Abstract
Odor perception plays a critical role in early human development, but the underlying neural mechanisms are not fully understood. To investigate these, we presented appetitive and aversive odors to infants of both sexes at 1 month of age while recording functional magnetic resonance imaging (fMRI) and nasal airflow data. Infants slept during odor presentation to allow MRI scanning. We found that odors evoke robust fMRI activity in the bilateral olfactory cortex and thalamus and that fMRI response magnitudes in the olfactory cortex differ across odors. However, in contrast to prior work in adults, we did not find compelling evidence that odor stimuli evoke discriminable fMRI activity patterns in the olfactory cortex or thalamus using two different multivariate pattern analysis techniques. Finally, the average inhale airflow rate was higher for appetitive odors than aversive odors, which tentatively suggests that infants could modulate their respiration to reflect odor valence. Overall, these results show strong neural responses to odors at this early developmental stage and highlight nasal airflow as a behavioral metric for assessing odor preference in infants.
Significance Statement
The sense of smell facilitates several adaptive behaviors in infants (e.g., feeding, soothing), but the brain areas supporting infant olfaction are understudied. Here, we delivered appetitive and aversive odors to sleeping infants during functional magnetic resonance imaging. We show that odors evoke activity in olfactory brain regions and the thalamus already at 1 month of age and activity levels vary across odors in some of these regions (e.g., piriform cortex, amygdala). However, we did not find strong evidence for pattern-based odor information in the same brain areas. Finally, preliminary nasal airflow findings suggest that infants inhale more vigorously in response to appetitive compared with aversive odors. Taken together, these findings advance our understanding of the neural mechanisms of infant olfaction.
Introduction
Although human infants are highly dependent on their caregivers to fulfill their basic needs, they are surprisingly savvy when it comes to their sensory capacities. This is particularly true for the chemical senses, which begin functioning in the womb. Anatomical and neurochemical evidence suggest that fetal olfaction is viable at around Gestational Week 29 (Schaal et al., 2004) and exposure to odorous chemicals prenatally though amniotic fluid facilitates the transition to the postnatal environment (Schaal et al., 2020). Prenatal odor exposure can also shape olfactory and flavor preferences in later development (Schaal et al., 2000; Mennella et al., 2001; Hepper et al., 2013; Wagner et al., 2019). For instance, exposure to anise through maternal diet in the final prenatal weeks has been shown to increase preference for anise odor soon after birth (Schaal et al., 2000).
The ability of newborns to recognize certain odors in the first few days of life is highly adaptive, since olfactory cues guide critical approach and avoidance behaviors [e.g., food intake (Forde and de Graaf, 2022), threat memory (Li and Wilson, 2024), kin recognition (Lundström and Olsson, 2010)]. For example, newborns exhibit a preference for their mother's scent and will orient toward objects that carry it (Varendi and Porter, 2001; Marin et al., 2015). Odors can also serve as powerful socioemotional cues for infants. For instance, the scent of breastmilk can facilitate soothing following painful stimuli (Nishitani et al., 2009; Tasci and Kuzlu Ayyildiz, 2020). Notably, olfactory dysfunction has been associated with neurodevelopmental disorders, such as autism spectrum disorder (Crow et al., 2020).
Despite the clear ecological relevance of olfaction in early human development, the neural mechanisms of odor perception are not well established in infants. In rodents, the olfactory system undergoes profound changes in the first few weeks of life (Maier and Zhang, 2023), both at the level of the olfactory bulb and its main downstream target, the piriform cortex. For instance, inhibitory synaptic circuitry is still developing in the piriform cortex during the first few postnatal weeks (Pardo et al., 2018), and high-frequency gamma oscillations begin to ramp up in the same timeframe (Zhang et al., 2021). In terms of human infant research, a few olfactory functional magnetic resonance imaging (fMRI) studies have been conducted (Arichi et al., 2013; Adam-Darque et al., 2018, 2020). These studies provide preliminary evidence that odors evoke neural activity in the piriform cortex and other olfactory brain areas in infants, similar to what has been demonstrated in adults (Torske et al., 2022).
The relative lack of research in this area is due, at least in part, to the challenges of olfactory neuroimaging (Olofsson and Freiherr, 2019) and neuroimaging in infants more broadly (Dubois et al., 2021). Critically, previous work did not measure nasal airflow, which is integral for determining the precise delivery of odors, confirming nasal inhalation during odor presentation, and controlling for respiratory confounds in fMRI time-series data. Nasal airflow may also provide an expedient tool to assess whether infants show behavioral responses to odors, as has been done for odor valence in children (Rozenkrantz et al., 2015) and adults (Bensafi et al., 2003; Arzi et al., 2012). Moreover, it is unclear whether infant olfactory cortices encode specific odor information, as they do in adults (Li et al., 2008; Howard et al., 2009; Sagar et al., 2023).
