1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 This author manuscript is copyrighted and published by Elsevier. It is posted here by agreement between Elsevier and MTA. The definitive version of the text was subsequently published in [Trends in Neurosciences, 40: (7), July 2017, DOI: https://doi.org/10.1016/j.tins.2017.05.003]. Available under license CC-BY-NC-ND. 29 biological underpinnings thereof. The domestic dog (Canis familiaris) is a promising model in 30 cognitive neuroscience. However, before it can contribute to advancements in such science in a 31 relevantly comparative, reliable, and valid manner, methodological questions warrant attention. 32 To base the research on rigorous foundations, we review non-invasive canine neuroscience 33 studies, primarily focusing on 1) variability across dogs and between dogs and humans in cranial 34 characteristics and 2) generalizability across dog and dog-human studies. Arguing not for 35 methodological uniformity but for functional comparability in study methods, experimental 36 design, and neural responses, we conclude that the dog may become an innovative and unique 37 model in comparative cognitive neuroscience, one that is complementary to traditional models. Canis familiaris as model for non-invasive comparative neuroscience Nóra Bunford1*, Attila Andics1,2, Anna Kis3, Ádám Miklósi1,2, and Márta Gácsi1,2 Eötvös Loránd University, Institute of Biology, Department of Ethology, 1117 Budapest, Pázmány Péter sétány 1/C 2 MTA-ELTE Comparative Ethology Research Group, 1117 Budapest, Pázmány Péter sétány 1/C 3 Institute of Cognitive Neuroscience and Psychology, Hungarian Academy of Sciences, 1117 Budapest, Magyar tudósok krt 2. 1 *Correspondence: bunfordnora@caesar.elte.hu (N. Bunford). Keywords: animal model, domestic dog, non-invasive neuroscience, comparative neuroscience, fMRI, EEG Abstract: There is ongoing need to identify and improve animal models of human behaviour and 2 38 Animal models in comparative neuroscience 39 Animal model research is grounded in the idea that animals share behavioural, physiological, and 40 other characteristics with humans. Benefits of such research include increased understanding of 41 phenomena that could not be directly studied in humans or without cross-species comparison. The 42 neuroscience of socio-cognition has been extended from traditional primate and rodent models to the 43 domestic dog – an alternative, complementary model that permits for non-invasive measurement of 44 behaviour and its neural correlates. There has been an upsurge in canine neuroscience studies, 45 necessitating establishment of methodological guidelines that ensure scientific rigor. To this end, 46 complementing available reviews that are heavily [1] or solely [2] focused on available fMRI findings 47 [1,2] from a conceptual perspective, we review the non-invasive canine neuroscience literature, focusing 48 on methodology and experimental design. Primarily guided by principles of comparative anatomy, we 49 highlight advantages of and remaining challenges of the dog as an animal model for comparative 50 cognitive neuroscience. 51 We begin with an overview of animal models of human behaviour, then narrow our focus into 52 neuroscience, leading to questions about the domestic dog as a model for comparative neuroscience. 53 Mainly focusing on non-invasive canine fMRI and EEG research, we reflect on such questions in light of 54 three main considerations. These centre on within- and between-species variability, in particular in cranial 55 characteristics, though are also varied in terms of the degree to which they potentiate (1) advantages and 56 disadvantages for the dog as an animal model and, in case of disadvantages, whether solutions (2) have or 57 (3) have not been developed to address those. 58 Animal models for comparative cognitive neuroscience 59 A goal of comparative research is to establish principles of proximate and ultimate causation 60 (see Glossary), via between-species comparisons and study of individual organisms. Animal models for 61 comparative cognitive science include avian [3–5] as well as rodent and primate models that have 62 emerged as primary models for comparative cognitive neuroscience [2]. Advantages of rodents include 63 feasibility of handling the animals under laboratory conditions; cost-efficiency; and utility in pre-clinical 3 64 and clinical studies [6]. Advantages of primates include similarity to humans in development, 65 neuroanatomy, physiology, and reproduction, as well as in cognition and social complexity and thus 66 suitability for studying a range of mental processes [7]. Yet, use of these models is increasingly 67 problematic for animal welfare and ethical reasons [8]. Conversely, the role of the domestic dog has been 68 becoming increasingly important, with research initially focused on informing treatment for human 69 medical diseases with laboratory dogs [e.g., 4] and more recently involving basic research on sensation, 70 perception, and socio-cognition with family dogs (Box 1). One reason for this increase in importance is 71 that dogs, having been encultured in human society, naturally exhibit cooperativeness and trainability, 72 obviating need for fluid and/or food restriction as a motivational tool. Thus, relative to other species, 73 preparation of the dog for an experiment is more similar to preparation of humans in terms of 74 corresponding physiological and social state and there is less limitation to generalizability of interaction 75 with experimenters and environmental (e.g., lighting and sound) and experimental stimuli [1]. 76 Cooperativeness and trainability also permit for non-invasive methods; although techniques have been 77 developed for awake scanning of monkeys, pigeons, and rats [1], unlike these animals but like humans, 78 dogs do not need to be restrained (e.g., via surgically implanted posts [10]) but can be trained to hold still, 79 yielding more valid cross-species comparisons. Finally, given their evolutionary history and integration 80 with humans, dogs and humans exhibit a range of socio-cognitive skills that share key behavioural and 81 functional characteristics [11]. It is for ability to study these very skills and corresponding functions (Box 82 1) that the dog may be one of the best model species for study of human socio-cognition [2] in 83 comparative neuroscience [11]. 84 Together, it stands to reason that the domestic dog is a suitable model for comparative neuroscience 85 and that the non-invasive methods of brain circuits, physiology, and behaviour used with the dog ideally 86 complement the invasive methods appropriate for studying molecules and cells used with traditional 87 models. In combination with over 20 years of canine ethological research [12] and capitalizing on 88 exciting possibilities of the species and non-invasive methods, there has been an increase in the number of 4 89 canine neuroscience studies, with an overwhelming majority conducted in the past 3 years. (Mostly fMRI 90 and EEG, although other methods have also been used [13]). 