This is the accepted version of the journal article: Sol, Daniel; Olkowicz, Seweryn; Sayol, Ferran; [et al.]. «Neuron numbers link innovativeness with both absolute and relative brain size». Nature Ecology & Evolution, Vol. 6 (July 2022), p. 1381-1389. DOI 10.1038/s41559-022-01815-x This version is available at https://ddd.uab.cat/record/288044 under the terms of the license The final version of this study can be found here: https://doi.org/10.1038/s41559-022-01815-x 1 2 Reference: Sol, D., Olkowicz, S., Sayol, F., Kocourek, M., Zhang, Y., Marhounová, L., Osadnik, C., 3 Corssmit, E., Garcia-Porta, J., Martin, T. E., Lefebvre, L. & Němec, P. (2022) Neuron numbers link innovativeness with both absolute and relative brain size in birds. Nat Ecol 4 Evol 6, 1381–1389. 5 6 Neuron numbers link innovativeness with both 7 absolute and relative brain size 8 9 Authors: Daniel Sol1†, Seweryn Olkowicz2, Ferran Sayol3, Martin Kocourek2, 10 Yicheng Zhang2, Lucie Marhounová2, Christin Osadnik4, Eva Corssmit1, Joan 11 Garcia-Porta5, Thomas E. Martin6, Louis Lefebvre7†, Pavel Němec2† 12 13 14 Affiliations: 1 15 CSIC-CREAF (Centre for Ecological Research and Applied Forestries), 16 Cerdanyola del Vallès, Catalonia E-08193, Spain 2 17 Department of Zoology, Faculty of Science, Charles University, Prague, Czech 18 Republic 3 19 Centre for Biodiversity and Environment Research, Department of Genetics, 20 Evolution and Environment, University College London, London, UK 4 21 Department of General Zoology, University of Duisburg-Essen, Essen, Germany 5 22 Department of Biology, Washington University in St. Louis, MO, 63130, USA 6 23 Montana Cooperative Wildlife Research Unit, University of Montana, Missoula, MT 24 USA 7 25 Department of Biology, McGill University, Montréal, Québec, Canada 26 27 † 28 Corresponding authors Emails: d.sol@creaf.uab.cat; louis.lefebvre@mcgill.ca; 29 pgnemec@natur.cuni.cz. 30 31 Short Title: Brains, neurons and cognition 1 32 Abstract 33 A longstanding issue in biology is whether the intelligence of animals can be predicted 34 by absolute or relative brain size. However, progress has been hampered by an 35 insufficient understanding of how neuron numbers shape internal brain organization 36 and cognitive performance. Based on estimations of neuron numbers for 111 bird 37 species, we show here that the number of neurons in the pallial telencephalon is 38 positively associated with a major expression of intelligence: innovation propensity. 39 The number of pallial neurons, in turn, is greater in brains that are larger in both 40 absolute and relative terms, and positively co-varies with longer post-hatching 41 development periods. Thus, our analyses show that neuron numbers link cognitive 42 performance to both absolute and relative brain size through developmental 43 adjustments. These findings help unify neuro-anatomical measures at multiple levels, 44 reconciling contradictory views over the biological significance of brain expansion. 45 The results also highlight the value of a life history perspective to advance our 46 understanding of the evolutionary bases of the connections between brain and 47 cognition. 48 49 50 51 52 53 54 55 56 57 58 2 59 Main 60 Encephalization—the evolutionary increase of the brain beyond that expected for a 61 given body size1—has long been thought to be a major factor in the evolution of 62 intelligence2,3. Comparisons across species have provided some support for this 63 theory, showing that encephalization is associated with several facets of intelligence 64 like innovativeness, learning and culture4–8. The theory has also stimulated an 65 extended research program on the ecological and evolutionary implications of brain 66 size and architecture9–11. Yet, the reasons why a disproportionately larger brain should 67 provide cognitive advantages remain unclear3,12,13. 68 The rationale of the encephalization theory, as originally envisioned by 69 Jerison14, is that the “extra tissue” that makes the brain larger than expected for a given 70 body size (i.e., larger relative brain size) reflects extra neurons that are available for 71 cognitive tasks. However, the notion that cognitive performance depends on neuron 72 numbers and increases with encephalization is backed by insufficient evidence3,15,16. 73 Moreover, given that neuron numbers increase with absolute brain size3,17, should we 74 not expect that cognitive differences across species will be better predicted by 75 absolute rather than relative brain size? There is indeed evidence that absolute brain 76 size sometimes predicts cognitive performance across species better than relative 77 measures of the brain18–20. Such results are not surprising given that increases in 78 relative brain size can be reached by both brain enlargement and body size reduction, 79 and therefore may not always be associated with an increase in brain information- 80 processing capacity21,22. The debate regarding the biological significance of absolute 81 and relative brain sizes has been further complicated by the finding that not all brains 82 are made in the same way; rather, brains may show different neuron densities and 83 distributions among brain areas across species12,15,23–25. Thus, the intuitively appealing 84 notion that larger brains translate into greater intelligence remains contentious3,12,13. 85 To address this longstanding controversy, we provide theoretical and 86 empirical grounds for the hypothesis that increased intelligence—operationally 87 defined here as the ability to solve problems through mental or behavioural 88 flexibility3— requires brains that are large in both absolute and relative terms (Fig. 89 1). This possibility has probably gone unrecognized because previous studies have 90 used an 'either-or' approach and pitted absolute against relative brain size 91 measures18,20. Yet if enhanced cognition requires more neurons in sensory, associative 3 92 and premotor areas of the telencephalon —the pallial areas in birds and the neocortex 93 in mammals3,24 — and these areas represent a large fraction of the telencephalon and 94 the whole brain26, then the accumulation of a disproportionately large number of 95 pallial neurons should produce brains that are larger both in absolute terms and 96 relative to body size27,28. Such co-variation between absolute and relative brain size 97 should be more accentuated if the selective advantages of accumulating greater 98 numbers of pallial neurons is higher for larger species than for small ones. This is to 99 be expected because body size is a major correlate of longevity29, and a long life 100 increases the fitness value of gathering information and learning30–33 while reducing 101 the costs of delaying reproduction34,35. One mechanism that may allow a greater 102 accumulation of pallial neurons is, according to some evo-devo models, an extension 103 of development periods, particularly the later stages in altricial offspring that are born 104 underdeveloped36,37. Thus, selection on cognition might link intelligence with larger 105 absolute and relative brain size through developmental adjustments (Fig. 1). 