To address these open questions, we simultaneously recorded fMRI activity and nasal airflow from young infants in response to four unique odors (two appetitive, two aversive) while they slept. We find that odors elicit strong fMRI responses in the olfactory cortex and thalamus. Moreover, while fMRI responses in the olfactory cortex differed significantly across the four odors in terms of magnitude, multivoxel pattern analysis (MVPA) did not reveal robust evidence for discriminable patterns of fMRI activity in olfactory brain areas. Finally, we found preliminary evidence that inhalation reflects odor valence, with greater nasal inflow for appetitive compared with aversive odors.
Materials and Methods
Study overview
Twelve full-term infants underwent olfactory fMRI scanning at ∼1 month of age. Infants were fed and swaddled prior to fMRI scanning to promote drowsiness (Copeland et al., 2021). Once asleep, infants were fitted with custom olfactometry equipment, including a small nasal mask to deliver odors and record nasal airflow, and then positioned in the MRI scanner (Fig. 1a,b). During fMRI scanning, a constant stream of clean air was directed to the sleeping infants via the nasal mask. Odorized air (or clean air for control trials) was inserted into the airstream with precise timing for 5 s blocks, with 5–10 s between blocks (Fig. 1c). Odor blocks contained one of four olfactory stimuli: two appetitive (isoamyl acetate or banana-like; ethyl hexanoate or strawberry-like) and two aversive (isovaleric acid or sweat-like; cyclopentanethiol or gasoline-like; Fig. 1d). Importantly, because clean air was constantly delivered during the intervals between odor blocks, odor delivery did not induce changes in overall airflow or somatosensory stimulation.
Infant olfactory neuroimaging approach. Sleeping infants were fitted with a nasal mask (a) and then secured in the MRI scanner (b). c, During fMRI scanning, odor trials (red and blue clouds) and clean air trials (gray cloud) were presented for 5 s blocks, with 5–10 s between blocks (clean air was also delivered between blocks). Nasal airflow was recorded in parallel. d, Olfactory stimuli included two appetitive odors (blue clouds) and two aversive odors (red clouds).
Participants
Thirteen infants (4 males, 9 females), all vaginally delivered at full term (mean gestational age, 39.16 weeks ± 0.28 standard error of the mean (SEM); range, 37.20–41.10 weeks), were recruited from the nursery and postpartum ward of Northwestern Medicine Prentice Women's Hospital. Parents provided written informed consent for their infant's participation in the study. To be considered eligible, infants had to be clinically well, not be receiving antibiotics, have normal nasal/facial anatomy, have English-speaking parents, and not have MRI contraindications (e.g., implanted medical devices, screws or metal fragments, etc.). All infants underwent the olfactory fMRI protocol at ∼1 month of age (mean age, 38.23 d ± 1.78 SEM; range, 29–48 d). One infant did not receive any odors because they did not stay asleep during fMRI scanning, leaving data from 12 infants for analysis (mean age, 37.58 d ± 1.80 SEM; range, 29–48 d). Microbiome samples (nasal, oral, and fecal) and neurobehavioral assessments were collected from the infants at three time points (microbiome, 1–2 d; ∼1 month, ∼5 months; neurobehavioral, ∼1 month, ∼5 months, ∼2 years). Microbiome and neurobehavioral components of the study are beyond the scope of this manuscript, with analyses in progress. The study was approved by the Institutional Review Board of Northwestern University (STU00212230) and Ann & Robert H. Lurie Children's Hospital of Chicago (IRB 2020-3521).
Olfactory stimuli and odor delivery
Olfactory stimuli included two appetitive odors (isoamyl acetate and ethyl hexanoate) and two aversive odors (isovaleric acid and cyclopentanethiol). Odors were diluted in mineral oil, and concentrations were adjusted to achieve approximately equal intensities based on the perception of the study team. Odor/mineral oil ratios were as follows: 4.5/1,000 (isoamyl acetate), 1/1,000 (ethyl hexanoate), 1/100 (isovaleric acid), and 1/100,000 (cyclopentanethiol). Odors were directed through a computer-controlled olfactometer at a stable flow rate of 3.6 L/min and then routed to the infant participants via a small nasal mask (Fisher & Paykel Healthcare). The mask was connected to a ventilation device (FlexiTrunk interface, Fisher & Paykel Healthcare), which was then connected to a spirometer via a custom 3D-printed adapter. This allowed us to deliver odors while sampling nasal airflow continuously using PowerLab acquisition hardware and software (ADInstruments).
