646 JOURNAL OF COSMETIC SCIENCE As researchers have come to realize that hair shaft properties may be emergent rather than due to a single cause, it has been recognized since the early 2000s in wool (18) and more recently in hair (19,20) that this requires investigation using a systems biology approach. Within this complex system, we expect to see some phenomena that correlate with curliness but do not cause curliness. This is a common problem in biology called a spurious correlation. A version of the issue was popularized by Gould and Lewontin (21) in the context of evolutionary biology. In the context of fiber growth, for example, various studies find that both the expression of inner root sheath keratins (22) and hair shaft ellipticity correlate with hair curl (23,24). But, does the keratin expression cause the ellipticity, or is it a consequence of ellipticity because follicles have a circular cross-section, and the inner root sheath fills in the difference between ellipse and circle? Or are both caused by a common causative mechanism, such as the organization of the neck of the dermal papilla that orchestrates the fate of cells into the various lineages including shaft and inner root sheath during hair growth (25)? This particular issue was first raised by Priestley (26), and despite decades of additional data slowly filling in our understanding of follicle biology, we still do not know the answer to this example question. An experimental approach is essential to directly tease out noncausative correlations from causation in addition to comparative analysis. WHAT FIBER FEATURES CORRELATE WITH CURL? WITHIN THE MATURE HAIR SHAFT Internal organization of the hair cortex, follicle morphology, the shape of the fiber cross- section, and differences in protein abundance have been described in the literature as features that correlate with curl. For more than 90 years, researchers have been trying to work out how fiber curvature correlation with mammalian hair features might explain a mechanism causing curliness. Studies have largely focused on three contexts: human scalp hair, midside wool from sheep, and mice (typically as models of genetic disorders affecting human scalp hair). Although there are some common features that correlate with hair shaft curl in scalp hair, sheep wool, and mouse hair (plus other species), the differences are particularly informative for identifying and discounting features underpinning hair curliness in all mammals. For example, studies of human scalp hair often point out that curly hair shafts tend to have a more elliptical cross-section than straight hairs (27–29). Ellipticity alone is not a mechanism, but could an elliptical profile be immutably connected to the process that generates curl? Sheep wool indicates that this is not the case, and there are clear examples of the reverse situation where hair shafts from merino and merino relatives that have a more elliptical profile are less curly than those with circular profiles (30). This demonstrates the challenge of correlating just one feature with curl. It has been appreciated since the 1950s that cell-type distribution in the cortex region of the fiber correlates with various measures of hair curliness (31). Sheep wool with lower- diameter hairs (15–25 μm) is the most curly and has been investigated as a model to study fiber curvature because it is made of three distinct modes of keratin organization (ortho-, meso-, and paracortex) embodied within macrofibrils composed of keratin intermediate filaments and matrix (32,33). The macrofibril-containing cells of the hair shaft, once living in the follicle, now dead and transformed to structural elements in the mature hair, are

647 WHAT CAUSES CURLY HAIR? themselves typed as ortho- or paracortical (or sometimes mesocortical). In high-curl wool fibers, such as those from the merino breed, the paracortical cells are clustered along the concave margin of the cortex, forming a strip with orthocortex on the convex side. The relation between cell type in low-diameter wools has become a definition applied to hair from other mammals. However, the direct transfer of wool cortical terminology to other hairs has been an oversimplification. In human, goat, alpaca, rabbit, deer, and higher- diameter sheep fibers (25–50 μm), cortical cell types can be ambiguous in both chemistry and structure, and single cells can contain macrofibrils of multiple architectures (10,17,32– 34). In particular, while the traditional view is that the intermediate filaments making up paracortical and mesocortical macrofibrils are arranged in parallel arrays, while those of the orthocortex are twisted into rope-like columns, recent studies show that the intensity of twisting varies. This variable twist means that many cortical cells contain features associated with both orthocortical and paracortical cell types. The details of cortical cell types are important because their organization not only correlates with curvature but has also been implicated as part of proposed mechanisms of curvature. Munro and Carnaby (35) developed a mathematical model that described how differential contraction of the cell types can explain fiber curvature in wool. Their hypothesis assumes that the fiber in the follicle is in a moisture-saturated state, and when the fiber dries during the transition to the mature state, the matrix located between intermediate filaments shrinks laterally. Because orthocortex is composed of helical wound ropes of intermediate filaments, lateral contraction should affect intermediate filament tilt and macrofibril length differently than in the paracortex. The resulting differential contraction then strains the fiber to curve toward the paracortex. With the Munro and Carnaby model in mind, studies on wool fibers have looked at differences between the orthocortex and paracortex using light and electron microscopy (13,36), atomic force microscopy (37,38), electron tomography (33,39,40), genomics (41,42), proteomics (43–47), microbeam small angle x-ray spectroscopy (48), elemental (49), and thermal analysis methods (50,51). The data have informed further models (52,53). Largely these studies have supported the models, but there have also been some findings that challenge the models, such as the discovery that macrofibrils can have variable tilt (46), that the proportion of orthocortex and paracortex does not correlate with curvature (13,51), and instead that the relative difference in length between orthocortical cells close to the outside of the curvature and paracortical cells close to the inside of the curvature correlates to curvature at that point in the wool fiber (13). Despite these new findings, the Munro and Carnaby–type models remain a good conceptual foundation for explaining curl in wool fibers. However, in human hair, the bilateral cell-type distribution is not typically observed in the cross-sectional images as it is in low-diameter wool fibers. Underlying this is the fact that cortical cell types are not as clearly differentiated in hair as they are in wool, and this occurs at the levels of macrofibril structure (33), cortical cell types and microscopy staining chemistry (33,39,54,55), and in intermediate filament angles (56). Historically, human scalp hair cortex has been described as the same as for wool (57,58), as orthocortex only (59), as like but not identical to wool (56,60), or using new terms (e.g., meta- or heterocortex) (61–64). Using a bespoke classification for human hair based on differences in macrofibril appearance using transmission electron microscopy, fluorescent stains, and electron tomography, Bryson et al. (39) demonstrated that differences analogous to those seen in wool in the cell-type distribution also occur in moderately curly and straight scalp