106 Testing the above tenets is challenging owing to the difficulties of accurately 107 estimating neuron numbers of different brain regions for many species15. The isotropic 108 fractionator—a new method of assessing neuron numbers developed by Herculano- 109 Houzel and collaborators38— now makes it possible. Our study is based on a 110 substantially updated dataset24,25,39 quantifying neuron numbers in the whole brain 111 and three brain areas (the pallium, the cerebellum and the brainstem) for 111 species 112 of 24 avian families, representing both basal and crown avian lineages (Fig. 1e; 113 Supplementary Fig. 1) and encompassing a large fraction of the morphospace 114 occupied by avian brains (Supplementary Fig. 2). To test associations with cognition, 115 we focus on a major component of intelligence, innovativeness40, by quantifying its 116 product —innovation frequency4–6,41. Our innovation data were extracted from a 117 database including >4400 published reports of bird species using novel foods or new 118 feeding techniques in the wild42. On this basis, we first ask whether innovation 119 propensity increases with the number of neurons in the pallium (and potentially also 120 in the cerebellum, which is thought to co-evolve and function in tandem with the 121 pallium43,44), but not with those in areas less directly involved in cognition like the 122 brainstem. Next, we investigate whether the proliferation of neurons in the pallium 123 makes the brain increase disproportionally with body size, linking innovativeness 124 with both absolute and relative brain size. Finally, we test whether the accumulation 125 of neurons in the pallium is associated with an extension of later stages of 4 126 development. We test these predictions by combining random forests, a type of 127 machine-learning algorithm that allows us to accommodate complex non-linear 128 interactions among predictors with minimal assumptions45, with Bayesian mixed 129 models that explicitly account for phylogenetic relatedness among species46. Because 130 nocturnal species are difficult to observe, and hence are not present in the innovation 131 data set, we exclude owls from all the analyses that follow; results with the entire 132 dataset are shown in the Supplementary Information. 133 134 Results and discussion 135 Cognitive performance has long been thought to depend on the number of 136 neurons in the brain47,48, but this hypothesis is currently backed by surprisingly little 137 empirical evidence15. A comparison of apes, corvids and pigeons in five cognitive 138 domains concluded that neuron number is a poor predictor of absolute cognitive 139 performance, but it may predict learning speed and the ability to plastically adjust 140 rules to novel situations47. A broader comparative analysis across primates and birds 141 revealed that performance in a cognitive task (the detour test) does tend to increase 142 with the total number of cortical/pallial neurons15, yet this study did not rule out the 143 possibility that the association was driven by phylogenetic relatedness. 144 Using Bayesian phylogenetic mixed models, we found that the number of 145 neurons in the entire brain is positively associated with behavioural innovation 146 propensity (Fig. 2a), particularly technical innovations that are assumed to require 147 more advanced cognition5 (Supplementary Table 1). The pattern holds when body 148 mass is included as a covariate in the model (Fig. 2b; Supplementary Table 2), 149 suggesting that innovation propensity is higher in birds with a disproportionately 150 larger number of neurons than expected on the basis of body size. While we find that 151 the brain of innovative species contains more neurons than the brain of less innovative 152 species, there is no parallel increase in neuron density; rather, innovation propensity 153 decreases with neuron density (Fig. 2c). Because avian neuron densities tend to 154 decrease with brain size and the total number of neurons, these results support the 155 notion that cognitive performance is primarily limited by the absolute and relative 156 number of neurons rather than by neuron densities. 157 Additional analyses revealed that the number of neurons in the pallium and, to 158 a lesser extent, the cerebellum are better predictors of innovation propensity than 159 neurons in the brainstem, a brain area less directly involved in cognition (Fig. 2; 5 160 Supplementary Tables 1 and 2). Although pallial areas are thought to co-evolve and 161 function in tandem with the cerebellum43,44, it remains to be determined whether the 162 avian cerebellum subserves motor skills only (as its association with technical 163 innovations but not resource innovations suggests; Supplementary Tables 2) or is 164 also directly involved in cognitive functions like the mammalian cerebellum43. 165 Nonetheless, our findings align with growing evidence that cognitive processes 166 associated with intelligence are controlled by widely-distributed networks integrating 167 several brain areas49. 168 Corvids and parrots are regarded as the most innovative birds, a conclusion 169 that is backed by ample experimental evidence47,48,50,51. These taxa also share both the 170 highest inferred rates of brain-body size evolution among Neoaves and the steepest 171 allometric slopes among all birds52. This contrasts with less innovative taxa like early- 172 diverging birds (Palaeognathae, basal Neognathae), Anseriformes (waterfowl), and 173 predatory core landbirds (hawks & eagles, falcons and owls), whose allometric 174 exponents have diverged little from the ancestral avian grade and hence represent low- 175 slope grades. To assess whether the proliferation of neurons in the pallium can explain 176 deviations from the “ancestral” allometric scaling relationship, we estimated the 177 allometric exponents of the neuron numbers for clades with the highest slope and low- 178 slope grades (sensu 52); we then compared these with the allometric exponents for the 179 cerebellum and brainstem. We find that while the allometric exponents for the 180 cerebellum and brainstem were similar between the two slope grade groups, clades 181 that share a high slope tended to accumulate disproportionately more neurons in the 182 pallium as they become larger (Fig. 3, Supplementary Figs. 3-5). Thus, as expected, 183 the accumulation of pallial neurons makes the brain increase in both absolute and 184 relative terms (Fig. 4, Supplementary Figs. 6 and 7). 185 While the number of neurons in the whole brain and the pallium increases in 186 a similar way with both absolute and relative brain size (Fig. 4), the number of neurons 187 in the cerebellum is more strongly related to absolute brain size alone; those in the 188 brainstem do not follow any clear pattern. These conclusions are consistent regardless 189 of the method used to estimate relative brain size (Supplementary Fig. 7); they also 190 hold when we include owls (Supplementary Figs. 6 and 7), whose large forebrain 191 results in part from expanding the visual Wulst for sensory rather than associative 192 purposes. Although the evolutionary repatterning of the brain-body relationship 193 cannot be circumscribed to selection on brain size alone, our results support the notion 6 194 that cognition can form a major driver of adaptive shifts to higher grades deviations 195 from the “ancestral” allometric scaling. 