The Adaptable Mind: What Neuroplasticity and Neural Reuse Tell Us about Language and Cognition - John Zerilli 2021
Implications of Neuroplasticity
Are Modules Innate?
Of all the instances of cortical map plasticity we reviewed in Chapter 2, undoubtedly the most impressive involve crossmodal changes in which brain regions deprived of their typical inputs come to subserve alternative uses. One example I mentioned there concerned early-blind Braille readers whose visual cortex appears to be functionally important for Braille character identification, suggesting a functional contribution of the reorganized occipital cortices during complex tactile discrimination tasks (Sadato et al. 1996). Moreover, when repetitive transcranial magnetic stimulation (rTMS) is used to impair the functioning of the occipital cortex, blind subjects appear to have difficulty performing embossed-character recognition, while sighted control subjects do not, again pointing up the functional significance of early-blind occipital cortices during tactile discrimination (Cohen et al. 1997). Probably the most famous case of crossmodal plasticity is that of the rewired ferrets whose visual cortex was induced to project into auditory cortex after their retinal nerves were rerouted so that, instead of feeding into primary visual cortex, they fed into primary auditory cortex via the auditory thalamus (Sharma et al. 2000; Melchner et al. 2000). The manipulation resulted in ferret auditory cortex’s taking on features typical of occipital cortex, such as columnar orientation and stimulus selectivity. Besides these cases, language studies suggest that this sort of plasticity is not confined to sensory-motor cortices alone, as the case of EB discussed in Chapter 2 illustrates very well.
While these results seem quite dramatic, nevertheless, some aspects of the evidence do not fit well with the idea of the brain as open-endedly malleable. In fact, rather than supporting the case for plasticity tout court, these results argue the case for what Laurence and Margolis (2015) call “constrained plasticity.” Take the ferret case. The clear suggestion here is that auditory cortex came to resemble the processing structures typically associated with occipital cortex. And indeed, to some extent, this is what seems to have happened. But in fact, primary occipital cortex is a complicated structure, “connected to a large number of distinct brain regions that support further specific types of visual processing, including computations responsible for downstream representations of location, direction of motion, speed, shape, and so on” (Laurence & Margolis 2015, p. 127). And there is no evidence that any of this complex processing structure was reproduced, for “the overall wiring of the ferrets’ auditory cortex was largely unchanged.” One interpretation of why a “largely unchanged” auditory cortex was able to process visual inputs is consonant with the theory of supramodal organization (and Pascual-Leone and Hamilton’s metamodal hypothesis). Recall that this theory posits a large number of intrinsically stable neural operators that are more or less suited to processing specific types of input, but are at the same time metamodal in that they receive inputs from many domains (i.e., they are really domain-general, or formally domain-specific). From this perspective, we would naturally expect there to be something about visual and auditory stimuli that makes them ideal for a neural operator whose processing disposition makes it suited to process one or the other of these specific types of input. Bregman and Pinker (1978) long ago postulated high-level analogies in computations that involve auditory and visual stimuli (e.g., different pitches are analogous to different locations, pronounced changes of pitch are analogous to sudden changes in the direction of motion, etc.). If such analogies between hearing and vision hold, it would suggest—consistently with the metamodal hypothesis—that auditory cortex did not really need to change when it began to receive inputs from a domain to which its processing capabilities were already well suited. As Laurence and Margolis interpret the ferret case:
even though the rewiring experiments show that the auditory cortex can be recruited for a certain amount of visual processing, this is because the auditory cortex and the visual cortex overlap in the types of computations they naturally support. Far from being a model case of the environment instructing an equipotential cortex, [the ferret] rewiring experiments illustrate the way in which cortical structure and function remain largely unchanged even in the extreme case of input coming from a different sensory system. (2015, p. 128)
Next, consider the case of EB from Chapter 2. EB recovered most of his language skills two years after undergoing a left hemispherectomy at the age of two and a half and tested as largely normal with respect to language at age 14, albeit with his language faculty now subserved by regions in his right cerebral hemisphere. Surely this argues for an almost equipotential cortex early in development, if anything does? Not quite. The fMRI evidence showed that the pattern of activation in his right hemisphere was almost isomorphic to that of the left hemisphere in normal control subjects, revealing a definite and predictable cortical pattern. Language did not arbitrarily migrate to a new location: it moved to the very site in the right hemisphere whose structural features most nearly resemble those of the left hemisphere’s language circuits. A truly equipotential brain would presumably reconfigure cortical sites selected on a far more ad hoc basis. The most important take-home message here, then, is not that the brain is open-endedly plastic, but rather that “the brain’s two hemispheres incorporate a large measure of potential redundancy of function that can be exploited at certain stages of development” (Laurence & Margolis 2015, p. 126; see also my § 7.5).