91 Basic standards for measures and methods include reliability and validity [14] and, in case of 92 comparative research, for them also to be relevantly comparative. Related pressing questions pertain to 93 the degree to which methods are comparable across dog-dog and dog-human studies as well as the degree 94 to which employed methods allow for comparability and generalizability across studies (Table 1, Key 95 Table); with the impetus behind such questions stemming from within- and between-species variability, 96 especially in cranial characteristics. Some of this variability presents advantages for the dog as a model 97 and some may be limiting. In the latter cases, methods to address limitations are either already being 98 developed and evaluated or are in need of development and evaluation. 99 Differences that present advantages 100 Differences in skull formation and brain anatomy. Across humans, variation in skull formation 101 and brain size is relatively trivial; the average female brain volume is 90% of the male [15] and the 102 average brain volume of a 7-11-year old child is 95% of the volume of a sex-matched adult [16]. 103 Conversely, there are large differences across dogs in skull shape and size and brain anatomy. Canine 104 skull length ranges from 7 to 28 cms [17] (i.e., the shortest dog skull is 25% of the longest), making Canis 105 familiaris the species with most within-species morphological variation in this regard [18]. 106 In addition to skull length, differences across dolichocephalic, brachycephalic, and 107 mesaticephalic dogs include dissimilarities in the craniofacial angle (angle between the basilar axis and 108 hard palate) [19], in neuroanatomy (e.g., in brachycephalic dogs the brain is rotated with respect to its 109 mediolateral axis) and the anatomy of the cerebral cortex [20], temporomandibular joint (i.e., jaw joint) 110 [21], and cribriform plate [22]. 111 These differences across dogs allow for examining the relation among brain structure, function, 112 and behaviour within the same species and the effects of differences in skull- and brain-morphology on 113 neuro-socio-cognition. As the ≥400 documented breeds exhibit a variety of genetically fixed morphologic 114 traits that correspond to differences in behaviour, longevity, size, skull shape, and disease susceptibility 5 115 [20], better understanding of these was proposed to increase understanding of mammalian biological and 116 embryonic development [20]. Although, to date, the number of dog breeds involved in fMRI studies is 117 considerably lower, they include subjects from diverse breeds suggesting that there is no limitation (e.g. 118 in trainability) to between-breed comparisons. 119 In support of stated advantages, differences in dog skull shape are associated with differences in 120 brain organization, e.g., brachycephalic brains are relatively rounded and shortened in the anterior- 121 posterior plane, the brain pitched ventrally at the anterior pole, with a pronounced shift in the position of 122 the olfactory lobe [18] (see Box 2 for additional examples). Differences in skull shape are further 123 associated with differences in behaviour in that brachycephaly, relative to dolichocephaly, is associated 124 with increased ability to focus and rely on human gestures [23]. Conversely, less morphological 125 differences across individuals in other species, such as humans, are less (or not) suitable for addressing 126 these questions and are thus largely overlooked. 127 Differences in experimental design: sample composition. Compared to the human neuroscience 128 literature, there is significant overlap in groups of dogs across studies. This is due, in part, to challenges 129 (e.g., limited subject availability and need for extensive training) and, in part, to advantages that make the 130 dog a multi-experiment model (e.g., ability to re-measure dogs as they do not need to be euthanized after 131 participation). For example, in canine fMRI studies, 100% of the sample of [24] was included in [25], and 132 there was a 92% overlap in the samples of [25] and [26], and a 67% overlap in the samples of [25] and 133 [27], and all dogs in [28] came from one of these samples. Similarly, in EEG studies, there was a 100% 134 overlap in the samples of [29] and [30], and a 68% overlap in the samples of [31] and [32]. 135 Awake fMRI testing necessitates that dogs are trained to get used to scanner coil; place their heads 136 in-between their paws [34,35,37–39] or on a chinrest [24–28,33,36,40], and hold this position until a 137 release signal and then while wearing canine ear muffs; get used to recordings of scanner noise and being 138 in a mock scanner; and to adhere to these procedures inside the scanner room and ultimately the scanner 139 [25,34]. Training is extensive and typically involves behavioural shaping, conditioning and social 140 learning (e.g., the “Model/Rival” training method [34]). Different training methods allow for different 6 141 lengths of time during which dogs are able to hold a position, which has implications for design. For 142 example, in some studies, consistent with human studies, dogs do not exit the scanner between runs 143 [34,35,37,39,41] whereas in others, they do [24–28,33,36,40]. Movement artefacts are also handled 144 differently: some authors, consistent with human studies, exclude scans with head translation >3mm or 145 rotation >1⁰ [34–36]; whereas others exclude scans with >1% scan-to-scan signal change [24]; >0.1 146 fraction of outlier voxels in each volume or >1% scan-to-scan signal change, in combination with >1mm 147 scan-to-scan displacement [26–28,40,42], and yet others exclude runs with .10mm total displacement 148 [37,41]. As the size of the dog brain is roughly one-third of the human, arguably, a >3mm translation in 149 dogs would approximate an unacceptable >9mm in humans. However, in most studies where the human 150 criteria were used, translation did not exceed 1mm [34,43]. Additionally, it has been shown that changes 151 in the time course of fMRI data are decreased when correlations are examined long-distance but increased 152 when they are examined short-distance, indicating that absolute movement is less and relative movement 153 is more important when pre-processing the data [44]. Finally, depending on study design and research 154 group, dogs need anywhere from five sessions [34] to 18 months of training [25]. For comparison, human 155 adults do not receive training and human children as young as 6 years of age receive minimal (a one-, 156 maximum two-occasion, 30-60-minute familiarization with a mock-scanner and recordings of scanner 157 noise) or no training [45] (Table 1). 