MRI scan and odor delivery protocol
After infant participants and their parent (or parents) arrived at the MRI facility, the parent (or parents) fed and swaddled the infant to facilitate sleep. Once asleep or sufficiently drowsy, infants were fitted with ear plugs, some of the odor delivery equipment (i.e., ventilation device), and a pulse oximeter attached to the infant's foot. The infant was secured in a swaddle device to limit movement (Pearl Technology BabyFix Coccoon) and then soothed as needed until soundly asleep. The sleeping infant was then placed on the MRI scanning table and fitted with the remaining odor delivery equipment (i.e., nasal mask delivering clean air), additional padding around the head to further restrict movement, and the MRI head coil. Finally, the infant was positioned in the bore of the MRI scanner. An MRI technologist stayed in the MRI scanner room for the duration of the scan to monitor the infant and sometimes entered the bore of the MRI scanner to soothe the infant as needed and to signal to the rest of the data collection team when to start the scan and initiate the odor delivery protocol (i.e., upon observing that the infant was sleeping soundly). For instance, if the infant was disturbed by the start of the fMRI scan, the MRI technologist would indicate to the rest of the data collection team that they should wait to initiate odor delivery until the infant fell back asleep. Thus, the study team relied on behavioral observations from the MRI technologist to determine whether the infant was sleeping and did not collect additional measures (i.e., electroencephalography, electromyography, and eye tracking) to confirm sleep or differentiate stages of sleep (i.e., rapid eye movement vs nonrapid eye movement). This approach was sufficient to achieve our study goal, which was to present odors to infants in the MRI scanner while limiting distress and motion confounds.
During the odor delivery protocol, odors were presented for 5 s blocks, with an interstimulus interval of 5–10 s between blocks. The 5 s stimulus duration was chosen to comfortably capture at least one respiratory cycle per trial, assuming a median respiratory rate at 1 month of age of ∼43 breaths per minute (Fleming et al., 2011). Odors were presented in a pseudorandom order, such that the different odor types were evenly distributed over time, interstimulus intervals were evenly distributed across the different odor types, and a given odor never appeared more than twice in a row. Odor trials were interspersed with clean air trials as a control, and there were two clean air trials for every four odor trials. Odor and clean air stimuli were organized in larger sequences of 36 trials [24 odor trials (6 trials per odor), 12 clean air trials], with 30 s between sequences. These longer pauses, as well as natural pauses due to infant arousal, allowed us to divide the odor delivery period into separate sessions for analysis purposes. This was important since we opted for continuous MRI scanning as much as possible to avoid sudden shifts in MRI scanner noise that could disturb infant sleep.
The total number of trials was contingent on the duration of infant sleep in the MRI scanner, which varied considerably across infants (mean number of trials, 105.25 ± 8.14 SEM; range, 68–166 trials). The odor delivery protocol was paused upon signs of infant arousal (e.g., significant movement, MRI technologist signaling arousal), and the odor delivery protocol and MRI scan were terminated if these signs persisted. When feasible, the data collection team made an effort to soothe the infant until they fell back asleep and then reinitiated the MRI scan and odor delivery protocol if successful. This allowed us to maximize the number of trials presented while navigating practical constraints.
fMRI acquisition
MRI data were acquired at the Northwestern University Center for Translational Imaging with a 3 T MRI scanner (Siemens PRISMA) and 64-channel head coil. Following a localizer scan, functional echoplanar images (EPIs) were acquired at a tilted acquisition angle to mitigate signal dropout in olfactory brain areas using the following parameters: TR, 1,600 ms; TE, 37 ms; flip angle, 52°; slice thickness, 2 mm; number of slices, 40; interleaved slice acquisition, field of view, 104 × 104 × 40 mm; and multiband factor, 4. Functional EPIs covered the majority of the infant brain, except the most dorsal portion of the frontal and parietal cortices in some cases.
fMRI preprocessing
We used SPM 12 software (https://www.fil.ion.ucl.ac.uk/spm/) to preprocess and analyze the MRI data. As a first step, the data were parsed into separate sessions, where each session consisted of one set of consecutive trials with no more than 25 s between trials. We retained 4 EPI volumes prior to the first trial in each session and 10 EPI volumes after the last trial of each session and discarded sessions that contained fewer than 12 trials. Next, we generated a mean EPI for each participant and retained the six standard motion parameters relative to the mean. We used these motion parameters to identify sessions where the end of the session was marked by excessive motion (this often coincided with pausing and/or terminating the odor delivery protocol due to signs of infant arousal) and then truncated these sessions to reduce motion artifacts. Thus, the refined dataset consisted of multiple sessions per infant (mean number of sessions, 2.92 ± 0.29 SEM; range, 2–5 sessions), where each session contained consecutive odor trials during which the infant remained relatively still (mean retained trials, 85.25 ± 6.80 SEM; range, 50–122 trials). Finally, retained EPI volumes were realigned to the mean EPI volume, spatially normalized based on an infant structural MRI template (Shi et al., 2011), and then smoothed with a 4 mm3 Gaussian kernel. The origin of the infant structural MRI template was shifted to facilitate successful normalization.
