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