for MRI acquisition, which would permit better spatial resolution. Further possibilities could be that higher olfactory acuity results in changes in piriform cortex functioning that are not reflected in volume increases, or simply, that correlations cannot be observed because our behavioral tasks do not sufficiently tap into piriform cortex's proposed functions (Wilson and Stevenson 2003). Future studies should tailor their behavioral assessments to such associative odor-learning tasks to potentially achieve a closer overlap with piriform cortex function. An additional limitation is that the present data set did not allow for thorough investigation of the extent to which the subject's sex mediates the observed influences of olfactory performance on brain morphometry. Given that the effect sizes for sex differences in olfaction tend to be rather small, larger subgroup sizes would likely be required to sufficiently address this question as part of a VBM analysis. Exploring this issue further would potentially provide valuable insights into the biological bases of sex-related differences in olfactory perception.
Taken together, our results demonstrate that interpersonal variability in olfactory functioning is reflected in morphological differences in both peripheral and cortical anatomical structures. For the first time, the present study can link volumetric measures from different processing nodes of the olfactory system, which until now have only been studied in separation. We show that both olfactory bulb volume and orbitofrontal cortex volume are strong predictors of olfactory perceptual performance, and uniquely contribute to the explained variance in TDI scores. Deconstructing our results by TDI subscores, we identified a functional division as a potential source of this separation. We propose that higher OFC volume benefits subtasks requiring a flexible adaptation of network activity to odors while a better ability to maintain separate representations of odorants of different quality is related to variability in olfactory bulb volume. Piriform cortex integrity and connection strength within the network may constitute additional sources of variance which need to be explored in future research. The overlap with results from functional neuroimaging and the animal literature illustrates that volumetric approaches have the power to narrow the gap between human and translational research, thereby supporting the formation of plausible network models to inform functional neuroimaging studies of olfactory processes.
This material is based on work supported by the U. S. Army Research Office under grant number W911NF-11-1-0087 and the Swedish Research Council (2009-2337), both awarded to J.N.L. J.S. is supported by a postdoctoral fellowship awarded by the German Research Foundation (DFG SE 2147/1-1). J.F. is supported by a postdoctoral fellowship of the Canadian Institutes of Health Research (CIHR).
The authors would like to thank Thilo Kellermann for his valuable methodological advice. Conflict of Interest: None declared.