fMRI univariate analysis
To identify brain areas that were activated by odors, we constructed a general linear model (GLM) for each infant. fMRI data were concatenated across sessions, and odor and clean air trials were modeled in two separate columns as 5 s blocks. The models also incorporated concatenated nuisance regressors representing the standard motion parameters from realignment, nasal airflow, nasal airflow volume, additional regressors to account for within-scan head motion (as in previous studies; Shanahan et al., 2021), and MRI session. Estimated models were subjected to one-tailed t tests at the group level (odor trials > clean air trials), and results were whole-brain FWE–corrected at the cluster level to detect brain areas with significant odor-evoked activity (pFWE < 0.05; cluster-defining threshold: puncorrected < 0.001). The bilateral olfactory cortex and bilateral thalamus regions of interest (ROIs) were functionally defined based on the significant clusters.
We constructed a second set of GLMs in a similar manner, where the only difference was that odor trials were modeled in four separate columns according to odor type (i.e., banana, strawberry, etc.). These models allowed us to estimate the magnitude of odor-specific fMRI responses within functionally defined olfactory cortex and thalamus ROIs. Parameter estimates were compared across odors and ROIs using one-way and two-way ANOVAs with repeated measures (Greenhouse–Geisser corrected when appropriate). We also used these models to compute average fMRI responses to appetitive and aversive odors in the olfactory cortex and thalamus and compared the responses using two-tailed t tests.
fMRI MVPA
Finally, we generated additional sets of GLMs for pattern analysis. In the first set, odor and clean air trials were modeled in 10 separate columns, 2 columns per trial type (i.e., odd banana trials, even banana trials, odd strawberry trials, even strawberry trials, etc.). Other aspects of the models were identical to those previously described. Next, we converted parameter estimate maps for the conditions of interest to t-maps and extracted multivoxel patterns of fMRI activity associated with each of the eight odor conditions from functionally defined olfactory cortex and thalamus ROIs (considered separately). Finally, we assessed correlations between even and odd patterns for the 16 possible combinations of odor pairs and compared the mean correlation for patterns associated with the same odors (e.g., even and odd banana patterns) to the mean correlation associated with different odors (e.g., even banana patterns and odd strawberry, sweat, and gasoline patterns) at the group level using a one-tailed t test. In a related analysis, we compared the mean correlation patterns associated with odors of the same valence to the mean correlation patterns associated with odors of different valence at the group level, also using one-tailed t tests.
For support vector machine (SVM) pattern analysis, we used the single-trial method (Mumford et al., 2012) to generate trial-wise fMRI activity patterns. Specifically, we generated a collection of GLMs for each participant, where each GLM represented a single odor trial. In these GLMs, the odor trial of interest was modeled in its own column (e.g., first banana trial from the first session), and the remaining trials from the same session were modeled together in a second column (e.g., all other odor trials from the first session). Other aspects of the models were identical to those previously described, except that sessions were modeled separately to maintain independence between data used to train and test the SVM classifier. Next, parameter estimate maps representing single trials were converted to t-maps. To achieve pattern classification, we trained and tested a four-way SVM classifier (LIBSVM; Chang and Lin, 2011) on the single-trial patterns using a leave-one-session-out cross–validation approach. In other words, the classifier was trained to discriminate patterns associated with the four different odors based on trials from all of the sessions except one and then tested on data from the left-out session. Decoding accuracy was construed as the mean rate of successful classification across iterations, and we used a one-tailed t test to compare actual decoding accuracy to the 25% success rate expected by chance for four-way classification. In cases where there was an extra trial for a subset of odors within a given session (e.g., when the odor delivery period was terminated early due to arousal), the extra trial patterns were discarded to ensure an equal number of trials across the four odor conditions for unbiased pattern classification. In a related analysis, we trained a two-way SVM classifier to discriminate appetitive and aversive odors and compared the resulting decoding accuracy to the 50% success rate expected by chance for two-way classification using a one-tailed t test.
fMRI anatomical ROI analysis
We then repeated the univariate and pattern-based MRI analyses in a set of five anatomically defined olfactory ROIs: piriform cortex, olfactory tubercle, anterior olfactory nucleus, amygdala, and entorhinal cortex. Anatomical ROIs were constructed by obtaining adult olfactory ROIs from previous work (Echevarria-Cooper et al., 2022) and then spatially warping these from Montreal Neurological Institute space to the infant structural template space. The resulting images were resliced, thresholded at 30%, and combined across left and right hemispheres.
