Introduction
In recent years, canine neuroimaging has seen a rise in interest (Berns et al.
2012; Bunford et al.
2017; Huber and Lamm
2017). The dog (
canis lupus familiaris) as a model organism provides us with interesting features, such as similar living conditions as humans (ManyDogs Project et al.,
2023) and a remarkable social cognitive ability (Hare and Tomasello
2005; ManyDogs Project et al.,
2023; Topál et al.
2009).
To understand how higher order cognition is processed in the dog´s brain, we also need to understand the brain´s fundamental organization and how it processes sensory input at the lower levels. To this end, recent research has mapped out the visual, olfactory and auditory cortices in awake and unrestrained dogs using functional magnetic resonance imaging (fMRI; Andics et al.
2014,
2016; Boch et al.
2021,
2023; Bunford et al.
2020; Cuaya et al.
2016,
2022; Dilks et al.
2015; Gillette et al.
2022; Jia et al.
2014; Phillips et al.
2022). However, our best understanding of the canine somatosensory cortex dates back to almost 70 years ago (Fritsch & Hitzig, 1870/
1963; Hamuy et al.
1956), using invasive methods, a small sample and focusing on selected parts of the canine cortex. Invasive methods have limitations in the investigation of the somatosensory cortex, even in humans (see Gordon et al.
2023, who found that a revision of the classic homunculus from Penfield and Boldrey
1937was needed). Particularly whenever lesions are created, not only the welfare of the (animal) test subject is heavily disrupted, but lesions can additionally disrupt normal brain function, and therefore also disturb scientific results. In this study, we examined somatosensory processing in a sample of healthy and awake pet dogs. Our main aims were to understand how a non-primate mammal processes touch, whether their touch processing is lateralized, and finding out what parts of the brain may be involved beyond the primary and secondary somatosensory areas. With this approach, our study aimed to provide another important puzzle piece in comprehensively mapping and understanding dogs’ functional neuroanatomy in vivo and using non-invasive methodology.
During fMRI scanning, dogs were dynamically touched using a wooden rod moved down their left and right flanks. In line with our preregistration (
https://osf.io/4gs9d/), we expected to see activation in the primary and secondary somatosensory cortices in response to touch (hypothesis 1), and that this activation would be higher in the hemisphere contralateral to the touched flank in primary somatosensory cortex SI, which in the dog is comprised of the postcruciate and rostral suprasylvian gyri, and in the ipsilateral hemisphere for secondary somatosensory cortex SII, the rostral ectosylvian gyri (Hamuy et al.
1956). Additionally, we hypothesized potential ipsilateral activation in the cerebellum, due to its (ipsilateral) involvement in motion control and potential involvement in suppressing touch elicited motion in our task (Uemura
2015). Importantly, since as of yet, our knowledge of canine brain function is still limited, we were also interested in uncovering additional brain areas related to in vivo touch processing. Hypothesis 2 focused on whether somatotopic mapping could be achieved with the relatively coarse temporal resolution of fMRI (Shmuel et al.
2007). In particular, we were interested in how activation shifts as a function of the dynamic somatosensory stimulation, possibly showing the trajectory of activation along the receptive fields coding for the parts of the back that were being stimulated, but also the progression of the signal into other parts of the canine cortex, to higher-order processing steps.
Moreover, we were interested in lateralization of somatosensory processing (hypothesis 3). While dogs may not possess a population wide side preference for one paw (Demirbas et al.
2023; Ocklenburg et al.
2019; Wells et al.
2018; but see Laverack et al.
2021), like humans favoring their right limbs (Papadatou-Pastou et al.
2020) or great apes do (Güntürkün et al.
2020; Hobaiter and Byrne
2013; Hopkins
2006), individually, most dogs seem to favor one over the other limb consistently (Ocklenburg et al.
2019). Even though somatosensation is not to be equated to motor behavior, we were interested in seeing whether somatosensation may be processed more strongly by one over another hemisphere within individuals (by looking at laterality quotients, hypothesis 3). Based on the existing literature for motor biases (Charlton and Frasnelli
2023; Ocklenburg et al.
2019; Simon et al.
2022), we expected the majority of dogs in our sample to show lateralized somatosensory processing.
Discussion
To understand touch processing in awake, healthy dogs, we touched dogs on their right and left flanks while undergoing non-invasive brain scanning using fMRI. Building on invasive work from the 1950s, we found activation (and de-activation) in areas previously identified as SI and SII of the dog, but also activation in other brain areas. We also found evidence for lateralization of responses and a shift of functional activity hotspots in line with the dynamics of the stimulation.