196 Presumably because brain cellular scaling rules can be clade-specific15,23–25,53 197 and because avian neuron densities decrease with brain mass24,25, the relationship 198 between neuron numbers and brain size is complex. The relationship tends to be 199 roughly linear for relative brain size, especially when we exclude owls, but only for 200 the entire brain and the pallium (Fig. 4b). In contrast, neuron numbers tend to 201 asymptote at larger absolute brain sizes in all cases (Fig. 4c). This latter finding agrees 202 with the notion that animals that have large brains merely because they have very big 203 bodies are not necessarily the most intelligent, as it is the case for Ratites and large 204 Galliformes. 205 Several developmental mechanisms, including differences in early morphogen 206 patterning and expansion of stem cell pool, diversification of neural progenitors, 207 variation of cell-cycle rates and protracted neurogenesis are responsible for expansion 208 of the telencephalon in amniotes37,54–56. We asked whether these mechanisms were 209 reflected in the duration of embryonic and post-hatching development periods. We 210 found that longer development time leads to a greater accumulation of number of 211 neurons in the pallium of clades with high-slope grades than in those showing low- 212 slope grades (Fig. 5; Supplementary Fig. 8). This accumulation of neurons is 213 associated with an extension of the post-natal (fledging) development relative to the 214 embryonic period. Importantly, scaling of pallial neuron counts with development 215 strongly resembles that found for the total number of brain neurons, reinforcing the 216 notion that the number of pallial neurons largely accounts for the much larger number 217 of neurons in the brains of altricial species. 218 Growing evidence suggests that different mechanisms underlie telencephalon 219 growth and maturation in precocial and altricial birds, leading to relatively larger 220 relative brains in the latter. Precocial birds like ducks and grouse enlarge their 221 telencephalon early in development (before the onset of neurogenesis) presumably by 222 an increase in the number of telencephalic progenitors57. In contrast, expansion of the 223 telencephalon in altricial birds like songbirds and parrots is associated with protracted 224 neurogenesis and delayed neuronal maturation58. Our analyses are consistent with 225 these patterns (Supplementary Fig. 9), showing that longer development time leads 226 to greater accumulation of neurons in the pallium in altricial than in precocial species. 227 Indeed, all species from our data set belonging to the highest slope grade category 7 228 show high degree of altriciality (i.e. they are classified as super-altricial). Altogether, 229 our results are consistent with the view that increases in absolute and relative brain 230 size are made possible through the evolutionary transition to altriciality and changes 231 in neurogenesis schedules related to a disproportionate lengthening of the later stages 232 of development. While this pattern is predicted by some models of brain development, 233 like the 'late is large' rule37, additional research is needed to elucidate the exact 234 mechanisms. 235 Our analyses unify neuro-anatomical measures at multiple levels. First, we 236 provide firm support for the intuitively appealing notion that cognitive performance 237 is limited by the number of neurons in the pallium and, to a lesser extent, the 238 cerebellum. Thus, our results support the hypothesis14 that intelligence reflects a 239 disproportionate allocation of neurons to cognitive tasks, but also aligns with 240 suggestions that cognitive performance ultimately depends on the total number of 241 neurons and the way neurons connect different brain areas3,15,23. Second, we show that 242 an increase in the number of neurons in the areas most closely involved in cognitive 243 performance, the pallium, increases brain size in both absolute and relative terms. 244 Although the number of neurons in the cerebellum scaled primarily with absolute 245 brain size, the effect of total neuron numbers on relative brain size persisted because 246 in birds larger brains contain increasing proportions of neurons in the pallium and 247 decreasing proportions in the cerebellum and other brain regions24. Third, we provide 248 an adaptive explanation for some of the patterns of brain-body co-variation in deep 249 time detected by Ksepka et al.52: Clades that have a higher brain-body slope than 250 others tend to be the ones that are most innovative. A higher brain-body slope means 251 that as body size gets bigger, the brain increases disproportionately more in size than 252 it does in non-innovative clades; this increase in both absolute and relative brain size 253 is, according to our analyses, mostly due to an increase in pallial neurons. Finally, we 254 provide a developmental rationale for the observed patterns, suggesting that the 255 elongation of the fledging period in altricial species links neuron numbers with 256 absolute and relative brain size. The failure to find a similar pattern in precocial 257 species supports the notion that not all brains are made in a same way25,53, highlighting 258 the key role of life history in brain evolution (Supplementary Fig. 10). 259 The reason why the dual role of absolute and relative brain size in cognition 260 has been under-appreciated in the past probably reflects the common practice of 261 removing the allometric effects of body size in comparative analyses of brain size. As 8 262 suggested by Jerison2, this is probably legitimate when comparing brains of species 263 with striking differences in body size, like an ostrich and a hummingbird. Yet by 264 treating body size as a statistical nuisance, we appear to be missing important 265 information. A larger body is often associated with greater longevity59 and can reduce 266 juvenile mortality risk35, which should increase the value of learning and reduce the 267 costs of a long development time30,31,60. Alternatively, the same environmental 268 pressures that favour a slow pace of life could generate correlated selection on both 269 cognition and body size31. Whether a large body facilitates selection for cognition or 270 co-varies with cognition due to either correlated selection or shared developmental 271 processes, the consequence for functional architecture of the brain is to link neuron 272 numbers and cognitive performance to both absolute and relative brain size. 273 274 Methods 275 Neuron numbers estimation. Our study is based on an updated database quantifying neuron 276 numbers in the whole brain and three brain areas – the pallium (comprising the hyperpallium, 277 mesopallium, nidopallium, acropallium and hippocampus), the cerebellum and the brainstem 278 (comprising the medula oblongata and midbrain tegmentum) for bird species. Information for 279 65 avian species were extracted from the literature24,25. Numbers of brain, pallial and 280 cerebellar neurons for an additional 81 individuals representing 46 species and number of 281 brainstem neurons for an additional 172 individuals representing 83 avian species were newly 282 estimated using the isotropic fractionator, following experimental procedures described in 283 Olkowicz et al.24. Briefly, animals were killed by an overdose of halothane, weighed and 284 immediately perfused transcardially with warmed phosphate-buffered saline containing 0.1% 285 heparin followed by cold phosphate-buffered 4% paraformaldehyde solution. The brains were 286 immediately removed, weighted, postfixed for an additional 7-21 days and then dissected into 287 the examined brain divisions. The cerebral hemispheres (including the olfactory bulbs) were 288 detached from the diencephalon by a straight cut separating the subpallium from the thalamus. 