158 The overlap in groups of dogs included across studies also has advantages for examination of 159 reliability and validity of measures as it allows for assessment of within-subject stability vs. change of 160 measures of neural function over time and of within-subject correspondence of neural correlates and 161 performance across social, cognitive, and affective paradigms. This ability to examine psychometric 162 properties of measures is comparable to research with humans but not most other species, where animals 163 easily habituate or are euthanized following participation. Regarding within-subject stability vs. change 164 over time, although the reliability, including test-retest reliability, of neuroimaging [14] has, until 165 recently, been a relatively neglected area of research in human neuroscience, the overlap in groups of 166 dogs across canine studies presents a natural opportunity to attend to questions of psychometrics [46]. 7 167 Regarding within-subject correspondence of neural correlates and performance across paradigms, 168 it is important that these exhibit convergence and divergence, where expected. Establishing 169 correspondence across different indices of phenomena of interest (e.g., social and cognitive indices of 170 self-regulation) but that these provide unique information about variables examined, is key to the 171 innovative dimensional frameworks that are currently championed (e.g., the Research Domain Criteria 172 [RDoC]; [47]). 173 Differences that potentiate disadvantages but solutions are available 174 Within-species differences in skull formation and brain anatomy. Thess within-species 175 variabilities (Figure 1) are relevant for normalization. In fMRI research, advantages of normalization are 176 that when a set of coordinates is referenced, the location to which those coordinates correspond is known 177 and that results can be: generalized to a larger population; compared across studies wherein the same 178 brain is used for normalization; and can be averaged across subjects for group-level analyses. 179 Disadvantages are that it reduces spatial resolution and increases probability of error in identification of 180 anatomical location. 181 Normalization requires a “standard” brain, i.e., template. In the adult human literature, the Montreal 182 Neurological Institute (MNI) template (MNI305) is commonly used (Table 1), which is based on 183 combination of 152 healthy adult MRI scans [48]. Given relatively little difference between adult and 184 child brains, the MNI-305 is suitable for use with children over age 6 years [49] and empirical studies 185 have generally followed suit, with some attempts at developing a child template for use with a wider 186 range of ages (e.g., from 2 weeks to 4.3 years [50] and 4.5 years through 19.5 years (on age increments of 187 6 months [51]). Conversely, at present, there is no widely-accepted and used dog template. Authors of 188 canine fMRI studies have addressed this issue by omitting group-level analyses altogether or, where 189 group-level analyses were conducted, by using the brain of a selected individual, or using a template 190 based on the brains of 15 mesaticephalic dogs (Table 1). 191 Besides the said advantages of population-based templates, there are advantages of study-specific 192 templates [52] (a special case of which is use of the brain of a selected individual). Regarding the Datta 8 193 atlas [53], one limitation is that head length and width may influence cortical folding in a manner that an 194 affine transformation of brain size may not correct for, indicating that the Datta template may not be 195 appropriate for non-mesaticephalic animals. 196 Challenges resulting from within-species differences in skull formation and brain anatomy across 197 dogs have been addressed differently in canine continuous EEG and in event-related potential (ERP) 198 studies. Regarding continuous EEG, presumably due to differences in skull morphology (e.g., thickness 199 of the frontal and parietal bones), absolute EEG power (μV2) varies greatly across dogs (e.g., 3-fold 200 across our samples; [31,32]). As a result, group-level analyses are best conducted using relative EEG 201 spectrum values [31,32], which is common practice in human EEG studies as absolute EEG power is less 202 psychometrically sound than relative EEG power. Regarding ERP research, challenges have been 203 addressed either via use of a homogenous group of dogs (e.g., laboratory-bred and -kept beagles, all of the 204 same age and similar weight [29,30]) or via report of results at the level of individual dogs [54]. 205 Relevant for both continuous EEG and ERP studies, an additional methodological issue is electrode 206 placement. Despite canine methods having been adopted from human studies, given variability in dog 207 head shape and size, the distance between electrodes placed on anatomical landmarks is different across 208 dogs. Although this difference is difficult to address, such variation in absolute distances are compatible 209 with the International 10-20 system used in human studies [55], which keeps not the absolute but the 210 relative distance between electrodes constant. 211 Between-species differences in skull formation and brain anatomy (Box 2; Figure 1). In fMRI, 212 these differences highlight consideration related to correction for multiple comparisons (Box 3). Given 213 smaller brain volume of dogs relative to humans, the multiple comparison problem is less relevant in 214 canine fMRI. If correction that takes voxel number into account is used in a human and a dog study or 215 across dog studies, results are comparable. If correction that does not take such number into account is 216 used, it is important that the search area is comparable in size. Both are feasible. Nevertheless, although 217 there are widely used methods for correction in human studies and these are now employed in most (if not 218 all) adult and child studies [56], there is heterogeneity across dog studies (Table 1). No meaningful 9 219 comparison can be made between results obtained without and with correction, with varying degrees of 220 stringency. If and when the aim is to compare results, consistency across studies will be important. 221 In EEG research, differences between dog and human skull and brain morphology necessitate 222 differences in electrode placement. Because dogs have a smaller but more muscular head than humans, 223 their heads permit less sites for electrode placement. The number of electrode holders in human EEG 224 head caps range from 16 to 256 compared to 3 [54,57], 4 [32], or 5-7 [29–31] electrodes placed on dogs’ 225 heads. Nevertheless, as these sites correspond to human electrode sites, a functional comparison between 226 species can be made, even if restricted to a small number of EEG channels, which may be further 227 increased with methodological advancements. 