These cases are only a beginning. By far the most significant evidence for constrained plasticity and the robust development of brain regions comes from studies revealing the brain’s latent supramodal organization. A flavor of this evidence was given in § 2.4.3, but it is instructive to consider a few more examples to drive the point home. It will be remembered that evidence of supramodal organization first came from studies of the two major visual processing streams; i.e., the dorsal (“action”) path for space and motion discrimination, and the ventral (“categorization”) path for object and shape recognition. What these studies suggest is that this dual stream processing structure persists with the same functional role and structural characteristics in both early and congenitally blind and sighted subjects. That is to say, even total and protracted visual input inhibition—from the very earliest developmental stages onwards—appears to have few if any adverse effects on the development of typical visual processing structures in humans. To repeat the conclusion one researcher drew from the case we examined in Chapter 2, “despite the vast plasticity of the cortex to process other sensory inputs,” these findings suggest “retention of functional specialization in this same region” (Striem-Amit & Amedi 2014, p. 4). The dorsal and ventral processing streams appear to be modular, developmentally constrained, and functionally preserved despite complete early and congenital visual impairment.
In one study (Renier et al. 2010), early-blind subjects were presented with paired auditory stimuli that differed either in type (in this case, different piano chords) or locality.3 The task required subjects to indicate whether the pairings were of the same type or emanated from the same location. Subjects exhibited differential activation in a region of the dorsal visual stream—specifically, the area rostral to the right middle occipital gyrus (MOG)—when engaged in the auditory spatial location task relative to the sound-type identification task. Similar results were obtained on an analogous tactile discrimination task using the same subjects. So while the MOG is clearly plastic, in that early-blind individuals recruit this area more intensively for auditory and tactile discrimination tasks than do sighted individuals, its plasticity reveals it to be functionally constrained and structurally preserved. It is classically supramodal in that it continues to perform a fixed computation despite receiving different sensory input. Other studies attest to the persistence of the spatial location function of the dorsal visual stream. Consider the posterior parietal cortex (PPC), implicated in the spatial representations that guide action. In healthy sighted subjects, caudal subregions play a relatively larger role in reaching and grasping than do rostral subregions, which are primarily engaged in the planning and execution of action. Lingnau et al. (2014) showed that the same response gradient occurs in the congenitally blind, concluding that “neural plasticity acts within a relatively rigid framework of predetermined functional specialization” (2014, p. 547). Other studies evidence preservation of the direction representation function of the dorsal visual stream, as judged by performance of congenitally blind subjects on analogous auditory discrimination tasks (Wolbers et al. 2011), as well as functions in the ventral visual stream in both congenitally blind and blindfolded sighted subjects (as we saw in Chapter 2) (Striem-Amit et al. 2012; Striem-Amit & Amedi 2014). Laurence and Margolis conclude their review of this evidence in the following way:
it would appear that the large-scale functional architecture of the visual cortex—the division of labor between the dorsal and ventral streams—develops in much the same way, and with the same functions being performed in various subregions of these streams, with or without visual experience. (2015, p. 133)
And of course all of this evidence once again testifies to the supramodal organization of the brain, and Pascual-Leone and Hamilton’s metamodal hypothesis in particular, since it is consistent with a brain that is composed of a number of “distinct computational systems whose functions are established independently of their sensory input” (Laurence & Margolis 2015, p. 428) and in which “multimodal sensory inputs feed into all cortical regions” (Pascual-Leone & Hamilton 2001, p. 432), even though the operations of a given region will dictate certain preferences. The metamodal hypothesis predicts that “when the preferred input is unavailable, the brain switches to the next best fit” (Laurence & Margolis 2015, p. 428) such that a region’s underlying computational structure and profile need undergo no truly radical alteration in the face of new processing inputs—in the standard case, it will perform in much the same way it always did, albeit on a new set of afferents. In this view, even many dramatic instances of crossmodal plasticity, where the equipotential nature of the cortex seems to be its most obvious feature, need involve little more than a straightforward remodeling of supramodal connection channels (Pascual-Leone & Hamilton 2001, p. 443).
