We included 16 healthy, young participants (12 women, mean age: 24; 20–29 years) in this study. The study was conducted at the University of Dresden Medical School, according to the Declaration of Helsinki and all subjects gave written informed consent prior to the study. It was approved by the local Ethics Committee (EK number 185062009).

Before the fMRI session, we assessed subjects' olfactory abilities and trigeminal chemoreception. Subjects' ability to identify odors was determined by means of the Sniffin' Sticks identification test kit (Kobal et al., 2000). In this test, subjects are presented with 12 pen-like odor dispensing devices. Their task was to choose the right descriptor from a list of four for each odor. We counted the number of correct responses. Further, we assessed subjects' ability to localize odors by means of the odor lateralization test for eucalyptol (Hummel et al., 2003). We stimulated subjects with a device which allows the delivery of predefined volumes of air to both nostrils simultaneously. In this test, subjects were stimulated with odorized air to one nostril and odorless air to the other; their task was to detect the side of odor stimulation. We used neat eucalyptol as the odor stimulus. The task was repeated 40 times with each trial separated by 40 s. We counted the number of correct localizations. We only included participants with a normal ability to identify odors [i.e., who were able to identify more than 10 out of 12 sticks (Hummel et al., 2001)] and the ability to localize odors above chance [more than 25 out of 40 (Frasnelli et al., 2008)].

We used 20% of eucalyptol saturated air (eucalyptus odor) and a mix of 20% of phenyl ethyl alcohol saturated air (rose odor) with 60% carbon dioxide (CO₂) (Boyle et al., 2007) as bimodal odors; both are known to activate the olfactory and the intranasal trigeminal system (Hummel et al., 2003; Boyle et al., 2007). The reasons why we decided to use a mixture of phenyl ethanol and carbon dioxide instead of a monomolecular substance are two-fold. First, phenyl ethanol is a pure odorant and therefore very difficult to be localized by humans (Frasnelli et al., 2009); carbon dioxide is virtually odorless and therefore very difficult to identify; the mixture of both, however, is both localizable and identifyable (Boyle et al., 2007). Second, it is difficult to match a monomolecular substance with regards to olfactory and trigeminal intensity. By using a mixture, we could adjust both trigeminal and olfactory intensity (by changing the concentrations of phenyl ethanol and CO₂ separately) to match eucalyptol's in pilote experiments. Stimuli were therefore isointense on both, the olfactory and the trigeminal dimensions. Subjects were familiarized with both odors and could easily distinguish them.

Odor stimuli were applied by means of a computer-controlled air-dilution olfactometer (OM6b; Burghart, Wedel, Germany). This stimulator allows application of rectangular-shaped chemical stimuli with controlled stimulus onset. Mechanical stimulation is avoided by embedding stimuli into a constant flow of odorless, humidified air of controlled temperature (80% relative humidity, total flow 8 L/min, 36°C) (Kobal, 1981). Thus, throughout the experiment, the subjects received humidified, warm air to their nostril. During stimulation an odor was embedded into this constant airflow. The olfactometer allows for stimulation of each nostril separately. Subjects were instructed to breathe through their mouth to avoid potential sniff-related activity.

We used a block design for stimulation. During the entire fMRI session, subjects focussed on a black cross on a screen. Nine seconds before the “on-period”, the cross switched to one of two questions [task; either “where?” (German: “wo?”) or “what?” (“was?”)]. The order of the questions was pseudo-randomized and counterbalanced. The text stayed on the screen for 5 s after which it switched back to the black cross. Four seconds later, the “on”-period begun during which odor stimuli were delivered five times, each 400 ms long, every 4 s. The chemosensory stimuli were either eucalyptol or the PEA/CO₂ mixture and delivered either to the left or the right nostril (all stimuli pseudorandomized and counterbalanced). After each stimulus, subjects responded to the task by pressing one of two buttons with the index of their right hand. Specifically, during the localization task they had to indicate whether their left or their right nostril was stimulated and during the identification task, they had to indicate whether they received eucalyptol or the PEA/CO₂ mix. The “on”-period was followed by a 30 s “off”-period, during which subjects received odorless air (AIR). During one run, we delivered 10 “on”- and 10 “off”-periods; subjects were tested in two runs. For data analysis we classified volumes during the “on”-periods as “where” or “what” conditions, whereas the “off”-periods were classified as “baseline” condition.