289 The cerebellum was cut off at the surface of the brainstem. The tectum (optic lobe) was 290 bilaterally excised from the surface of the brainstem. The excised parts included most of the 291 tectal gray, optic tectum and torus semicircularis. The remaining structures were dissected 292 into diencephalon (rostral part) and brainstem (caudal part comprising the medula oblongata 293 and midbrain tegmentum) along the plane connecting the posterior commissure dorsally and 294 hypothalamus-mesencephalon boundary ventrally. The latter is visible macroscopically as a 295 groove between the convex ventral part of the midbrain and the hypothalamus, caudally to 296 the infundibulum and mammillary bodies. In one individual per species, one hemisphere was 9 297 dissected into the pallium and the subpallium. These hemispheres were embedded in agarose 298 and sectioned on a vibratome at 300–500 µm (depending on size of a hemisphere) in the 299 coronal plane. Under oblique transmitted light at the stereomicroscope and with the use of a 300 microsurgical knife (Stab Knife Straight, 5.5 mm, ref 7516, Surgical Specialities Corporation, 301 Reading, PA, USA) we manually dissected the pallium from subpallium on each section by 302 cutting along the pallial-subpallial lamina, as defined by Reiner et al.62. 303 The dissected structures were dried with a paper towel, weighed to the nearest 0.1 304 milligram, incubated in 30% sucrose solution until they sank, then transferred into antifreeze 305 (30% glycerol, 30% ethylene glycol, 40% phosphate buffer) and frozen for further processing. 306 The examined brain parts were homogenized in 40 mM sodium citrate with 1% Triton X-100 307 using Tenbroeck tissue grinders (Wheaton, Millville, NY, USA) to obtain a suspension of free 308 cell nuclei. The fluorescent DNA marker 4′,6-diamidino-2-phenylindole (DAPI) was added 309 (0.5 mg / l) to stain the nuclei. Afterwards the homogenate was adjusted to defined volume 310 and the mixture was kept homogenous by agitation. The total number of cells was estimated 311 by counting at least five aliquots of 10 µl using a Neubauer improved counting chamber 312 (BDH, Dagenham, Essex, UK) with an Olympus BX51 microscope equipped with 313 epifluorescence and appropriate filter settings; additional aliquots were counted if needed to 314 reach the coefficient of variation among counts ≤0.10. The proportion of neurons was 315 determined by immunocytochemical detection of the neuronal nuclear marker NeuN63. This 316 neuron-specific protein was detected by an anti-NeuN mouse monoclonal antibody (clone 317 A60, Sigma-Aldrich; dilution 1:800), which was characterized by Western blotting with chick 318 brain samples and shown to react with a protein of the same molecular weight as in mammals, 319 indicating that it does not cross-react with other proteins in birds64. The binding sites of the 320 primary antibody were revealed by Alexa Fluor 546-conjugated goat anti-mouse IgG antibody 321 (Life Technologies, Carlsbad, CA, USA; dilution 1:400). An electronic hematologic counter 322 (Alchem Grupa, Torun, Poland) was used to count the proportion of double-labelled nuclei in 323 the Neubauer chamber. At least 500 nuclei were examined for each sample. The final data set 324 included information on neuron numbers for 240 specimens belonging to 111 species. For 325 Caloenas nicobarica and Eudromia Formosa, information on pallial neurons was missing and 326 had to be imputed to avoid comparing results with different sample sizes. We estimated these 327 missing data by combining phylogenetic imputation with multivariate data (brain and body 328 size, and number of neurons in the entire brain, the pallium, the cerebellum and the brainstem), 329 as implemented in the R-package phytools65. We note that results hold whether or not these 330 two species were included in the analyses. 331 10 332 Innovation data. Our innovation data were taken from Lefebvre42, compiled by 333 systematically searching for reports of new behaviours in the short notes of 204 ornithology 334 journals published between 1960 and 2020. The criterion to accept an innovation was that the 335 report described the behaviour with key words such as ‘novel’, ‘not noted before’ and/or 336 ‘unusual’. Each innovation was classified as a resource innovation —if it involved a novel 337 food item—, or a technical innovation —if the searching and handling techniques were 338 themselves novel regardless of whether the food type was novel or not5. Nocturnal clades 339 were excluded due to the difficulty of being observed. The frequency with which a species 340 was observed innovating in the wild was used to characterize the propensity of the species to 341 innovate. Innovation propensity depends not only on innovative ability, however, but also on 342 the probability that new behaviours are observed and reported. Thus, a species may have a 343 low number of innovations not because it cannot innovate but because it is rare or secretive, 344 and hence difficult to observe and study. We tackled this issue by considering research effort 345 in the analyses5,11,31, using data on number of papers published per species66. The probability 346 of reporting an innovation may also increase with geographic range, urbanisation and island 347 living, and it can decrease with migratory behaviour5,11,31. Therefore, we also included these 348 variables as covariates in the models (see below). Data were drawn from previously 349 assembled datasets. Geographic range (number of 1 degree x 1 degree grid cells overlapping 350 breeding/resident range), mobility (resident, nomadic, migrant and altitudinal migrant) and 351 insularity (proportion of breeding/resident range intersecting with islands of landmass below 352 2,000 sq km) were extracted from67, while the occurrence in urban environments was taken 353 from11. 