One final study is especially worth mentioning for the illumination it provides on the precise extent to which predefined cortical functionality is developmentally robust. A group of mice whose brains were genetically modified so that they were incapable of synaptic transmission, and therefore incapable of releasing any neurotransmitters at all, were compared to normal control littermates. Mice in whom the potential for all synaptic transmission has been inhibited in this way have effectively no potential for learning or indeed any activity-dependent cell differentiation. Verhage et al. (2000) reported that, at least prior to birth, the two brain types were assembled correctly, and were in fact essentially similar. As they state their own findings:
Neuronal proliferation, migration and differentiation into specific brain areas were unaffected. At [embryonic day 12], brains from null mutant and control littermates were morphologically indistinguishable. . . . At birth, late-forming brain areas such as the neocortex appeared identical in null mutant and control littermates, including a distinctive segregation of neurons into cortical layers. . . . Furthermore, fiber pathways were targeted correctly in null mutants. (2000, p. 866)
This means activity-independent changes are robust enough to withstand severe synaptic privation, and “that many features of even the fine-grained structure of the brain can develop without any sensory input or feedback” (Laurence & Margolis 2015, p. 130).
Notice, incidentally, just what this sort of neuroconstructivist nativism implies: that while there is a certain (and relative) sense in which M-networks and other functionally significant brain regions are developmentally robust, the same cannot be said for higher-level cognitive functions. There is a weak sense, of course, in which the innateness of M-networks translates to the innateness of higher-level/gross functional composites, which are innate insofar as the parts used in assembling them are innate. But this claim is different from the claim that such higher-level composites are innate as organized (Jusczyk & Cohen 1985). If the “derived” innateness of a functional composite were sufficient for its being considered innate as an organized ensemble, all complex cognitive functions would be innate by default, which is plainly absurd. I shall revisit this matter in Chapter 7 (see pp. 120—121).
Before I leave this chapter, it will be useful to delineate once again the relationship between neuroplasticity (qua Hebbian learning) and neural reuse, for there is a good deal of complementarity on offer here that is easy to miss amid the detail of specific cases. The supramodally organized brain in effect constitutes the architectural foundation upon which Hebbian synaptic mechanisms operate. That is to say, Hebbian plasticity presupposes reuse, inasmuch as it consists in the strengthening (or weakening) of existing supramodal connection channels. Synaptic pruning, synaptogenesis, and other forms of interneural transmission can no doubt account for the more drastic examples of plastic change and post-pathological recovery we examined in Chapter 2 (yielding “a change in use from a change in working,” in the language of Chapter 3), perhaps joining a suite of mechanisms that could account for the very youngest cortico-cortical pathways established in the developing brain (in effect supplying us with a supramodal architectural foundation).4 But Hebbian mechanisms remain an important part of the story of how patterns of neural reuse are regularly refined and remodeled in the course of normal development, learning, and recovery after injury (yielding “a change in use without a change in working,” as we saw in Chapter 3).