The study was performed using a 1.5 MRI scanner (Sonata; Siemens, Erlangen, Germany). For anatomical overlays, a T1-weighted (turboflash sequence) axial scan with 224 slices, voxel size of 1.6 × 1.1 × 1.5 mm, a repetition time (TR) of 3000 ms, echo time (TE) of 3.93 ms, and 2 averages (2130/3.93/2) was acquired. Functional data acquisition was performed in the axial plane (oriented parallel to the planum sphenoidale to minimize artifacts) using a multislice spin-echo echo-planar imaging sequence. Scan parameters included a 64 × 64 matrix, voxel size of 3 × 3 × 3.75 mm, TR of 3000 ms, and a TE of 35 ms. A total of 207 images were acquired at each of 24 slice locations per run over the course of a total functional acquisition session of approximately 10 min in length. The three imaging conditions consisted of (1) subjects identifying a chemosensory stimulus (“what”), (2) subjects localizing a chemosensory stimulus (“where”), and (3) chemosensory-free low-level baseline.

The functional MRI data was analyzed by means of SPM8 (Wellcome Trust, http://www.fil.ion.ucl.ac.uk/spm/software/spm8/) implemented in Matlab (Mathworks Inc., Natick, MS). Functional data were registered, motion-corrected, and resliced using SPM8 preprocessing procedures. Mean functional images were coregistered to the anatomical T1 volume. We then performed the analysis on spatially normalized stereotactically transformed into ICBM152-space and smoothed images (8 mm full width at half maximum Gaussian kernel).

In order to first verify that our imaging paradigm reliably activated chemosensory processing areas, we initially assessed the main effect of odor stimulation by comparing both odor conditions against no odor condition (where + what vs. no-odor baseline). We observed activations areas commonly associated with chemosensory processing, such as left and right insula, the right OFC, as well as multisensory integration centers such as the right inferior parietal lobule and the left supramarginal gyrus (see Table 1).

To verify task specific brain activations, we compared the two stimulation conditions to each other. When contrasting odor localization against odor identification (where vs. what), we observed activations of a cluster in the left intraparietal sulcus, and one in the right superior frontal sulcus (Table 2).

The opposite contrast (what vs. where) did not reveal any significant activation above threshold criteria.

We used ICA to obtain specific component maps for individual subjects. We then extracted, for each subject, the component with the best fit to the time course of each condition. Information on correlation coefficients for individual components in both tasks is outlined in Table 3.

The resulting statistical maps were submitted to a subsequent second level analysis where the component for odor identification (“what”) revealed two significant clusters within an area of the left central fissure and the left superior frontal gyrus (Table 4).

For odor localization (“where”), we detected two clusters above set criterion, one located in the right hippocampal region and another in left paracentral lobule (Table 5).

In Figure 1 we provide an overview of activations in the parietal cortex obtained in different conditions (Figure 1).

When subjects were localizing the odorous objects, both means of analysing the neuroimaging data identified significant activation of the left posterior parietal lobe, in addition to its general activation independent of the task. These observations fit well with the literature where object localization consistently activates posterior parietal regions. For instance, in analogy to the findings in non-human primates (Mishkin et al., 1983) a visual spatial localization task led to activation in the lateral superior parietal cortex (Haxby et al., 1991). In the auditory system, object localization activated a dorsal stream from the caudal primary auditory cortex to the inferior parietal cortex (somewhat lower than for visual stimuli) to middle and inferior frontal gyri, whereas anterior primary auditory cortex to posterior frontal and orbitofrontal regions formed the ventral stream for object identification (Rauschecker and Tian, 2000; Maeder et al., 2001).

Different subregions of the parietal lobe play particular roles in the dual stream dichotomy. SPL and IPS, whose activation is often associated with activation of the dorsoloateral frontal lobe, are part of the dorsal frontoparietal system for directing spatial attention. IPL on the other hand, is activated, together with more ventral frontal regions when individuals perform non-spatial tasks. Thus, there is a gradient from more spatial tasks in SPL to less spatial tasks in IPL, with the IPL's suggested role to sustain attention over time (Husain and Nachev, 2007). The data obtained within the present study corresponds closely with these earlier reports, especially with regards the parietal lobe: unilateral chemosensory stimuli triggered activation of right IPL independent of the task subjects performed, indicating multisensory integration. Localization of these stimuli, however, led to a significantly stronger activation of the left IPS than what we observed for stimulus identification. Therefore, the cortex in and around the IPS is part of a dorsal stream responsible for object localization in different sensory systems, including the chemical senses. Activation of this particular brain region was observed when subject localized monomodal stimuli, such as visual (Haxby et al., 1991), auditory (Rauschecker and Tian, 2000; Maeder et al., 2001), and somatosensory (Reed et al., 2005) ones as well as multimodal stimuli such as audio-somatosensory (Renier et al., 2009) and olfactory-trigeminal ones (present study). It is commonly activated with mixed olfactory trigeminal stimuli (Boyle et al., 2007; Lombion et al., 2009).