Nasal airflow analysis
Nasal airflow data were sampled at 1,000 Hz and postprocessed prior to analysis. Specifically, the nasal airflow trace was parsed into separate sessions in the same manner as the MRI data, smoothed using a 125-point moving average, and then normalized by applying a z-score transformation. Next, we identified the start and end point of each respiratory cycle, and discarded short respiratory cycles. To accomplish this, we used all of the respiratory cycles identified from all of the participants to calculate the mean cycle duration (1.34 s) and cycle duration standard deviation (0.49 s). We then discarded cycles lasting <350 ms, which corresponded to approximately two standard deviations below the mean cycle duration. For each remaining respiratory cycle, we calculated the maximum inhale airflow rate, as well as the inhale volume. We then discarded cycles where both of these parameters exceeded two standard deviations from the mean on an individual participant basis. Our goal was to remove outlier respiratory cycles from the nasal airflow data more broadly, before considering cycles affiliated with odor and clean air trials. Finally, we extracted the nasal airflow parameters associated with the first inhale following each odor (or clean air) presentation, including the mean airflow rate, maximum airflow rate, inhale volume, and inhale duration. We used two-tailed t tests to compare mean parameters across appetitive and aversive odors for each participant at the group level. In a follow-up analysis, we considered all of the inhales that occurred during the 5 s odor and clean air trials rather than limiting the analysis to the first inhale.
Results
Odors evoke robust fMRI activity in infant olfactory brain areas
To probe neural correlates of infant olfaction, we compared fMRI activity between trials in which odors and clean air were delivered (odor trials > clean air trials). Importantly, we leveraged nasal airflow data to confirm nasal inhales during odor delivery and account for respiration in our statistical models (see Materials and Methods). Compared with clean air trials, odor trials evoked a strong and localized fMRI response in a bilateral olfactory cortical region overlapping with the piriform cortex and amygdala (left; t(11) = 6.51; pFWE cluster-corrected = 2.78 × 10−4; right; t(11) = 6.03; pFWE cluster-corrected = 0.002; Fig. 2a) and in the bilateral thalamus (t(11) = 6.15; pFWE cluster-corrected = 1.07 × 10−6; Fig. 2b). Thalamic activity was primarily evident in the mediodorsal nucleus, which receives input from olfactory structures and has been implicated in olfactory function (Tham et al., 2009). This finding extends previous work (Arichi et al., 2013; Adam-Darque et al., 2018, 2020) and demonstrates that olfactory brain areas respond to odor stimuli already in young infants.
Odors evoke robust fMRI activity in infant olfactory brain areas. Significant odor-evoked activity in (a) the bilateral olfactory cortex (left; t(11) = 6.51; pFWE cluster-corrected = 2.78 × 10−4; right; t(11) = 6.03; pFWE cluster-corrected = 0.002; circled) and (b) bilateral thalamus (t(11) = 6.15; pFWE cluster-corrected = 1.07 × 10−6; circled). fMRI response magnitudes differed significantly across odors in (c) the olfactory cortex (F(3,44) = 5.02; p = 0.004), but not in (d) the thalamus (F(3,44) = 0.87; p = 0.47). e, Anatomically defined olfactory ROIs: piriform cortex (red), olfactory tubercle (cyan), anterior olfactory nucleus (magenta), amygdala (blue), and entorhinal cortex (green). f, Significant odor-evoked activity in the anatomically defined piriform cortex (t(11) = 3.41; p = 0.006; two-tailed), olfactory tubercle (t(11) = 3.48; p = 0.005; two-tailed), amygdala (t(11) = 4.93; p = 4.52 × 10−4; two-tailed), and entorhinal cortex (t(11) = 2.34; p = 0.04; two-tailed), but not the anterior olfactory nucleus (t(11) = 0.03; p = 0.98; two-tailed). Statistical maps in a and b are thresholded at puncorrected < 0.001 and overlaid on an infant structural MRI template. The thalamus sections displayed coincide with peak fMRI activation, while piriform cortex sections were selected to show activity clusters from both left and right hemispheres. Gray lines and points in c, d, and f represent individual participant data and error bars depict SEM. ROIs in e are overlaid on an infant structural MRI template.
Next, we tested whether fMRI responses in these brain areas differed across the four odors. To do this, we compared odor-specific responses extracted from the significant clusters in the olfactory cortex and thalamus (averaged across voxels in each ROI). Intriguingly, distinct odors triggered significantly different response magnitudes in the olfactory cortex (F(3,44) = 5.02; p = 0.004; Fig. 2c), but not in the thalamus (F(3,44) = 0.87; p = 0.47; Fig. 2d). Moreover, there was a significant interaction, such that the difference in response magnitudes across the four odors was larger in the olfactory cortex when compared with the thalamus (F(2.64,29.03) = 3.06; p = 0.05). This demonstrates that average activity in the infant olfactory cortex may differentiate distinct odors to a greater extent than downstream brain areas.