354 355 Life history data. We extracted published information on the duration of incubation 356 (embryonic stage) and fledging periods (post-natal growth) from previously compiled 357 datasets11,31, updated with information from the online edition of the Handbook of birds of 358 the world (https://birdsoftheworld.org). Information was available for 108 species for 359 incubation duration and 102 species for fledging duration. Post-natal growth fraction was 360 estimated as [fledging/(incubation + fledging)]0.5, following68. To assign species to different 361 developmental modes, we used the classification recently proposed by Bothelo69, which 362 divides species in super-precocial, precocial, semi-precocial, semi-altricial, altricial and 363 super-altricial. However, for the questions addressed in the study, and because some 364 categories were insufficiently represented in our data set, we pooled together super-precocial, 365 precocial and semi-precocial species in a general precocial category and semi-altricial, 366 altricial and super-altricial species in an altricial category. 367 11 368 Modelling neurons, brains and innovations. We used Bayesian Generalised Linear Mixed 369 models (BGLMM) based on Markov Chain Monte Carlo (MCMC) approximations to model 370 variation in neuron numbers, brain size and innovation propensity, as implemented in the R- 371 packages MCMCglmm46 and BRMS70. To ensure that neuron numbers and brain measures 372 were species characters, we first used intra-specific data to assess within-species consistency 373 by means of Gaussian BGLMM, including sex as fixed effect and species as random effect 374 (Supplementary Table 3). Consistency was estimated as the Intraclass Correlation 375 Coefficient (ICC), calculated by dividing the variation among species by the total variation 376 (i.e. variation among species plus variation within species, the latter including natural 377 variation and measurement error). The consistency attributed to shared ancestors was 378 estimated in a similar model, but incorporating a variance-covariance matrix of phylogenetic 379 distances as a random effect. This allowed us estimating varying intercepts among species 380 adjusted by phylogenetic dependency. Phylogenetic heritability was estimated as the fraction 381 of total variation accounted for the phylogenetic distance between species (Supplementary 382 Table 3). 383 To test whether neuron numbers affect innovation frequency, we then averaged the 384 values for each species and used phylogenetic BGLMMs with the response variable 385 (innovation frequency) fitted as Gaussian. In these models, biogeographic realm71 was 386 included as a random effect together with the phylogenetic variance-covariance matrix to 387 allow the integration of global information originating from different regions31. To control 388 for potential confounding effects, research effort, geographic range, tolerance to urbanisation, 389 insularity and mobility were included as fixed effects. 390 Species-level phylogenetic BGLMMs were also used to model neuron numbers as a 391 function of body mass, development duration (incubation and fledging) and incubation 392 fraction. We generally used BRMS with Gaussian responses, switching to Weibull 393 distributions when divergent transitions affected model convergence. To assess whether the 394 relationship varied between low-slope and highest-slope grades, sensu Ksepka et al.52, we 395 included in the models an interaction with a variable coding for these two groups. Differences 396 between precocial and altricial species was investigated in a similar way. 397 The phylogenetic hypothesis was a summary trees based on 10,000 trees from one of 398 the backbones of the complete phylogeny of birds72 available at www.birdtree.org. We note 399 that using the alternative phylogenetic backbone yielded similar results. 400 For all models, the number of MCMC iterations and the burn-in interval were chosen 401 so as to ensure satisfactory convergence. The priors settings are described in the 402 Supplementary R code. The parameters reported for fixed and random effects are the posterior 403 mode and the 95% lower and upper credibility intervals (CI). We considered the fixed effects 12 404 statistically significant when 95% CIs did not include the zero. Conditional effects plots and 405 95% credible intervals were used to visualize the relationship between predictors and response 406 variables. 407 408 409 Describing neuron numbers as a function of relative and absolute brain size. We used 410 regression-based Random Forests (RF), a type of machine-learning algorithm, to describe 411 neuron numbers as a function of both absolute and relative brain size73. When modelling 412 quantitative response variables, RF uses linear regressions to recursively partition the data by 413 means of decision trees. Instead of selecting a best tree, however, the method does so by 414 taking a random sample of the training data and a random selection of variables at each step45. 415 For each tree in the RF, the fitted value of each terminal node is the mean of the response 416 variable values, which is averaged over all trees to estimate the fitted values of the RF. The 417 data not used to train the model, the out-of-bag (OOB) sample, provides a way to stabilise the 418 error without having to sacrifice training data to use for validation. In this way, RF allows to 419 efficiently model non-linear relationships and deal with complex interactions between 420 predictors while avoiding over-fitting, producing stable patterns that are more difficult to 421 change with new data and that are less sensitive to outliers. 422 We modelled neuron numbers (response variable) as a function of relative and 423 absolute brain size (predictors) with the R-package randomForest74. Following the protocol 424 suggested by Brieuc et al.45, we ran 500 trees twice and compared the stability of the results 425 (correlation >0.97 in all cases). Deviations between the fitted and observed values were used 426 to compute a “pseudo” R2. Bi-variate partial dependence (i.e., marginal effects) plots for the 427 last tree in the forest, once the model had converged, were used to visualize the co-variation 428 of neuron numbers with absolute and relative brain size while univariate plots were used to 429 visualize the influence of each predictor separately. 430 To be included together with absolute brain size in the RF, we estimated relative brain 431 size by means of the normalized scaled brain index61 (NSBI). This approach uses the equation 432 of allometric growth to adjust the brain size of species to that which they would have if all 433 had the same body size, making the values directly comparable: 𝛸" # 434 𝑁𝑆𝐵𝐼 = 𝑌! ( + 𝑋! 435 where b is the allometric exponent, Yi and Xi are, respectively, brain size and body size for the 436 individual i and X0 is the ancestral body size used to scale all species to the same size. We 437 estimated the allometric exponent b based on a log-log phylogenetic Gaussian BGLMMs of 438 absolute brain size against body mass. To this purpose, we used a previously assembled 439 dataset of brain mass (g) and volume (ml)75, updated with information on brain mass from 13 440 the specimens used to estimate neuron numbers (see above) and with new endocast measures 441 of specimens from museums in Europe and North America (n = 114 specimens). Volumes 442 were obtained by means of the endocast method and converted to mass by multiplying by the 443 density of brain tissue (i.e. 1.036g ml–1)75. We only used specimens of known body mass, 444 yielding information for 10,523 specimens belonging to 1,977 species. To scale all species to 445 the same size, we used the body mass of the presumed ancestor of current birds (2400 g), as 446 suggested by Torres et al.76. In our dataset, for example, the greater adjutant (Leptoptilos 447 dubius) has the largest brain (~34g) of the 1976 species from our dataset but this mainly 448 reflects the fact that it is a large bird (~7400g). Using the above normalization technique to 449 scale the brain, the NSBI of the greater adjutant (Leptoptilos dubius) —estimated around 450 6.29— brings the species down to the 300th position of the ranking. 