Firstly, we found clusters in the left and right hemisphere that responded more strongly to touch than baseline (no touch), as well as in a medial cluster. However, in addition to the hypothesized activation, we found additional areas outside the somatosensory areas that responded to touch stimulation. Touch activated the rostral and caudal sylvian gyri left and right, areas ventral and slightly caudal to SII, as well as the right cingulate cortex and right precruciate. The rostral sylvian gyrus has been linked to emotion perception (Hernandez et al. 2017; Karl et al.
2021), the detection of familiar vs. unfamiliar speech (Cuaya et al.
2022), or action observation (Boch et al.
2023), and therefore seems to act as a multisensory area that could play a role in sensory integration and association processes (see also Boch, Huber, et al.
2023 for review). Interpreting the findings, it seems unlikely that the areas outside of SI and SII would be directly related to primary sensory processing, based on findings in dogs but also other mammalian species, when it comes to touch processing. Rather, they might be related to higher-order processing of touch. In the case of the gyrus cinguli, affective processing of the touch stimulation may be localized there, as the gyrus cinguli has been shown to relate to affective and even social processes in both humans and non-human primates (Devinsky et al.
1995; Rudebeck et al.
2006; Vogt et al.
2005) but also to motor control (Wang et al.
2001). Little is known about the functionality of the canine cingulate cortex so far, but it may play a role in integration of information and functional processing (Szabó et al.
2023). Additionally, the central cluster was biased towards the right side. In dogs, different side biases for processing of sensory inputs exist, and namely, the right hemisphere may play a larger role in processing arousing (or negative) stimuli (Simon et al.
2022; Siniscalchi et al.
2017). Through the domestication process and intensive handling by humans, experienced by most dogs, it is conceivable that their brains show particular adaptations to process touch administered by humans in a social way, which may be linked to increased arousal. Finding involvement of higher order cognitive processes highlights the methodological advantage of using a non-invasive method in healthy functioning animals: these findings can be made impossible when other parts of the brain are removed or damaged surgically (Hamuy et al.
1956; Marshall et al.
1937).
When contrasting left or right flank touch separately to baseline, we did find major activation in the contralateral hemispheres in locations consistent with SI/SII (rostral suprasylvian and ectosylvian gyri right, left suprasylvian gyrus). Ipsilateral processing was only found for the left touch, and also only at trend level in the SVC analysis, but remained absent for right flank touch. Peak locations of activations were almost identical for both hemispheres. These results suggest a strong preference for contralateral processing and potentially a bigger role of SI in flank stimulation, in line with previous findings on touch processing in mammals (e.g. Catania and Remple
2002; Santiago et al.
2007; Welker
1976).
We also found cerebellar activation, particularly in the vermis, which is generally associated with motor control and conditioning (Glickstein and Yeo
1990; Ohyama et al.
2003): both aspects may have been relevant here, as the dogs are both relying on their learned behavior to lie still in the scanner and not respond with an orienting response to the touch stimulation, but also need to actively suppress motor responses. This is also in line with the observed deactivation in the postcruciate gyrus (rostral SI right). Interestingly, we found the deactivation only in the right hemisphere, for both directions of touch. Due to its proximity to areas that have been suggested to play a role in motion (precruciate gyrus), it may be that the right postcruciate gyrus stands in close connection to downregulation of precruciate activity, playing a role in the inhibition of motor responses (but see Cook et al.
2016, which showed frontal areas involved in inhibition, however no precruciate activity).
Our second hypothesis related to the possibility of tracking activation shifts across the 4s stimulation period starting at the shoulder and ending at the hip. Contrasting the first and third time bin, we found activation in the left and right hemispheres as well as a rostro-central cluster, aligning with parts of the postcruciate gyrus left and right: Importantly with time, processing moved away from the primary sensory areas and towards potentially higher-order association related parts of the cortex (see Fig.
7), as revealed by higher engagement of rostro-central cortical areas. Thus, the FIR analysis approach allows some insights into the progression of sensory signal processing across higher-order cortical areas (i.e., outside the primary and secondary somatosensory cortices). However, possibly due to the lower resolution representation of the back (see e.g. Gordon et al., 2020), and an ebbing off of activity after the initial touch in the primary cortex, other touch procedures (different start points and different body parts) should be considered by future studies. This will also be important for mapping caninculi in a more comprehensive fashion. Note that one reason we had chosen to stimulate the back of the dogs in this first exploratory study on somatosensory responses were the feasibility of stimulation and their tolerance by the dogs in the challenging MRI environment.
Finally, our third aim was to investigate whether dogs show lateralization (hemispheric dominance) of somatosensory processing. Already in the GLM analysis, we found some effects that indicated substantial lateralization, such as the deactivation in the right postcruciate gyrus. While somatosensory processing should not be equated with lateralized motor preferences, such as pawedness, which are more commonly investigated, lateralization in somatosensory processing can be related to motor biases and could be indicative of a generalized feature shared by many vertebrates (Güntürkün et al.