Finally, we repeated these analyses in five anatomically defined ROIs: the piriform cortex, olfactory tubercle, anterior olfactory nucleus, amygdala, and entorhinal cortex (Fig. 2e). We first compared the fMRI activity (averaged across voxels in each ROI) between odor trials and clean air trials (Fig. 2f). The comparison was significant in the piriform cortex (t(11) = 3.41; p = 0.006; two-tailed), olfactory tubercle (t(11) = 3.48; p = 0.005; two-tailed), amygdala (t(11) = 4.93; p = 4.52 × 10−4; two-tailed), and entorhinal cortex (t(11) = 2.34; p = 0.04; two-tailed), but not in the anterior olfactory nucleus (t(11) = 0.03; p = 0.98; two-tailed). We went on to test whether fMRI responses differed across the four odors. This was the case in the piriform cortex (F(3,44) = 4.24; p = 0.01) and amygdala (F(3,44) = 6.34; p = 0.001), but not in the olfactory tubercle (F(3,44) = 1.77; p = 0.17), anterior olfactory nucleus (F(3,44) = 1.54; p = 0.22), or entorhinal cortex (F(3,44) = 2.01; p = 0.13). These results reinforce those derived from the functionally defined ROIs and implicate additional brain areas (i.e., olfactory tubercle, entorhinal cortex) in infant olfactory processing.
No evidence for discriminable patterns of odor-evoked fMRI activity in infant olfactory brain areas
After establishing that odors elicit strong fMRI activity in olfactory brain areas, we sought to investigate whether distributed patterns of neural activity in these regions encode odor-specific information, as this is a hallmark of odor coding in adults (Li et al., 2008; Howard et al., 2009; Sagar et al., 2023). To this end, we used two MVPA approaches. First, we extracted average odor-specific fMRI activity patterns from functionally defined olfactory cortex and thalamus ROIs for even and odd odor trials separately (Fig. 3a). Then we computed correlations between the even and odd patterns across all odor pairs and compared pattern correlations between the same odors (e.g., correlation between even and odd banana patterns) with pattern correlations between different odors (e.g., correlations between even banana patterns and odd strawberry, sweat, and gasoline patterns). These correlations did not differ in either region (olfactory cortex; t(11) = 0.53; p = 0.30; one-tailed; thalamus; t(11) = −0.66; p = 0.74; one-tailed; Fig. 3b,c).
No evidence for discriminable patterns of odor-evoked fMRI activity in infant olfactory brain areas. Two different MVPA approaches were implemented. For the pattern correlation approach (a), fMRI activity patterns were extracted from even and odd trials and then correlated across same and different odors. b, c, Average activity patterns from even and odd trials were not significantly more correlated when comparing the same odors versus different odors in functionally defined olfactory cortex or thalamus ROIs (olfactory cortex; t(11) = 0.53; p = 0.30; one-tailed; thalamus; t(11) = −0.66; p = 0.74; one-tailed). For the decoding approach (d), patterns were extracted from single trials, and a four-way SVM classifier was trained and tested on the patterns. e, Odor identity could not be decoded significantly above the chance level (25%) from trial-specific activity patterns in functionally defined olfactory cortex or thalamus ROIs (olfactory cortex; decoding accuracy = 24.9%; t(11) = −0.08; p = 0.53; one-tailed; thalamus; decoding accuracy = 25.0%; t(11) = 0.01; p = 0.50; one-tailed). Gray lines and points in c and e represent individual participant data and error bars depict SEM.
In a second analysis, we used SVM classification to discriminate between fMRI activity patterns evoked by the four different odors (Fig. 3d). To accomplish this, we extracted trial-specific fMRI activity patterns from the same functionally defined ROIs and trained and tested the SVM on these patterns in a leave-one-session-out fashion. As in the pattern correlation analysis, the SVM could not classify odor identity significantly above chance [olfactory cortex; decoding accuracy = 24.9% (chance level, 25%); t(11) = −0.08; p = 0.53; one-tailed; thalamus; decoding accuracy = 25.0%; t(11) = 0.01; p = 0.50; one-tailed; Fig. 3e].
Lastly, we repeated the pattern correlation and decoding analyses in the anatomically defined ROIs described previously. Results (summarized in Table 1) were not significant (all p > 0.05), except for the decoding analysis in anterior olfactory nucleus, which did not correct for multiple comparisons.
Multivariate pattern analysis results for anatomically defined ROIs
Thus, we did not find compelling evidence of discriminable patterns across the four odors. Taken together, these results could mean that odors do not evoke discriminable patterns of fMRI activity in olfactory brain areas at this early developmental stage. However, alternative explanations related to both methodological constraints and the sleep state must be considered (see Discussion).
