Modeling transformations of neurodevelopmental sequences across mammalian species

Abstract

A general model of neural development is derived to fit 18 mammalian species, including humans, macaques, several rodent species, and six metatherian (marsupial) mammals. The goal of this work is to describe heterochronic changes in brain evolution within its basic developmental allometry, and provide an empirical basis to recognize equivalent maturational states across animals. The empirical data generating the model comprises 271 developmental events, including measures of initial neurogenesis, axon extension, establishment, and refinement of connectivity, as well as later events such as myelin formation, growth of brain volume, and early behavioral milestones, to the third year of human postnatal life. The progress of neural events across species is sufficiently predictable that a single model can be used to predict the timing of all events in all species, with a correlation of modeled values to empirical data of 0.9929. Each species' rate of progress through the event scale, described by a regression equation predicting duration of development in days, is highly correlated with adult brain size. Neural heterochrony can be seen in selective delay of retinogenesis in the cat, associated with greater numbers of rods in its retina, and delay of corticogenesis in all species but rodents and the rabbit, associated with relatively larger cortices in species with delay. Unexpectedly, precocial mammals (those unusually mature at birth) delay the onset of first neurogenesis but then progress rapidly through remaining developmental events.

Figure 1.

The distribution of the neural location (left) and general class of developmental process (right) of events used to develop the model, plotted against the event scale. The event scale, which is described in more detail in Figure 3, is a ranking of all these events on a scale from 0 to 1, and principally represents their order of occurrence across all of the species. New events added in this data analysis compared with the original model (Clancy et al., 2001) are in red and show not only the addition of late behavioral, physiological, and overall growth information to the model but also addition to the database throughout; events common to both in black.

Figure 2.

Predicted developmental schedules for human (blue circle), macaque (red circle), cat (yellow circle), short-tailed opossum (green circle), and mouse (black circle), selected from the 18 species to illustrate the full range of developmental durations. In this graph the event scale is the x-axis, to which we have added a subset of the 271 events that were observed. This scale ranges from 0 to 1, but in this case, event scale numerical values are replaced by these example events. As will be described in Figure 3, the event scale is a common ordering of developmental events across all species. The y-axis is the estimated date of occurrence of each event in each species from conception (log scale). To determine when a particular event would be predicted to occur in any species from this graph, using the name of the event on the event scale, find where it intersects the regression line for that particular species. The y-axis value will be the predicted PC day for that event/species combination. In future graphical representations of the event scale, the event scale value for any named event can be found in Table 1. Also represented on this graph are interaction terms for corticogenesis and retinogenesis, with interaction terms always associated with individual species. The parallel lines for a subset of events in four of the species (black bordered circles for human, macaque, cat, and possum) represent delays in cortical neurogenesis with respect to their time of occurrence in the rodent and rabbit. In the cat, a second parallel line can be seen representing the delay of retinal neurogenesis (yellow circle with a black dot).

Figure 3.

The relation of the event scale used in this analysis to the ordering of events that was used in the prior model. Plotted on the x-axis are the event scores determined by GLM as in Clancy et al. (2001), estimating values for all species-event combinations by determining their overall ranking, and fitting those scores on a scale from 0 to 1.5. The event scale of the present analysis plotted on the y-axis was derived by the quasi-Newton optimization method described in the text. The two values derived for each event are very highly correlated, and also conform to the necessary biological ordering of events (e.g., cell birth before cell death). Each of these two scales was designed to correlate perfectly, within its own dataset, with the estimated Y values for any rodent and rabbit species. This shows that the quasi-Newton optimization method essentially ranks the order of occurrence of events in development, but it differs in two important respects. First, it shows a statistically significant nonlinearity, most obvious in the higher values on each scale. Second, some events fall noticeably above the main regression line, the largest cluster are those to the left of the arrow on the figure, and may represent events in early axonogenesis and GABA-neuron generation, which are either very substantially shifted in their order between species or nonhomologous events.

Figure 4.

Fit of empirical data to model values in four representative species. The x-axis is the predicted date of occurrence of all the events for which there were also empirical data points; the y-axis is the empirical value of the measured event in that species. As described in several technical aspects in the texts, the fit of predicted to actual days is very high.

Figure 5.

Plotted are all of the empirically measured developmental events (y-axis) in the cat, versus predicted developmental time for each event (x-axis), before the fitting of a retinal neurogenesis interaction term. All of the events in retinal neurogenesis are highlighted by black circles and all other events are in gray. The retinal neurogenesis events all occur later than would be predicted by the general model for the cat. Consistent with the hypothesis that this developmental delay has occurred for the purpose of increasing the retinal cell populations associated with nocturnality, within the retinal neurogenesis group only retinal ganglion cells are delayed. Retinal ganglion cells are the only measured cell class in the cat retina that are neither rods nor have a connection with rods.