2020; Rogers et al.
2013). Roughly 86% of dogs (19 out of 22) showed a lateralization bias towards one hemisphere, as quantified through beta weights, a significant deviation from 0, with roughly equal numbers for left and right hemispheric biases. While there is no population wide preference for use of the left or right paw, around 70% of dogs generally show an individual side preference. This number is lower, but in a similar ballpark, than the number of dogs that showed a lateralized processing in our sample here. Thus, sensory processing is lateralized in most dogs. These data give no reason to assume a population wide side bias of processing, in contrast to a leftward grey matter volume bias in dogs in general (Barton et al.
2023), suggesting a complex relation between functional and structural asymmetries.
Canine fMRI research is often limited by its sample sizes, an issue we circumvented by extending data collection until reaching a sample of 22 dogs, however, the future of canine and comparative neuroscience should be collaborative, to maximize power for important questions on canine cognition and cognitive evolution (ManyDogs Project et al.,
2023). Our main interest lay in the identification of touch elicited cortical activation against baseline, a very strong effect for which we deemed our sample size to be more than sufficient. However, statistical power with the current sample size may have been suboptimal for some analyses, but in particular the FIR analysis which comes with a number of statistical tests that need to be corrected for (thus decreasing power; see also Huber and Lamm
2017, for further discussion regarding issues of power and sensitivity in dog fMRI research).
In our approach of touch stimulation, a few limitations should be noted: first of all, the stimulations were performed by the dog trainers, and not a robotic apparatus. While timings and the procedure itself were both practiced and cued, it is very likely that a degree of variability in stimulations was present, which is of relevance to the FIR analysis, in addition to the coarse neural somatosensory representation of the back in comparison to other areas (Coq et al.
2004; Kaas
1993; Woolsey et al.
1942). We chose this area pragmatically: stimulating the dog from the front would have brought the trainer into the view of the dog, possibly inducing distraction and other effects. From the back of the scanner, only the flanks and the paws were accessible, and many dogs do not like being touched on their paws and show this by moving them out of reach. Thus, only the back stimulation was viable. However, even though the back may not be the somatosensorily best represented area of the dog´s body, we were still able to find differences between touch bins on this area. Finally, for the analysis of lateralization, within fMRI research, it is a continuing debate how to proceed with negative values. Since no consensus exists in the literature, one can a) look at negative and positive values separately, or b) use absolute values in the denominator of the fraction used to calculate laterality indeces (Seghier
2008). We opted for the latter version, since we would have shrunk the sample if only looking at either negative or positive values (not all dogs
had both positive and negative values). The accuracy of activation localizations in fMRI is always limited by the signal that can be collected. While we were able to collect data at a (relatively high) resolution of 1.5 × 1.5. x 2 mm
3 functionally, dogs´ brains are also quite small, and vary in shape depending on breed. However, our localizations are based on relatively large anatomical regions of interest (see Fig.
2), within a small range of dog sizes and relatively uniform breeds (all dogs were mesaticephalic). We thus suggest that the spatial resolution was well-matched to detect our main areas of
a priori interest.
In general, however, we want to point out the strengths of (f)MRI in the investigation of non-human cognition. While fMRI may seem limited to humans, or highly trainable species, advances have been made with a broad range of animals, from farm animals (Pluchot et al.
2023), to crocodiles (Behroozi et al.
2018), and birds (Behroozi et al.,
2019; DeGroof et al.
2013). MRI allows us to not only investigate the overall structural anatomy of brains, but also their function in a non-invasive manner. This not only comes at the benefit of animal welfare, making sure the animal can continue living healthy lives, in this case with their human caregivers, but also enhances the bandwidth of possible findings: Especially links to higher-order processing areas that may be unexpected can be hindered by invasive studies, which require pathways and areas to be partly lesioned, or which only allow to target specific pre-selected brain areas (while MRI allows assessment of the whole brain at once). Thus, using fMRI, it is possible to investigate not only lateralization, but most importantly, higher-order processing of the intact brain.
Using fMRI, we confirmed the roles of the dog’s postcruciate and rostral suprasylvian gyri (SI) as well as the rostral ectosylvian gyri (SII) in the processing of touch. Beyond these areas that are likely most closely tied to sensory-perceptual computation, we pinpointed additional areas, namely parts of the rostral cingulate cortex, which may relate to affective processing, as well as the sylvian gyri, which may play a putative role in higher-order or associative somatosensory processing. We found somatosensory processing to be lateralized in most dogs, with the sample being evenly split between left-, and right-hemisphere biased dogs. First indications for a somatotopic organization based on analyses of activation dynamics need further corroboration. Our findings add another puzzle piece in the dogs’ functional neuroanatomy and will be helpful in charting brain function across diverse species.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.