228 Differences in experimental design: sample composition. Available findings having been obtained 229 with a small group of dogs and the noted overlap in included dogs may be disadvantageous for 230 generalizability to larger dog populations. This can be addressed through sample selection that increases 231 generalizability potential, e.g., ensuring that dogs of different ages, breeds, sexes, and level of prior 232 training (e.g., from training-naïve to service dogs), are included and then tested. Selection of a 233 biologically and demographically heterogeneous sample with variation in training history has been 234 attended to with varying degrees, with some variability in laboratory [29,30,37,39,41] vs. family [24– 235 28,31–36,40,54,57] dogs, single [29,30,37,39,41,57] vs. multiple [24–27,31,32,34–36,40,54] breeds (with 236 [28,33] not specified), and ages ranging from 1 to 12 years. 237 The noted small sample sizes and overlap in included dogs also means a very small overall number 238 of tested dogs. The sample sizes of all but one [32] canine neuroscience studies published to date are <15, 239 leaving the research underpowered and effects difficult to detect. Although the obtained results may 240 reflect effects that are so large and robust that they are detectable even with small samples, they may 241 alternatively reflect effects that are fragile, non-generalizable, or spurious. Power analysis indicates that 242 larger samples are needed for confidence in results [58]. Yet, it is also the case that in early and 243 exploratory stages of a research area, small N studies are not only warranted but also desired to establish 244 that larger (necessitating more funds and participant and researcher time) studies are indicated. 10 245 Differences that potentiate disadvantages and solution need to be identified 246 Between-species differences in skull formation, brain anatomy, and physiology. Although further 247 research is needed about the degree to which dogs’ anatomical structures and circuits correspond to 248 humans’, knowledge about canine brain anatomy and the similarities between such anatomy and that of 249 humans’ is encouraging regarding the dog as an animal model in comparative neuroscience. There is 250 evidence of correspondence between the species in, for example, primary sensory areas and associated 251 functions [34]. Yet, whether other areas, especially the frontal and prefrontal cortex are organized in a 252 manner that allows for characterization of structures and circuits as associated with similar cognitive 253 functions across dogs and humans is largely unknown. As such, when a specific human structure is 254 referenced (e.g., rostral anterior cingulate cortex [rACC] or dorsolateral prefrontal cortex [DLPFC]), it is, 255 at present, unclear whether the rACC in dogs is anatomically delineable from other areas of the ACC and 256 functionally (e.g., attentional control over emotional conflict or distracters [46,59]) the same or at least 257 meaningfully comparable across the species. 258 The solution to this challenge is unclear as from a biological perspective, there is no “reference 259 species” that is uniformly appropriate for addressing pertinent questions. Would it be prudent to take 260 rodents as a reference? Although rodent brains are more dissimilar from human brains than dog brains, 261 evidence obtained via invasive methods indicates correspondence in certain structures across rodents and 262 humans [60]. Alternatively, would it be useful to take humans as a reference and identify areas of 263 activation to stimuli, present dogs with comparable stimuli and search for correspondence in the canine 264 brain? Then again, in addition to or instead, is there need for research that identifies parallels through 265 ontogeny? For example, although there are differences between birds and apes in neural structures, e.g., 266 birds do not have a cerebral cortex for processing complex mental tasks [5], both species have prefrontal 267 structures that control comparable executive functions [5]. It has been argued that these similarities either 268 originated from the last common ancestor passing down neuronal bases of executive functions or evolved 269 independently due to the species facing similar challenges [5]. 11 270 Between-species differences in skull formation and brain anatomy are also source of 271 methodological shortcomings in fMRI as the obtained images are of poor quality due to use of 272 radiofrequency (RF) coils (human head/neck coils [24–28,33] or knee coils [34–39]) whose geometries 273 have been optimized for different purposes and have not been tailored to dogs’ heads and neuroanatomy, 274 making them less than ideal for canine fMRI. Together, as was the case with other species (e.g., 275 marmosets, rats, mice, and rhesus monkeys) where use of dedicated animal coils has been shown to 276 improve signal-to-noise ratio (SNR) [10], there is need for development of dedicated dog coils that satisfy 277 the anatomical constraints imposed by these animals. Until such coils are available, it will be important 278 for research to determine which coil type is best for performing fMRI in awake dogs with sensitivity, 279 specificity, and large functional contrast-to-noise ratio [1]. 280 Between-species differences in cranial musculature and size are relevant for artefact rejection in 281 EEG (Box 4). In human studies, artefact rejection includes correction for ocular artefacts and quantitative 282 procedures (e.g., removing artefacts with voltage step between sample points that is greater than e.g., 283 50μV; with voltage difference of e.g., 300μV within a trial; and maximum voltage difference within e.g., 284 100msec intervals of e.g., <0.5μV [61]) and rejection via visual inspection. In dog studies, there are no 285 well-established quantitative procedures, given difficulty in distinguishing muscle artefact from EEG 286 signal and artefact rejection is typically done using simpler methods. The authors of ERP studies used 287 only a single crude method [62] for rejecting trials with artefacts, in which a trial is rejected if the voltage 288 during the epoch exceeds a user-defined threshold (amplitudes higher than 100μV [54,57] or 200μV 289 [29,30]) and the authors of sleep EEG studies conduct artefact rejection by visual inspection only [31,32]. 290 Although the user-defined method works for rejection of artefacts resulting from blinks, it is 291 inadequate for detecting more subtle artefacts, such as those resulting from eye (or ear) movements [62]. 