451 The NSBI is equivalent to the residual approach used in previous studies and hence 452 presumes that species share the same brain:body allometric equation. However, there is 453 evidence that the allometric exponent can exhibit some differences across lineages52. 454 Consequently, we used a second NSBI (NSBIgrades) based on an allometric exponent b 455 estimated excluding clades that have been found to exhibit substantial grade shifts in 456 brain:body allometries (i.e. Anseriformes and Neoaves)52. This exponent represents the 457 scaling relationship of birds before some lineages experienced grade shifts (Fig. 1), and it is 458 remarkably close to that estimated with the entire dataset (Supplementary Fig. 11). Thus, in 459 the main text we present the results based on the NSBIgrades, which better fit to the proposed 460 theoretical framework (Fig. 1). 461 462 463 References 464 465 1. Shultz, S. & Dunbar, R. Encephalization is not a universal macroevolutionary 466 phenomenon in mammals but is associated with sociality. Proceedings of the National 467 Academy of Sciences of the United States of America 107, 21582–21586 (2010). 468 2. Jerison, H. J. Animal intelligence as encephalization. Philosophical Transactions of 469 the Royal Society of London. B, Biological Sciences 308, 21–35 (1985). 470 3. Roth, G. & Dicke, U. Evolution of the brain and intelligence. Trends in Cognitive 471 Sciences 9, 250–257 (2005). 472 4. Lefebvre, L., Whitle, P., Lascaris, E. & Finkelstein, A. Feeding innovations and 473 forebrain size in birds. Animal Behaviour 53, 549–560 (1997). 474 5. Overington, S. E., Morand-Ferron, J., Boogert, N. J. & Lefebvre, L. Technical 475 innovations drive the relationship between innovativeness and residual brain size in 476 birds. Animal Behaviour 78, 1001–1010 (2009). 477 6. Reader, S. M., Hager, Y. & Laland, K. N. The evolution of primate general and cultural 14 478 intelligence. Philosophical Transactions of the Royal Society B: Biological Sciences 479 366, 1017–1027 (2011). 480 7. Benson-Amram, S., Dantzer, B., Stricker, G., Swanson, E. M. & Holekamp, K. E. 481 Brain size predicts problem-solving ability in mammalian carnivores. Proceedings of 482 the National Academy of Sciences 113, 2532–2537 (2016). 483 8. Reader, S. M. & Laland, K. N. Social intelligence, innovation, and enhanced brain size 484 in primates. Proceedings of the National Academy of Sciences 99, 4436–4441 (2002). 485 9. Fristoe, T. S., Iwaniuk, A. N. & Botero, C. A. Big brains stabilize populations and 486 facilitate colonization of variable habitats in birds. Nature Ecology & Evolution 1, 487 1706–1715 (2017). 488 10. van Woerden, J. T., van Schaik, C. P. & Isler, K. Effects of seasonality on brain size 489 evolution: evidence from Strepsirrhine primates. The American Naturalist 176, 758– 490 767 (2010). 491 11. Ducatez, S., Sol, D., Sayol, F. & Lefebvre, L. Behavioural plasticity is associated with 492 reduced extinction risk in birds. Nature Ecology & Evolution 4, 788–793 (2020). 493 12. Herculano-Houzel, S. Brains matter, bodies maybe not: The case for examining 494 neuron numbers irrespective of body size. Annals of the New York Academy of 495 Sciences 1225, 191–199 (2011). 496 13. Logan, C. J. et al. Beyond brain size: Uncovering the neural correlates of behavioral 497 and cognitive specialization. Comparative Cognition & Behavior Reviews 13, 55–89 498 (2018). 499 14. Jerison, H. Evolution of the Brain and Intelligence. (Academic Press, 1973). 500 15. Herculano-Houzel, S. Numbers of neurons as biological correlates of cognitive 501 capability. Current Opinion in Behavioral Sciences 16, 1–7 (2017). 502 16. Van Schaik, C. P., Triki, Z., Bshary, R. & Heldstab, S. A. A Farewell to the 503 Encephalization Quotient: A New Brain Size Measure for Comparative Primate 504 Cognition. Brain, Behavior and Evolution 96, 1–12 (2021). 505 17. Striedter, G. F. Principles of brain evolution. (Sinauer Associates, Inc. Sunderland, 506 Mass, 2005). 507 18. MacLean, E. L. et al. The evolution of self-control. Proceedings of the National 508 Academy of Sciences 111, E2140–E2148 (2014). 509 19. Matějů, J. et al. Absolute, not relative brain size correlates with sociality in ground 510 squirrels. Proceedings of the Royal Society B: Biological Sciences 283, 20152725 511 (2016). 512 20. Deaner, R. O., Isler, K., Burkart, J. & Van Schaik, C. Overall brain size, and not 513 encephalization quotient, best predicts cognitive ability across non-human primates. 514 Brain, Behavior and Evolution 70, 115–124 (2007). 515 21. Smaers, J. B., Dechmann, D. K. N., Goswami, A., Soligo, C. & Safi, K. Comparative 516 analyses of evolutionary rates reveal different pathways to encephalization in bats, 517 carnivorans, and primates. Proceedings of the National Academy of Sciences 109, 518 18006–18011 (2012). 519 22. Smaers, J. B. et al. The evolution of mammalian brain size. Science Advances 7, 520 eabe2101 (2021). 521 23. Němec, P. & Osten, P. The evolution of brain structure captured in stereotyped cell 522 count and cell type distributions. Curr. Opin. Neurobiol 60, 176–183 (2020). 15 523 24. Olkowicz, S. et al. Birds have primate-like numbers of neurons in the forebrain. 524 Proceedings of the National Academy of Sciences USA 113, 7255–7260 (2016). 525 25. Kverková, K. et al. The evolution of brain neuron numbers in amniotes. Proceedings 526 of the National Academy of Sciences of the United States of America 119, 527 e2121624119 (2022). 528 26. Iwaniuk, A. N. & Hurd, P. L. The evolution of cerebrotypes in birds. Brain, Behavior 529 and Evolution 65, 215–230 (2005). 530 27. Timmermans, S., Lefebvre, L., Boire, D. & Basu, P. Relative size of the hyperstriatum 531 ventrale is the best predictor of feeding innovation rate in birds. Brain, Behavior and 532 Evolution 56, 196–203 (2000). 533 28. Sayol, F., Lefebvre, L. & Sol, D. Relative brain size and its relation with the 534 associative pallium in birds. Brain, Behavior and Evolution 87, 69–77 (2016). 535 29. Healy, K. et al. Ecology and mode-of-life explain lifespan variation in birds and 536 mammals. Proceedings of the Royal Society B: Biological Sciences 281, 20140298– 537 20140298 (2014). 538 30. Deaner, R. O., Barton, R. A. & van Schaik, C. P. Primate brains and life histories: 539 Renewing the connection. in Primate life histories and socioecology 233–265 540 (University of Chicago Press, 2003). 