Infant nasal airflow differs between appetitive and aversive odors
To explore infant respiration during odor delivery, we extracted the nasal airflow trace for the first respiratory cycle after each odor onset and then compared airflow parameters across appetitive and aversive odors (Fig. 4a). We considered the mean inhale airflow rate, maximum inhale airflow rate, inhale volume, and inhale duration, in keeping with a prior study in children (Rozenkrantz et al., 2015). We observed significant differences in the mean airflow rate, such that infants inhaled more vigorously during the presentation of appetitive compared with aversive odors (t(11) = 2.38; p = 0.04; two-tailed; Fig. 4b), and there was a similar trend when considering maximum airflow rate (t(11) = 2.06; p = 0.06; two-tailed; Fig. 4c). We did not observe significant differences for the other two parameters (inhale volume; t(11) = 1.61; p = 0.14; two-tailed; inhale duration; t(11) = 0.05; p = 0.96; two-tailed) or when pooling parameters across all respiratory cycles in each 5 s block (mean inhale airflow rate; t(11) = 1.41; p = 0.18; two-tailed; maximum inhale airflow rate; t(11) = 1.45; p = 0.18; two-tailed; inhale volume; t(11) = 1.32; p = 0.21; two-tailed; inhale duration; t(11) = 0.21; p = 0.84; two-tailed). Although this result should be treated with caution given the modest p value and limited sample size, it suggests that infants may adjust their respiratory behavior in a valence-dependent manner, even while asleep.
Infant nasal airflow differs between appetitive and aversive odors. a, Average nasal airflow trace for the first respiratory cycle following the onset (0 ms) of appetitive odors (blue), aversive odors (red), and clean air (gray). b, The mean inhale airflow rate was significantly higher for appetitive odors than aversive odors (t(11) = 2.38; p = 0.04; two-tailed). c, There was a similar trend for the maximum inhale airflow rate (t(11) = 2.06; p = 0.06; two-tailed). Gray lines in b and c represent individual participant data and error bars depict SEM.
Interestingly, and in contrast with respiratory behavior, odor-evoked fMRI activity did not significantly differ between appetitive and aversive odors in the functionally defined olfactory cortex or thalamus ROIs (olfactory cortex; t(11) = 0.37; p = 0.72; two-tailed; thalamus; t(11) = 0.80; p = 0.44; two-tailed). This suggests that while the olfactory cortex encodes some odor-specific information, this information does not seem to be valence-based. Along similar lines, fMRI activity evoked by appetitive and aversive odors could not be significantly discriminated in either ROI using the pattern correlation MVPA approach (olfactory cortex; t(11) = −0.28; p = 0.61; one-tailed; thalamus; t(11) = 1.28; p = 0.11; one-tailed) or the decoding MVPA approach (olfactory cortex; decoding accuracy = 52.3% (chance level = 50%); t(11) = 0.79; p = 0.22; one-tailed; thalamus; decoding accuracy = 52.4%; t(11) = 0.86; p = 0.20; one-tailed), and we did not find evidence of pattern discrimination when considering anatomically defined ROIs either (Table 1).
Discussion
These results constitute strong evidence for robust fMRI activity localized to infant olfactory brain areas and the thalamus in response to a diverse set of four odor stimuli. This extends prior work that suggests consistency across brain areas supporting olfaction in infants and adults (Arichi et al., 2013; Adam-Darque et al., 2018, 2020). However, earlier studies were largely preliminary. One of them was limited to seven infants and a single odor (Arichi et al., 2013), while another focused on characterizing fMRI responses to individual odors, and aggregate responses were less robust (Adam-Darque et al., 2018). A third study focused on the relationship between odor-evoked fMRI responses and infant colic, without establishing a main effect of odors on brain activity (Adam-Darque et al., 2020). Moreover, none of these earlier studies used anatomically defined ROIs to examine fMRI activity in the broader olfactory network. Thus, our findings add much-needed credence to the basic premise that odors recruit olfactory brain areas in infants and provide a more complete picture of where these effects manifest.
Critically, we also demonstrate that different odors trigger unique levels of fMRI activity in the infant olfactory cortex. This demonstrates that olfactory cortical activity tracks differences between odors in infants rather than simply indicating whether or not an odor is present. Earlier fMRI work suggests that the adult olfactory cortex responds to various dimensions of odor perception, such as valence and intensity (Bensafi, 2012). Although fMRI responses in our study did not track the appetitive or aversive nature of the odors presented, future work could utilize larger odor sets to explore finer-grained aspects of odor encoding in the infant olfactory cortex. Interestingly, thalamic fMRI activity did not differ significantly across odors, suggesting that odor-specific responses are restricted to earlier stages of sensory processing.