Figure 6.

Phylogenetic tree plotting the positions of relative delay in cortical neurogenesis, to determine whether delay in onset of cortical neurogenesis, associated with a taxonomic grade shift up in neocortical volume, had occurred at single or multiple times in mammalian phylogeny. Bold pluses and minuses mark branches with delays described in this dataset, and which also have species with increased adult cortical volume, while the unmarked ones represent branches where only relative cortical volume differences in adults are known (Reep et al., 2007). Increases in relative cortical volume appear to have occurred at least three times in mammals, in Metatheria (various marsupials), in the branch giving rise to carnivores, various ungulates, and dolphins, and in primates. It cannot be determined from these data as yet whether a large cortex is the “basal” state, and reduced cortical volume has been selected for in glires, bats, shrews, and Afrotheria, or the reverse. Phylogeny from Song et al. (2012).

Figure 7.

The maximum duration of neurogenesis across species, plotted on the basic segmental divisions of the embryonic vertebrate nervous system. Only one side of the neural plate is plotted, with the midline to the right, and more alar or lateral locations to the left. Rhombomeres 1–11, which give rise to the medulla, pons, and cerebellum, and the prosomeres, which give rise to the diencephalon and telencephalon, are indicated. The neural plate representation compresses the segmental divisions described previously (Garcia-Lopez et al. (2009) into four medial to lateral divisions. The latest date of neurogenesis observed in each division is plotted on the z-axis, as given by the mean of the species' event scores found there. Extended duration of neurogenesis is more likely in more alar (lateral) and more rostral divisions.

Figure 8.

Comparison of the modeled relative rates of brain development for a metatherian (marsupial) and a eutherian (placental) mammal of similar adult brain sizes. The marsupial species is the Polyprodonta dunnart, the marsupial mouse, and the placental mammal is a rodent, the common mouse M. musculus. The x-axis is the event scale, and the y-axis shows the predicted PC day of the occurrence of each event in both species. In this graph, the y-axis is a linear, not a log scale, so that the differences in duration can be better appreciated in this case where brain sizes are comparable. The elevated points near the dunnart line represent the effect of the non-glires interaction term delaying cortical neurogenesis with respect to the mouse. The dunnart takes nearly twice as long to reach 80% of its adult brain size compared with the laboratory mouse, with later developmental events disproportionately protracted compared with early events.

Figure 9.

The position of birth for six placental mammals relative to the event scale (x-axis); the age of each mammal in PC days can be read for birth (or any event scale value) on the y-axis. The five placental mammals are chosen to represent close to the full range of the dataset and include one highly precocial mammal, the guinea pig. For an example, in the mouse at birth cortical neurogenesis is still underway and synaptogenesis in the forebrain is only beginning, while in the guinea pig at birth, cortical neurogenesis, cortical cell migration, and basic axonogenesis is entirely complete, and the point of peak synaptic density has passed (human (blue dot), macaque (red dot), cat (yellow cat), guinea pig (green dot), and mouse (black dot).

Figure 10.

The relationship of adult brain weight (x-axis, log scale), modeled slope (left), and intercept (right) values for all eutherian mammals. The slope of the modeled regression line (left) can be used as a measure of the number of days, and thus the total developmental duration required to reach the last maturational point measured is closely related to brain weight. The intercept (right), which gives the PC age when the first neurodevelopmental events occur, distinguishes precocial from altricial animals, and also relates to brain size. Of the rodents, for example, the precocial spiny mouse and guinea pig have the highest intercepts, while the altricial mouse and hamster have the lowest.

Figure 11.

Top, Modeled regression slopes for two highly precocial rodents, the spiny mouse and guinea pig, compared with the altricial mouse (M. musculus) whose brain weight is similar to the spiny mouse, and the relatively altricial ferret, whose brain weight is close to the guinea pig; the regression slope for the macaque is also plotted. x-axis is the event scale, and the y-axis is predicted PC day (logged). Both precocial cases show the delayed onset of in utero neurodevelopment and very rapid progress through developmental stages characteristic of the precocial group. Elevated points associated with the ferret and macaque regression lines are the non-glires*corticogenesis interaction term. Bottom, Empirical data density and distribution, and fit of data to model, for the three most precocial species, the spiny mouse, guinea pig, and sheep. All modeled points are represented in gray, while the points for which there are empirical measurements are black circles.