541 31. Sol, D., Sayol, F., Ducatez, S. & Lefebvre, L. The life-history basis of behavioural 542 innovations. Philosophical Transactions of the Royal Society B 371, 20150187 (2016). 543 32. Dukas, R. Evolutionary biology of animal cognition. Annual Review of Ecology, 544 Evolution, and Systematics 35, 347–374 (2004). 545 33. Ricklefs, R. E. The cognitive face of life histories. Wilson Bull. 116, 119–133 (2004). 546 34. Martin, T. E., Oteyza, J. C., Boyce, A. J., Lloyd, P. & Ton, R. Adult mortality 547 probability and nest predation rates explain parental effort in warming eggs with 548 consequences for embryonic development time. American Naturalist 186, 223–236 549 (2015). 550 35. Unzeta, M., Martin, T. E. & Sol, D. Daily Nest Predation Rates Decrease with Body 551 Size in Passerine Birds. The American Naturalist 196, 743–754 (2020). 552 36. Charvet, C. J. & Striedter, G. F. Developmental modes and developmental 553 mechanisms can channel brain evolution. Frontiers in Neuroanatomy 5, 1–5 (2011). 554 37. Finlay, B. L. & Darlington, R. B. Linked regularities in the development and evolution 555 of mammalian brains. Science 268, 1578–1584 (1995). 556 38. Herculano-Houzel, S. Isotropic fractionator: A simple, rapid method for the 557 quantification of total cell and neuron numbers in the brain. Journal of Neuroscience 558 25, 2518–2521 (2005). 559 39. Massen, J. J. M. et al. Brain size and neuron numbers drive differences in yawn 560 duration across mammals and birds. Communications Biology 4, 1–10 (2021). 561 40. Ramsey, G., Bastian, M. L. & Schaik, C. Van. Animal innovation defined and 562 operationalized. Behavioral and Brain Sciences 30, 393–437 (2007). 563 41. Reader, S. M. & Laland, K. N. Social intelligence, innovation, and enhanced brain size 564 in primates. Proceedings of the National Academy of Science, USA 99, 4436–4441 565 (2002). 566 42. Lefebvre, L. A global database of feeding innovations in birds. The Wilson Journal of 16 567 Ornithology 132, 803–809 (2021). 568 43. Barton, R. A. Embodied cognitive evolution and the cerebellum. Philosophical 569 Transactions of the Royal Society B: Biological Sciences 367, 2097–2107 (2012). 570 44. Gutiérrez-Ibáñez, C., Iwaniuk, A. N. & Wylie, D. R. Parrots have evolved a primate- 571 like telencephalic-midbrain-cerebellar circuit. Scientific Reports 8, 9960 (2018). 572 45. Brieuc, M. S. O. O., Waters, C. D., Drinan, D. P. & Naish, K. A. A practical 573 introduction to random forest for genetic association studies in ecology and evolution. 574 Molecular Ecology Resources 18, 755–766 (2018). 575 46. Hadfield, J. D. & Nakagawa, S. General quantitative genetic methods for comparative 576 biology: Phylogenies, taxonomies and multi-trait models for continuous and 577 categorical characters. Journal of Evolutionary Biology 23, 494–508 (2010). 578 47. Güntürkün, O., Ströckens, F., Scarf, D. & Colombo, M. Apes, feathered apes, and 579 pigeons: differences and similarities. Current Opinion in Behavioral Sciences 16, 35– 580 40 (2017). 581 48. Ströckens, F. et al. High associative neuron numbers could drive cognitive 582 performance in corvid species. Journal of Comparative Neurology (2022) 583 doi:10.1002/cne.25298. 584 49. Shanahan, M., Bingman, V. P., Shimizu, T., Wild, M. & Güntürkün, O. Large-scale 585 network organisation in the avian forebrain: A connectivity matrix and theoretical 586 analysis. Frontiers in Computational Neuroscience 7, 1–17 (2013). 587 50. Emery, N. J. Cognitive ornithology: the evolution of avian intelligence. Philosophical 588 Transactions of the Royal Society B: Biological Sciences 361, 23–43 (2006). 589 51. Lambert, M. L., Jacobs, I., Osvath, M. & von Bayern, A. M. P. Birds of a feather? 590 Parrot and corvid cognition compared. Behaviour vol. 156 505–594 (2019). 591 52. Ksepka, D. T. et al. Tempo and pattern of avian brain size evolution. Current Biology 592 30, 2026-2036.e3 (2020). 593 53. Herculano-Houzel, S., Manger, P. R. & Kaas, J. H. Brain scaling in mammalian 594 evolution as a consequence of concerted and mosaic changes in numbers of neurons 595 and average neuronal cell size. Frontiers in Neuroanatomy 8, 1–28 (2014). 596 54. Smaers, J. B., Mongle, C. S., Safi, K. & Dechmann, D. K. N. Allometry, evolution 597 and development of neocortex size in mammals. Progress in Brain Research 250, 83– 598 107 (2019). 599 55. Cárdenas, A. & Borrell, V. Molecular and cellular evolution of corticogenesis in 600 amniotes. Cell Mol Life Sci 77, 435–1460 (2020). 601 56. García-Moreno, F. & Molnár, Z. Variations of telencephalic development that paved 602 the way for neocortical evolution. Progress in Neurobiology 194, 101865 (2020). 603 57. Charvet, C. J. & Striedter, G. F. Developmental basis for telencephalon expansion in 604 waterfowl: Enlargement prior to neurogenesis. Proceedings of the Royal Society B: 605 Biological Sciences 276, 3421–3427 (2009). 606 58. Striedter, G. F. & Charvet, C. J. Developmental origins of species differences in 607 telencephalon and tectum size: Morphometric comparisons between a parakeet 608 (Melopsittacus undulatus) and a quail (Colinus virgianus). Journal of Comparative 609 Neurology 507, 1663–1675 (2008). 610 59. Sibly, R. M. & Brown, J. H. Effects of body size and lifestyle on evolution of mammal 611 life histories. Proceedings of the National Academy of Sciences 104, 17707–17712 17 612 (2007). 613 60. Uomini, N., Fairlie, J., Gray, R. D. & Griesser, M. Extended parenting and the 614 evolution of cognition. Philosophical transactions of the Royal Society of London. 615 Series B, Biological sciences 375, 20190495 (2020). 616 61. Lleonart, J., Salat, J. & Torres, G. J. Removing allometric effects of body size in 617 morphological analysis. Journal of Theoretical Biology 205, 85–93 (2000). 618 62. Reiner, A. et al. Revised Nomenclature for Avian Telencephalon and Some Related 619 Brainstem Nuclei. Journal of Comparative Neurology 473, 377–414 (2004). 620 63. Mullen, R. J., Buck, C. R. & Smith, A. M. NeuN, a neuronal specific nuclear protein 621 in vertebrates. Development 116, 201–211 (1992). 622 64. Mezey, S. et al. Postnatal changes in the distribution and density of neuronal nuclei 623 and doublecortin antigens in domestic chicks (Gallus domesticus). Journal of 624 Comparative Neurology 520, 100–116 (2012). 625 65. Revell, L. J. phytools: An R package for phylogenetic comparative biology (and other 626 things). Methods in Ecology and Evolution 3, 217–223 (2012). 627 66. Ducatez, S. & Lefebvre, L. Patterns of Research Effort in Birds. PLoS ONE 9, e89955 628 (2014). 629 67. Sheard, C. et al. Ecological drivers of global gradients in avian dispersal inferred from 630 wing morphology. Nature Communications 11, (2020). 631 68. Cooney, C. R. et al. Ecology and allometry predict the evolution of avian 632 developmental durations. Nature Communications 11, 2383 (2020). 633 69. Botelho, J. F. & Faunes, M. The evolution of developmental modes in the new avian 634 phylogenetic tree. Evolution & Development 17, 221–223 (2015). 635 70. Bürkner, P.-C. Brms: An R Package for Bayesian Multilevel Models using Stan. 636 Journal of Statistical Software 80, 1–28 (2017). 