292 As such, the used methods are problematic for awake continuous EEG measurement and ERP data 293 collection where there is need for more stringent artefact rejection, given greater canine cranial muscle 294 mass; another example where methodological uniformity between human and dog studies is neither 295 possible, nor warranted. As an example, if the dog moves its eye (or ear) every time there is an event (i.e., 12 296 stimulus), it is difficult to determine whether what appears to be a voltage change reflects the movement 297 or differential neural activation. It may be for this reason that there are no established methods for non- 298 invasive measurement of ERPs in dogs, albeit some non- [29,30] and semi-invasive studies suggest 299 progress [42,44]. 300 Potential solutions to the artefact problem in non-invasive canine ERP research is to collect data 301 from dogs with less cranial muscle and/or in a state of drowsiness (i.e., canine equivalent of light sleep) or 302 sleep. In support, the Mismatch Negativity (MMN) component can be elicited during light sleep in 303 humans [63,64], indicating an auditory ERP method may be useable with drowsy dogs. Notably, dogs 304 spend at least 30 minutes in drowsiness during a 3-hour-long spontaneous EEG recording [32]. Not unlike 305 sleep, drowsiness is characterized by lowered muscle tone, indicating it permits a considerable amount of 306 artefact-free EEG data that ERP studies could potentially capitalize upon. 307 Between-species differences are pertinent beyond skull formation, brain anatomy and include 308 differences in resting state physiology. Specifically, normal respiratory rate in newborn puppies may be as 309 low as 15 breaths/minute and in an average adult dog it is 24 breaths/min [65]. Conversely, respiratory 310 rate in human neonates (<1 year old) is 30-40 breaths/min, in older children/young adolescents (5-12- 311 year-olds) it is 20-25 breaths/min [66] and in a healthy adult it is 12–20 breaths/min [67]. With regard to 312 heart rate, <2-week-old puppies have 160-200 beats/min (bpm), ≥2-week-old puppies have up to 220 313 bpm, and adult dogs have 60-140 bpm [65,68]. For comparison, human neonates (<1 year old) have 110- 314 160 bpm and older children and young adolescents (5-12-year-olds) have 80-120 bpm [66]. The heart rate 315 of a healthy adult is between 50–90 bpm [69]. 316 These between-species differences are important as differences in brain shape and size also results 317 in between-species differences in the hemodynamic response function (i.e., the course of the 318 hemodynamic response to an external stimulus – the most common functional imaging signal; HRF) [1] 319 and respiratory rate and heart rate are major sources of fMRI confounds as they are correlated with 320 changes in BOLD signal [70]. The shape of the canine HRF is currently unknown [1] potentially due to 321 the temporal resolution in canine fMRI studies, where repetition time (TR) varies between 1-2secs, which 13 322 is insufficient to sample respiratory or heart rate in dogs. Related, the number of acquired datasets is 323 limited by how long dogs are able to hold still (with experiments necessitating 5- [24], 6- [27,34,40], 7.5- 324 [35], 10- [71], and some 14-minute-runs [26]) (no information is provided in [28,36,39]). As such, the 325 measurement duration that maximizes data quality is unknown. To identify an optimal parameter setup, 326 different anatomical and functional sequence parameters should be tested with phantom and ex-vivo 327 measurements. Similarly, protocols should be optimized with respect to signal- and contrast-to-noise ratio 328 in pilot samples sufficiently similar to the intended experimental samples, but without the constraints on 329 measurement time and motion of in vivo measurements. The ultimate goal of adapting sequence 330 parameters to the dog brain is combination of high spatial and high temporal resolution. Such adaptation 331 will have account for the smaller size of the dog brain, differences in dog compared to human physiology, 332 and limits on run length by how long dogs are able to hold still. Importantly, there are methodological and 333 ethical advantages to shorter runs as these minimize image deterioration due to motion artefacts and 334 prevent rises in specific absorption rates (SAR) of radio frequency levels (see Ethics and Safety) [1]. 335 Differences in skull formation and brain anatomy: within- and between-species. Combined, 336 differences across dogs and between dogs and humans in cranial characteristics will make it difficult to 337 determine whether measured electrocortical signal originates from a meaningfully comparable population 338 of neurons across dogs and dogs and humans. Even the human source localisation literature is in its early 339 stages, with only a few studies on the association between BOLD signal and ERPs recorded during the 340 same session [72]. As the human literature advances, it will be important for canine research to make 341 parallel progress. As noted, little is known about the degree to which certain neural structures in dogs are 342 anatomically and functionally the same as humans’ and advancing the literature in this domain will also 343 be important for source localization. 344 Differences in experimental design: active vs. passive paradigm. In the human neuroscience 345 literature, there are examples of studies where no behavioural response is required (passive task) and 346 where a response is required (active task). From the perspective of introducing additional movement that 347 results in additional motion artefact, as passive tasks do not involve movement, they are not problematic. 14 348 In humans, active tasks are also feasible with behavioural responses like a button press. In dogs, requiring 349 an active response would mean that images obtained following an active condition have to be discarded. 350 Indeed, in all but one of canine fMRI studies, the functions that have been examined are ones that 351 do not necessitate an active response, including in passive auditory paradigms [34,35], passive visual 352 paradigms [24,27,28,36,42], passive olfactory paradigms [37] or, finally, probing resting state activity. In 353 the only canine fMRI study, with an active, go/no-go paradigm, a “go” signal indicated an active 354 behavioural response is to be executed, which, in this case involved dogs touching a target with their 355 noses while in the scanner. When analysing human go/no-go data, go trials are typically compared to no- 356 go trials [73]. Here, however, activation during inhibition trials was compared with activation during 357 neutral trials as successful “go” trials could not be analysed due to the head motion produced by the nose- 358 touch. This is an important limitation to the current state of the canine neuroscience field as there are 359 socio-cognitive functions that are best probed in active paradigms. 