Next to the parietal activations, we also observed activations in the frontal lobe. First, object localization led to a significant activation of the right superior frontal sulcus, as shown by the GLM contrast “where” vs. “what.” Second, the ICA demonstrated odor object identification to be associated with the left superior frontal gyrus. In addition to these hemispheric differences, the latter activation was located more posteriorily than the former. It has been demonstrated that there are distinct working memory systems for spatial and verbal information predominantly located in the (dorsolateral) prefrontal cortex of both hemispheres. The right hemisphere stores and maintains information on spatial features, whereas the left hemispheres does the same for verbal and object identity information (Smith and Jonides, 1997; Belger et al., 1998). This appears to be modality independent, as both visual and tactile working memory evoked similar frontoparietal networks including the posterior parietal cortex and the dorsolateral prefrontal cortex, with a leftwards tendency for object discrimination (Ricciardi et al., 2006). In Figure 2 we highlight how our results fit into the same framework. After activation of chemosensory regions common to both tasks, both tasks activated a parieto-frontal network, from the posterior parietal cortex to prefrontal areas, with activation of a left sided and a right sided frontal area for object identification and localization, respectively (Figure 2).

ICA revealed a puzzling finding in the activation of the right posterior hippocampus when subjects were localizing odors. Hippocampal activation is usually linked to spatial navigation and episodic memory (Igloi et al., 2010). Localizing odorants to the left or right nostril is clearly a spatial task; however, our paradigm did not explicitly involve a mnesic component. We know that hippocampus also stores serial working memory of spatial locations even in the encoding phase (Toepper et al., 2010). Importantly, this is true even in implicit conditions: hippocampal activation can be observed when subjects learn the temporal structure of sequences even without any conscious sequence knowledge (Schendan et al., 2003). In our paradigm, we presented our subjects with a series of spatial locations. Therefore, implicit spatial sequence learning may therefore explain the hippocampal activity we observed.

There has been a prior attempt to investigate the dissociation between object localization and identification in the chemical senses (Porter et al., 2005). Here, subjects smelled four odorants. Similarly to our study, odors were delivered monorhinally, and subjects were asked to either identify or localize the odors. Although the main focus of the study was to investigate nostril specific receptive fields within the piriform cortex, the authors also compared brain activations between both tasks. They did indeed observe dissociations between tasks which differentially activated three specific brain areas: odor identification activated the occipital gyrus and the paracentral lobule to a larger extent than odor localization; odor localization, in turn, activated the superior temporal gyrus more than odor identification. These findings contradict the existing literature where activation of different regions of the occipital cortex has been reported mainly for visual stimuli, during object identification [e.g., the occipitotemporal junction for face recognition (Haxby et al., 1991)] rather than for object localization. On the other hand, temporal areas have been associated with object identification rather than object localization (Mishkin et al., 1983), with the exception of sound localization (Maeder et al., 2001). Further, the existing literature, paired with the present results, suggest that activation of superior parietal areas, such as the paracentral lobule, appears to be more commonly linked to object localization (Haxby et al., 1991; Rauschecker and Tian, 2000; Maeder et al., 2001; Reed et al., 2005; Renier et al., 2009). The exact implications of Porter and colleagues (Porter et al., 2005) findings therefore remain unclear.

An earlier study on chemosensory stimulation with CO₂ has compared regression-based analysis of fMRI based data using the GLM with that of group analysis using the ICA methods. Some activations were only detected by group ICA, but not by GLM; this could be explained by the fact that activity in these regions was shifted temporally and therefore delayed with respect to the expected response. Furthermore, it showed a variation of CO₂-stimulus-evoked responses which was different for the selected ROIs within one subject (Schopf et al., 2011). This finding of differing hemodynamic responses across subjects, brain regions and sessions is a known constraint of regression-based methods such as GLM (Aguirre et al., 1998; Cunnington et al., 2003; Neumann et al., 2003; Handwerker et al., 2004; Menz et al., 2006). Fully exploratory analysis methods such as ICA [introduced by (McKeown et al., 1998)] do not require the specification of a model or a hemodynamic response function. A number of approaches have been developed to extend ICA from the analysis of a single data set to the group level (Calhoun et al., 2009); the most widely adopted method is to concatenate single-subject data in time prior to performing ICA (Calhoun et al., 2001; Beckmann and Smith, 2005) [for a comparison of toolboxes using temporal concatenation ICA see (Schopf et al., 2010)].

A challenge in group ICA is the need to identify and evaluate group components. This can either be done by temporally correlating the model time course with the corresponding time courses of the group components or by template matching, which includes the spatial correlation of a predefined template with the group component maps. For the present data, we used temporal correlation to find spatial activity patterns across subjects. As hypothesized earlier (Schopf et al., 2011) our study showed that group ICA provides supplemental information—in our case regarding parallel pathways processing—in addition to a priori defined model-dependent regression-based analysis.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.