637 71. Pigot, A. L. et al. Macroevolutionary convergence connects morphological form to 638 ecological function in birds. Nature Ecology & Evolution 4, 230–239 (2020). 639 72. Jetz, W., Thomas, G. H., Joy, J. B., Hartmann, K. & Mooers, A. O. The global diversity 640 of birds in space and time. Nature 491, 444–448 (2012). 641 73. Berk, R. A. Statistical learning from a regression perspective. (Springer International 642 Publishing Switzerland, 2017). 643 74. Liaw, A. & Wiener, M. Classification and regression by randomForest. R News 2, 18– 644 22 (2002). 645 75. Sayol, F., Downing, P. A., Iwaniuk, A. N., Maspons, J. & Sol, D. Predictable evolution 646 towards larger brains in birds colonizing oceanic islands. Nature Communications 9, 647 2820 (2018). 648 76. Torres, C. R., Norell, M. A. & Clarke, J. A. Bird neurocranial and body mass evolution 649 across the end-Cretaceous mass extinction: The avian brain shape left other dinosaurs 650 behind. Science Advances 7, 1–10 (2021). 651 652 653 18 654 Acknowledgments. This research was funded by MINECO (PID2020-119514GB- 655 I00, to D.S.), the Czech Science Foundation (18-15020S, to P.N.), the Grant Agency 656 of Charles University (1438217 to M.K.) and the European Union’s Horizon 2020 657 research and innovation program under the Marie Skłodowska-Curie postdoctoral 658 fellowship (grant agreement 838998, to F.S). We thank Andrew Iwaniuk, Simon 659 Reader, Simon Ducatez and Julie Morand-Ferron for discussions, and the R-team and 660 package contributors for facilitating analytical tools. This paper is dedicated to Julie 661 Morand-Ferron. 662 663 Author contributions: LL and FS proposed the study; DS, LL, FS and PN conceived 664 the study; DS and TM elaborated the life history framework; PN, SO, LC, MK, YZ, 665 CO collected the neuroanatomical data, LL the innovation data, and JGP and FS the 666 endocast data; DS designed and conducted the analyses, with contributions of FS, LL, 667 PN and TM; DS wrote the manuscript and all authors edited and approved it. 668 669 Competing interests: The authors declare no competing interests. 19 670 Figures 671 672 Fig. 1. Framework linking cognition, neuron numbers and brain size. a, Enhanced 673 cognition is assumed to require more neurons in the pallial telencephalon, and perhaps also 674 in the cerebellum. Thus, an increase in pallial neurons relative to the ancestors is expected in 675 lineages that have been selected for higher intelligence. b, Because pallium comprises a large 676 fraction of the mass of the brain, a disproportionate accumulation of pallial neurons should 677 enlarge the brain relative to body size. c, If the benefits of enhanced cognition increase with 678 body size, selection for cognition should further increase brain size in large species. As a 679 result, species that excel at cognitive performance should have brains that are large in both 680 absolute and relative terms. d, A mechanism that may allow accumulation of more neurons 681 in the pallium is to extend the period of development, particularly in the later stages. 682 According to some evo-devo theories, extending the later stages of development increases 683 neurogenesis in the areas of the brain where progenitor cell multiplication stops later, i.e. the 684 pallial areas of the telencephalon. Thus, if a longer development period facilitates 685 neurogenesis in pallial regions, it may be targeted by selection for increased intelligence. e, 686 Phylogenetic relationships among the species analysed for neuron numbers to address 687 hypotheses b-c (for an enlarged tree with species names see Supplementary Fig. 1). Silhouette 688 illustrations are from PhyloPic (http://phylopic.org), contributed by Anthony Caravaggi and 689 Ferran Sayol under public domain licence. 690 20 691 Fig. 2. Neurons and innovation propensity. Relationship between neuron numbers and 692 innovation propensity for the entire brain and the pallium, cerebellum and brainstem, as 693 predicted by models. a, Absolute neuron numbers. b, Neuron numbers adjusted by body size 694 by including body mass (previously subtracting brain mass) as co-variate in the model. c, 695 Density of neurons (cells per mg). All models account for the effect of phylogeny, 696 biogeographic realm and confounding variables (see Supplementary Tables 1 and 2 for 697 details). Lines and credibility intervals are derived from Bayesian phylogenetic mixed 698 models. Sample size is 99 species, as nocturnal specialists (i.e. owls) are excluded from the 699 innovation database. 700 21 701 Fig. 3. Neuron numbers and brain mass as a function of body size. a-b, Distribution of 702 neuron numbers among pallium, cerebellum and brainstem for clades belonging to low-slope 703 (a) and the highest slope (b) grades. The assignation of species to each slope grade group is 704 based on Ksepka et al.52. c, Variation in neuron numbers in the entire brain as a function of 705 body size. d-e, Variation in brain mass as a function of body size for the sample of species 706 used in analyses of neurons (d) and for the entire brain-body data set (e). In c-d, clades with 707 low-slope grades are shown in blue while clades with the highest slope grades are shown in 708 red. In all plots, owls have been excluded. For plots based on the entire sample of species, see 709 Supplementary Fig. 3. 710 22 711 Fig. 4. Neuron numbers as a function of absolute and relative brain size. a, Bivariate 712 dependence plots representing neuron numbers in the entire brain and main brain regions as 713 a function of absolute and relative brain size, based on the predictions from random forests. 714 Colours describe neuron numbers, with low numbers represented by dark-blue colours and 715 higher numbers by yellow-green colours. Relative brain size was estimated by means of the 716 normalized scaled brain index61, with the allometric exponent estimated excluding clades that 717 have been found to exhibit substantial grade shifts in brain:body allometries 718 (NSBIgrades, see Methods).. b, Univariate representations (partial dependence plots) for 719 relative brain size to further interpret the bivariate dependence plots. The plots show the 720 dependence between neuron numbers and relative brain size, marginalizing over the values 721 of absolute brain size. c, Univariate representation of the bivariate dependence plot for 722 absolute brain size. In all analyses, owls have been excluded. For analyses with the entire 723 sample of species, see Supplementary Figs. 6 and 7. 724 23 725 Fig. 5. Neurons and development in species belonging to low- and the highest slope 726 grades. Neuron numbers as a function of the duration of development (embryonic stage plus 727 post-natal growth)(a) and the fraction of total development time represented by the post-natal 728 growth (b), for low-slope grades (blue bar) and the highest slope grades (red bar). Lines and 729 credibility intervals are derived from Bayesian phylogenetic mixed models. In all analyses, 730 owls have been excluded (for analyses with the entire sample size, see Supplementary Fig. 731 8). 24