Additionally, we used two different MVPA techniques to analyze fMRI response patterns in infant olfactory brain areas. This is an important advance, since earlier infant olfactory studies are limited to univariate fMRI analyses. Intriguingly, both pattern correlation and decoding analyses yielded null results. Namely, we did not find compelling evidence that patterns of odor-evoked fMRI activity can be significantly classified based on odor identity or valence. This is in stark contrast to adult olfactory work, where activity patterns have been repeatedly shown to carry olfactory information, such as odor category (e.g., citrusy, woody, or minty; Howard et al., 2009; Bao et al., 2016), affiliation with foods (Bhutani et al., 2019; Shanahan et al., 2021), and fine-grained odor percepts (Sagar et al., 2023). While these null findings could suggest that odor information is not represented by distributed patterns of neural activity until later in development, it is also possible that odor-based pattern differences did not emerge due to the technical limitations of infant MRI or because infants were sleeping during odor presentation (see final paragraph of Discussion). On the technical side, the resolution of MRI images is relatively lower for infants than adults because of the smaller size of the infant brain, and the number of stimulus presentations is constrained by the depth and duration of the infant sleep period. These issues could be mitigated in the future by implementing higher-resolution neuroimaging and by finding new ways to facilitate infant comfort and sleep in the MRI scanner.
Finally, we went beyond prior work by collecting nasal airflow data during neuroimaging. This metric not only confirms inhalation of olfactory stimuli and corrects for potential respiration-related signals in fMRI models but also allowed us to investigate an important behavioral read-out of infant olfaction. In doing so, we found that nasal airflow seems to be modulated by odor valence, such that infants inhaled more vigorously when appetitive odors were present in the airstream as compared with aversive odors. This finding should be taken as preliminary due to the modest p value and small sample size, and further work in this area is needed to arrive at a more definitive conclusion. Still, this valence-based trend is similar to what has been observed in adults during both wakefulness (Bensafi et al., 2003; Prescott et al., 2010; Ferdenzi et al., 2015) and sleep (Arzi et al., 2012) and could serve to limit intake of potentially noxious odors and inform approach and avoidance behaviors. Moreover, given that prior olfactory work in infants often relies on more subjective interpretations of various behaviors (e.g., physical orientation, mouthing, facial expressions; Schaal et al., 2000; Varendi and Porter, 2001; Delaunay-El Allam et al., 2006; Marin et al., 2015; Wagner et al., 2019), our approach also represents an important methodological advance. Future research could harness nasal airflow to learn more about odor perception during very early development, when collecting verbal reports is not yet viable.
It is worth emphasizing that fMRI and nasal airflow measures were collected predominantly during the sleep state. This was necessary to enable MRI scanning of infants but also makes it difficult to compare infant and adult findings directly. This is because olfactory fMRI data are almost always collected in the wake state in adult studies, with the exception of a handful of memory reactivation studies, where odors delivered during sleep were associated with specific memory content and activated limbic brain areas (Rasch et al., 2007; Shanahan et al., 2018). Thus, differences between our findings and those from adult studies (e.g., no evidence for fMRI pattern differences across odors) could be attributable to the sleep state rather than developmental stage. Perhaps future advances enabling infant fMRI data collection in the wake state will allow for more direct comparisons. Until then, that odors can activate olfactory brain areas and alter inhalation in infants even during sleep further demonstrates the importance of olfaction in early human development.
Footnotes
We thank Todd Parrish and Yufen Jennie Chen for their assistance with the MRI protocol, as well as Sumehda Attanti and Rachael Young for their assistance with data collection. We also acknowledge Sebastian Otero for his assistance with participant recruitment and logistics and Sheila Krogh-Jespersen for her input on infant sleep MRI. This research was funded by grants from the National Heart, Lung, and Blood Institute (T32HL007909 to L.K.S.) and the National Institute on Deafness and other Communication Disorders (R01DC015426 to T.K.), as well as the Rhodes College Faculty Development Endowment Grant (to L.K.S.), the Visionary Grant from Stanley Manne Children’s Research Institute and Ann & Robert H. Lurie Children’s Hospital of Chicago (to L.B.M.), and the National Institute on Drug Abuse Intramural Research Program (ZIA DA000642 to T.K.). The opinions expressed in this work are the authors’ own and do not reflect the view of the National Institutes of Health/Department of Health and Human Services.
↵*L.K.S. and L.B.M. contributed equally to this work.
The authors declare no competing financial interests.
- Correspondence should be addressed to Laura K. Shanahan at shanahanl{at}rhodes.edu or Thorsten Kahnt at thorsten.kahnt{at}nih.gov.