360 In addition, the likelihood of prematurely attributing connections between brain structure and 361 function is enhanced when the aim is to separate active and passive processing in dogs, as in the absence 362 of concurrent behavioural response, the relevant cognitive processes are unknown. Being able to 363 differentiate between active and passive processing in dogs will be key, as there are differences in 364 activation to these two forms of processing in humans. One solution to ameliorate risk of reverse 365 inference (i.e., post hoc attribution of presence of a certain cognitive process given activation) is ensuring 366 that dogs have pre-fMRI training on a behavioural paradigm that probes the same cognitive process the 367 fMRI task in question is intended to probe [1] (see, for example, [27]). On a related note, as discussed in 368 relation to the overlap in groups of dogs included across studies, the most ideal assessment battery will 369 comprise measurement methods representing different levels of the measurement continuum (ranging 370 from micro level measurement of brain circuits via fMRI, through less micro level measurement of 371 physiology through EEG, to macro level measurement of observable behaviour via observation or rating 372 scales; [74]) as data obtained at these different levels provide unique information on characteristics of 373 interest [46,61,75–77]. 15 374 Ethics and safety 375 As noted, a main advantage of dogs is that being a domestic animal they can be tested without 376 need for laboratory breeding, raising and keeping. As such, focus on family dogs is what makes the 377 advantage of the dog model ethically permissible. Nevertheless, as aptly discussed by others [1], care 378 should be exercised that no harm is caused, e.g., that scanner noise and high sound pressure levels do not 379 lead to discomfort and hearing damage or that specific absorption rates (SAR) of radio frequencies do not 380 reach harmful levels of rise in tissue temperature [1]. 381 During tests, dogs’ well-being should be continuously monitored and undue stress eliminated 382 both for reasons of ethics and because stress can lead to increases in physiological activity such as 383 increased respiration and tachycardia, which, as noted, may introduce non-neural noise. The techniques 384 used by canine neuroscience laboratories address stress reduction via use of sound-attenuating earmuffs 385 and in training [1]. Stress reduction can be further improved through careful selection of sequence 386 parameters combined with pre- and post-scanning measurement of physiological indices (e.g., cortisol) of 387 stress such as from saliva or urine [1]. SAR should be measured throughout MR scans and in the absence 388 of established guidelines for nonhuman animals, researchers may adhere to standards established for 389 humans. 390 Concluding remarks 391 There has been a notable, recent increase in canine neuroscience studies, necessitating 392 establishment of methodological guidelines and standardisation to inform the next generation of studies in 393 the area. We discussed foremost questions related to methodology and experimental design in the canine 394 neuroscience literature. As a result, we identified areas for further empirical inquiry. Capitalizing on 395 advantages of the dog such as its cooperativeness and trainability, further areas of exploration include the 396 relation among brain structure, function, and behaviour in dogs, within-subject temporal stability of 397 neural measures, and within-subject correspondence of neural correlates. In addition, we suggest to 398 evaluate and performance across social, cognitive, and affective paradigms, in particular probing socio- 399 cognitive skills that share key behavioural and functional characteristics across dogs and humans. 16 400 Regarding challenges for which solutions are already being employed, it will be important that such 401 solutions are adopted and used in a reasonably standardised fashion. Regarding unresolved challenges, it 402 will be important to ensure that samples of dogs reflect variation in the larger population to increase 403 generalizability. Specific to fMRI, it will be key to improve sensitivity of imaging protocols and image 404 quality including via improved spatial and temporal resolution that also allow for sampling heart and 405 respiratory rate as well as development of sequence parameters and dog coils and that are tailored to the 406 specifics of dogs and their neuroanatomy. It is unknown whether non-invasive ERP research is possible 407 with dogs. Addressing this question may necessitate more sophisticated methods either for minimizing 408 eye-movement and muscle artefact during experiments and/or for artefact rejection (e.g., filtering) that is 409 appropriate to the magnitude and type of artefact that occurs in dogs. The degree to which neural 410 structures in dogs are anatomically and functionally comparable to those of humans will need to be 411 established, including to set the stage for future studies with simultaneous neuroimaging and 412 electrophysiological measurement aimed at source localisation. Source localisation will, in turn, help 413 uncover the degree to which what appears to be meaningfully comparable electrode placement across 414 dogs (and across dogs and humans) reflects signal from a meaningfully comparable population of 415 neurons. Regarding difficulty with active behavioural paradigms, methods need to be identified that either 416 permit for dogs to exhibit a behavioural response without data loss or, alternatively, passive paradigms 417 that probe functions that currently can only be manipulated in active paradigms need to be developed. 418 In closing, we argue that, carefully considering inherent advantages, the domestic dog may become 419 an innovative and unique model for comparative cognitive neuroscience. This becomes relevant if the 420 highlighted advancements take place as these will be necessary for measuring the neural bases of canine 421 socio-cognition in a relevantly comparative, reliable, and valid manner. Addressing the noted challenges 422 with dogs appears appreciably more feasible than addressing those with traditional models, such as their 423 non-cooperativeness, them not sharing a social environment with humans, and, in case of primates, cost- 424 inefficiency and paucity. 17 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 Glossary Basilar axis: the axis corresponding to the base of the skull Bradicephalic: short skulled Calvaria: the bone that covers the cranial cavity containing the brain, i.e., the skullcap Continuous EEG: continuous measurement of electrocortical signal, i.e., not measurement of change in such signal in response to a stimulus Cribriform plate: a structure that forms the caudal boundary of the nasal cavity Dolichocephalic: long skulled ERP: measurement of negative and positive voltage changes in electrocortical signal in response to specific events (e.g., stimuli) Gyrencephalic brain: with brain folds (gyri) and grooves (sulci), i.e., folded brain Hard palate: a thin horizontal bony plate of the skull, located in the roof of the mouth Homology: shared ancestry between a pair of genes or structures, in different taxa. A common example is the vertebrate forelimb, where bat wings, primate arms, whale front flippers, and dog forelegs are all derived from the same ancestral tetrapod structure. The opposite of homologous genes or structures are analogous ones, i.e., ones that serve a similar function across two taxa but were not present in their last common ancestor but evolved independently. For example, the wings of a bird and a sycamore maple seed are analogous (but not homologous), as they developed from different structures. International 10–20 system: a method used to describe the location- and guide the application of scalp electrodes in an EEG examination or experiment, based on the relation between placement of an electrode and underlying cortex. The 10-20 system was developed to ensure reproducibility and standardisation. The “10” and “20” refer to the distances between adjacent electrodes being 10% and 20% of the total front–back or right–left distance of the skull, respectively. Lissencephalic brain: without brain folds (gyri) and grooves (sulci), i.e., smooth brain Mesaticephalic: a mesaticephalic skull is neither markedly dolichocephalic or brachycephalic and is of intermediate length and width Model/Rival method: a social learning training method where during the training of an individual, another individual can be present and when the model is rewarded and praised for the wanted behaviour the rival is ignored Prehensile organ: an organ adapted for seizing or grasping especially by wrapping around Proximate causation: an explanation of biological functions and traits in terms of the effects of immediate environmental forces Somatotopic organization: various portions of the body are represented topographically on specific regions of the cerebral gyri Somesthetic cerebral cortex: the primary cortical processing mechanism for sensory information originating at the body-surfaces (e.g., touch) and in deeper tissues such as muscle, tendons, and joint capsules (i.e., position sense). Ultimate causation: an explanation of biological functions and traits in terms of the effects of evolutionary forces 18 468 469 470 471 472 473 474 475 476 477 478 Acknowledgements During the preparation of this article, Nóra Bunford and Márta Gácsi were supported by the National Research, Development and Innovation Office grant (115862 K); Attila Andics, Ádám Miklósi, and Márta Gácsi were supported by a Hungarian Academy of Sciences grant (F01/031); Attila Andics was additionally supported by a Hungarian Academy of Sciences Bolyai Scholarship and by a Hungarian Scientific Research Fund grant (OTKA PD116181); Anna Kis was supported by Nestlé Purina and the BIAL Foundation (grant no 169/16). We thank Árpád Dobolyi and Kálmán Czeibert for their comments on this manuscript. 19 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 References Thompkins, A.M. et al. (2016) Functional Magnetic Resonance Imaging of the Domestic Dog: Research, Methodology, and Conceptual Issues. Comp. Cogn. Behav. Rev. 11, 63–82 Berns, G.S. and Cook, P.F. (2016) Why Did the Dog Walk Into the MRI? Curr. Dir. Psychol. Sci. 25, 363–369 Seed, A.M. et al. (2006) Investigating Physical Cognition in Rooks, Corvus frugilegus. Curr. Biol. 16, 697–701 Prior, H. et al. (2008) Mirror-induced behavior in the magpie (Pica pica): evidence of self-recognition. PLoS Biol. 6, e202 Güntürkün, O. and Bugnyar, T. (2016) Cognition without Cortex. Trends Cogn. Sci. 20, 291–303 Vandamme, T.F. (2014) Use of rodents as models of human diseases. J Pharm Bioallied Sci 6, 2–9 Phillips, K.A. et al. Why primate models matter. , American Journal of Primatology, 76. (2014) , 801–827 Jennings, C. et al. (2016) Opportunities and challenges in modeling human brain disorders in transgenic primates. Nat. Neurosci. 19, 1123–1130 Gaspo, R. et al. (1997) Functional mechanisms underlying tachycardia-induced sustained atrial fibrillation in a chronic dog model. Circulation 96, 4027–4035 Papoti, D. et al. (2013) An embedded four-channel receive-only RF coil array for fMRI experiments of the somatosensory pathway in conscious awake marmosets. NMR Biomed. 26, 1395–1402 Miklósi, Á. and Topál, J. (2013) What does it take to become “best friends”? Evolutionary changes in canine social competence. Trends Cogn. Sci. 17, 287–294 Miklósi, Á. (2014) Dog Behaviour Evolution and Cognition, Oxford University Press. Gygax, L. et al. (2015) Dog behavior but not frontal brain reaction changes in repeated positive interactions with a human: A non-invasive pilot study using functional nearinfrared spectroscopy (fNIRS). Behav. Brain Res. 281, 172–176 Siegle, G.J. (2011) Beyond depression commentary: Wherefore art thou, depression clinic of tomorrow? Clin. Psychol. Sci. Pract. 18, 305–310 Cosgrove, K. et al. (2007) Evolving knowledge of sex differences in brain structure, function, and chemistry. Biol. Psychiatry 62, 847–855 Caviness, V.S. et al. (1996) The human brain age 7–11 years: a volumetric analysis based on magnetic resonance images. Cereb. cortex 6, 726–736 McGreevy, P. et al. (2003) A Strong Correlation Exists between the Distribution of Retinal Ganglion Cells and Nose Length in the Dog. Brain. Behav. Evol. 63, 13–22 Roberts, T. et al. (2010) Human induced rotation and reorganization of the brain of domestic dogs. PLoS One 5, e11946 Regodon, S. et al. (1993) Craniofacial angle in dolicho-, meso- an brachycephalic dogs: radiological determination and application. Ann. Anat. 175, 361–363 Schoenebeck, J.J. and Ostrander, E.A. (2013) The genetics of canine skull shape variation. Genetics 193, 317–325 Dickie, A.M. and Sullivan, M. (2001) The effect of obliquity on the radiographic appearance of the temporomandibular joint in dogs. Vet. Radiol. Ultrasound 42, 205– 217 Schwarz, T. et al. (2000) Radiographic anatomy of the cribiform plate (Lamina cribosa). Vet. Radiol. Ultrasound 41, 220–225 Gacsi, M. et al. (2009) Effects of selection for cooperation and attention in dogs. Behav. Brain Funct. 5, 31 20 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 Berns, G.S. et al. (2012) Functional MRI in awake unrestrained dogs. PLoS One 7, Berns, G.S. et al. (2013) Replicability and heterogeneity of awake unrestrained canine fMRI responses. PLoS One 8, e81698 Berns, G.S. et al. (2015) Scent of the familiar: An fMRI study of canine brain responses to familiar and unfamiliar human and dog odors. Behav. Processes 110, 37– 46 Cook, P.F. et al. (2014) One pair of hands is not like another: caudate BOLD response in dogs depends on signal source and canine temperament. PeerJ 2, e596 Dilks, D.D. et al. (2015) Awake fMRI reveals a specialized region in dog temporal cortex for face processing. PeerJ 3, e1115 Kujala, M. V et al. (2013) Reactivity of dogs’ brain oscillations to visual stimuli measured with non-invasive electroencephalography. PLoS One 8, e61818 Törnqvist, H. et al. (2013) Visual event-related potentials of dogs: A non-invasive electroencephalography study. Anim. Cogn. 16, 973–982 Kis, A. et al. (2017) The interrelated effect of sleep and learning in dogs (Canis familiaris); an EEG and behavioural study. Sci. Rep. 7, 41873 Kis, A. et al. Development of a non-invasive polysomnography technique for dogs (Canis familiaris). Physiol. Behav. 130, 149–156 Cook, P. (2016) Awake canine fMRI predicts dogs’ preference for praise versus food. Soc. Cogn. Affect. Neurosci. 11, 1853–1862 Andics, A. et al. (2014) Voice-sensitive regions in the dog and human brain are revealed by comparative fMRI. Curr. Biol. 24, 574–578 Andics, A. et al. (2016) Neural mechanisms for lexical processing in dogs. Science (80-. ). 353, 1030–1032 Cuaya, L. V. et al. (2016) Our faces in the dog’s brain: Functional imaging reveals temporal cortex activation during perception of human faces. PLoS One 11, 1–13 Jia, H. et al. (2014) Functional MRI of the olfactory system in conscious dogs. PLoS One 9, e86362 Jia, H. et al. (2015) Enhancement of odor-induced activity in the canine brain by zinc nanoparticles: A functional MRI study in fully unrestrained conscious dogs. Chem. Senses Kyathanahally, S.P. et al. (2015) Anterior-posterior dissociation of the default mode network in dogs. Brain Struct. Funct. 220, 1063–1076 Cook, P.F. et al. (2016) Neurobehavioral evidence for individual differences in canine cognitive control: an awake fMRI study. Anim. Cogn. DOI: 10.1007/s10071-0160983-4 Jia, H. et al. (2015) Enhancement of odor-induced activity in the canine brain by zinc nanoparticles: A functional MRI study in fully unrestrained conscious dogs. Chem. Senses 0, 1–15 Berns, G.S. et al. (2013) Replicability and heterogeneity of awake unrestrained canine fMRI responses. PLoS One 8, e81698 Andics, A. et al. (2016) Neural mechanisms for lexical processing in dogs. Science (80-. ). 353, 1030–1032 Power, J.D. et al. (2012) Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion. Neuroimage 59, 2142–2154 de Bie, H.M. et al. (2010) Preparing children with a mock scanner training protocol results in high quality structural and functional MRI scans. Eur. J. Pediatr. 169, 1079– 1085 Bunford, N. et al. (2017) Threat distractor and perceptual load modulate test-retest reliability of anterior cingulate cortex response. Prog. Neuro-Psychopharmacology 21 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 Biol. Psychiatry 77, 120–127 Morris, S.E. and Cuthbert, B.N. (2012) Research Domain Criteria: cognitive systems, neural circuits, and dimensions of behavior. Dialogues Clin. Neurosci. 14, 29–37 Evans, A.C. et al. (1993) , 3D statistical neuroanatomical models from 305 MRI volumes. , in Proc. IEEE-Nuclear Science Symposium and Medical Imaging Conference, pp. 1813–1817 Muzik, O. et al. (2000) Statistical parametric mapping: assessment of application in children. Neuroimage 12, 538–549 Sanchez, C.E. et al. (2011) Neurodevelopmental MRI brain templates for children from 2 weeks to 4 years of age. Dev Psychobiol 54, 77–91 Sanchez, C.E. et al. (2012) Age-specific MRI templates for pediatric neuroimaging. Dev. Neuropsychol. 37, 379–99 Grabner, G. et al. (2014) A study-specific fMRI normalization approach that operates directly on high resolution functional EPI data at 7 Tesla. Neuroimage 100, 710–714 McGreevy, P.D. et al. (2013) Dog behavior co-varies with height, bodyweight and skull shape. PLoS One 8, Howell, T.J. et al. (2011) Development of a minimally-invasive protocol for recording mismatch negativity (MMN) in the dog (Canis familiaris) using electroencephalography (EEG). J. Neurosci. Methods 201, 377–380 Jasper, H.H. (1958) Report of the committee on methods of clinical examination in electroencephalography. Electroencephalogr. Clin. Neurophysiol. 10, 370–375 Wu, M. et al. (2016) Age-related changes in amygdala-frontal connectivity during emotional face processing from childhood into young adulthood. Hum. Brain Mapp. 37, 1684–1695 Howell, T.J. et al. (2012) Auditory stimulus discrimination recorded in dogs, as indicated by mismatch negativity (MMN). Behav. Processes 89, 8–13 Arden, R. et al. (2016) A Review of Cognitive Abilities in Dogs, 1911 Through 2016: More Individual Differences, Please! Curr. Trends Psychol. Sci. 25, 307–312 Bunford, N. et al. Neurofunctional correlates of behavioral inhibition system sensitivity during attentional control are modulated by perceptual load. Wallis, J.D. (2012) Cross-species studies of orbitofrontal cortex and value-based decision-making. Nat. Neurosci. 15, 13–19 Bunford, N. et al. (2016) Neural Reactivity to Angry Faces Predicts Treatment Response in Pediatric Anxiety. J. Abnorm. Child Psychol. DOI: 10.1007/s10802-0160168-2 Luck, S.J. (2014) An Introduction to the Event-related Potential Technique, (2nd edn) MIT Press. Loewy, D.H. et al. (1996) The mismatch negativity to frequency deviant stimuli during natural sleep. Electroencephalogr. Clin. Neurophysiol. 98, 493–501 Näätänen, R. et al. (2007) The mismatch negativity (MMN) in basic research of central auditory processing: A review. Clin. Neurophysiol. 118, 2544–2590 Peterson, M.E. and Kutzler, M. Small animal pediatrics: the first 12 months of life. , Small animal pediatrics: the first 12 months of life. (2010) Fleming, S. et al. (2011) Normal ranges of heart rate and respiratory rate in children from birth to 18 years of age: A systematic review of observational studies. Lancet 377, 1011–1018 Barrett, K. et al. (2016) , Ganong’s Review of Medical Physiology. , New York, NY: McGraw-Hill. Klein, B.G. (2013) Cunningham’s textbook of veterinary physiology, Elsevier Health Sciences. 22 69 70 71 72 73 74 75 76 77 Spodick, D.H. (1993) Survey of selected cardiologists for an operational definition of normal sinus heart rate. Am. J. Cardiol. 72, 487–488 Murphy, K. et al. (2013) Resting-state fMRI confounds and cleanup. Neuroimage 80, 349–359 Berns, G.S. et al. (2013) Replicability and heterogeneity of awake unrestrained canine fMRI responses. PLoS One 8, Liu, Y. et al. (2012) Neural Substrate of the Late Positive Potential in Emotional Processing. J. Neurosci. 32, 14563–14572 Falkenstein, M. et al. (1999) ERP components in Go/Nogo tasks and their relation to inhibition. Acta Psychol. (Amst). 101, 267–291 Nigg, J.T. (2010) Attention-deficit/hyperactivity disorder endophenotypes, structure, and etiological pathways. Curr. Dir. Psychol. Sci. 19, 24–29 Wheaton, M.G. et al. (2014) Perceptual load modulates anterior cingulate cortex response to threat distractors in generalized social anxiety disorder. Biol. Psychol. 101, 13–17 Bunford, N. et al. (2016) Correspondence between Heart Rate Variability and Emotion Dysregulation in Children, Including Children with ADHD. J. Abnorm. Child Psychol. DOI: 10.1007/s10802-016-0257-2 Bunford, N. et al. (2016) Attenuated neural reactivity to happy faces is associated with rule breaking and social problems in anxious youth. Eur. Child Adolesc. Psychiatry DOI: 10.1007/s00787-016-0883-9