Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Apr 1.
Published in final edited form as: J Invest Dermatol. 2024 Sep 3;145(4):800–811.e8. doi: 10.1016/j.jid.2024.08.014

Deep Hair Phenomics: Implications in Endocrinology, Development, and Aging

Jasson Makkar 1, Jorge Flores 1, Mason Matich 1, Tommy T Duong 1, Sean M Thompson 1, Yiqing Du 1, Isabelle Busch 1, Quan M Phan 1, Qing Wang 3, Kristen Delevich 2,3, Liam Broughton-Neiswanger 4, Iwona M Driskell 1, Ryan R Driskell 1,2,*
PMCID: PMC11873809  NIHMSID: NIHMS2025756  PMID: 39236901

Abstract

Hair quality is an important indicator of health in humans and other animals. Current approaches to assess hair quality are generally non-quantitative or are low throughput due to technical limitations of ‘splitting hairs’. We developed a deep learning-based computer vision approach for the high throughput quantification of individual hair fibers at a high resolution. Our innovative computer vision tool can distinguish and extract overlapping fibers for quantification of multivariate features including length, width, and color to generate single-hair phenomes (shPhenome) of diverse conditions across the lifespan of mice. Using our tool, we explored the effects of hormone signaling, genetic modifications, and aging on hair follicle output. Our analyses revealed hair phenotypes resultant of endocrinological, developmental, and aging-related alterations in the fur coats of mice. These results demonstrate the efficacy of our deep hair phenomics tool for characterizing factors that modulate the hair follicle and developing new diagnostic methods for detecting disease through the hair fiber. Finally, we have generated a searchable, interactive web tool for the exploration of our hair fiber data at skinregeneration.org.

Keywords: A.I., Hair, Aging, Development, Endocrinology

Graphical Abstract

graphic file with name nihms-2025756-f0006.jpg

Introduction

Hair is an important evolutionary trait conserved across Mammalia for its function in thermoregulation, protection from UV exposure, and environmental mechanosensing (Agramunt et al., 2023; Hamann, 1995; Lasisi et al., 2023; Li et al., 2016). Hair loss from diseases, such as androgenetic alopecia, affect a significant proportion of men and women (Ho et al., 2024, p. 202). In addition to the functional role hair serves, its appearance plays a considerable part in an individual’s perception by society and their mental health and well-being (Schielein et al., 2020). Dissection of the cellular and molecular processes that govern the development and maintenance of the hair follicle have greatly advanced the fields of dermatology and trichology (Duverger & Morasso, 2009; Millar, 2002; Schmidt-Ullrich & Paus, 2005). However, the characterization of phenotypes associated with the hair fiber itself have been largely neglected.

Deviations in normal hair fiber growth, such as thinning and graying, have been noted in conjunction with hormonal imbalance, chronic stress, and dementia (Grymowicz et al., 2020; Hasan et al., 2022; Mendelsohn & Larrick, 2020; Neau et al., 2014; Thom, 2016). Typically, the changes seen in aging and disease are documented only once clearly visible, but the age of onset for these changes and the rate of follicular deterioration are currently difficult to study (Almashali et al., 2023; Liu et al., 2016). For instance, the effects of androgenetic alopecia are typically noted once it has progressed to obvious hair loss (Wolff et al., 2016). While therapies to prevent hair loss are effective, the inability to reverse damage leads to permanent hairlessness (Wolff et al., 2016). However, monitoring of hair fiber phenotypes may facilitate early detection and subsequent treatment, improving therapeutic outcomes. The largest hurdle in detecting the gradual degradation of hair lies in the inability to accurately quantify them en masse at the level of individual fibers. Due to technical limitations, hair has been predominantly assessed visually, and the few fiber-level evaluations performed required tremendous time and labor, limiting reproducibility (Chi et al., 2013, p. 201; Driskell et al., 2009; Schlake, 2005; Takeo et al., 2023; Weger & Schlake, 2005b, 2005a). With the revolutionary capabilities of artificial intelligence (A.I.), state-of-the-art computer vision techniques provide the ability to overcome the constraints of fiber analysis and enhance our understanding of hair biology.

Recently, advances in computer vision frameworks have propelled applications for the detection and monitoring of skin-related ailments, including skin cancer and scarring (Jeong et al., 2023; Kim et al., 2023; Tschandl et al., 2020; Wei et al., 2024). Similar approaches have also been utilized to facilitate automated analysis of trichoscopy images (Di Fraia et al., 2023; GAO et al., 2022). While trichoscopy is effective for evaluating hair density and scalp health, it is unable to capture the information contained within the complete, intact hair fiber. The ability to extract full length hairs using machine learning algorithms has previously been restricted due fiber overlapping and the irregular nature of their curvature and orientation (Arteta et al., 2013; Wang et al., 2022). Deep learning - a method of teaching computers to recognize complex patterns - has revolutionized the development of scalable methods of object identification and quantification in image data for a wide variety of fields (Chai et al., 2021; Esteva et al., 2021). When applied to images of scientific interest, these tools enable the collection of metrics that were previously challenging to capture or could only be described qualitatively (Archila et al., 2022; Bae & An, 2023). Implementations of deep learning have been shown to be powerful instruments for research in instances such as distinguishing seed varieties and counting nematodes (S. Mori et al., 2022; Pun et al., 2023; Toda et al., 2020). Altogether, the use of deep learning approaches to characterize phenotypic variability can be referred to as deep phenomics (Ubbens & Stavness, 2017).

In this manuscript, we have developed a computer vision methodology for the dissection of images of full-length hair fibers for phenotyping at a single fiber resolution, known as deep hair phenomics. We used synthetically generated data for computer vision model training, a sliding-window approach for efficient processing, and an object oriented convolutional neural network (CNN) for accurate object segmentation (Supplemental Figure 1). Our findings indicate that our tool has the capacity to detect changes in fur composition that are invisible to the naked eye and identify phenotypes resulting from endocrinological, developmental, and aging-related dysfunction.

Results

Development and validation of a computer vision-based deep hair phenomics pipeline

Hair fiber growth is variably affected by fluctuations in the skin and dermal microenvironment during aging and development (Figure 1A) (Gokce et al., 2022; Grymowicz et al., 2020; Porter, 1971; Williams et al., 2021). Due to the heterogeneity of fibers, understanding how these variables impact hair growth requires large sample collections to capture the full extent of variability (Figure 1B). Current and historical approaches for hair fiber analysis are time and labor intensive due to the need for manual separation of fibers for imaging followed by manual measurement of length and type (Chi et al., 2013, p. 201; Driskell et al., 2009; Schlake, 2005; Takeo et al., 2023; Weger & Schlake, 2005b, 2005a), often limiting the scope of hair-related analyses as well as the reproducibility of the data.

Figure 1: Deep Learning-Based Computer Vision Approach for Hair Fiber Analysis.

Figure 1:

Open in a new tab

(A) Graphical representation of factors that impact hair growth and development. (B) Graphical representation of hair fiber size and diversity. (C) Overview of deep hair phenomics pipeline. (D) Representation of the steps in our deep phenomics pipeline for extracting hair fibers from sample image. (E) Digitally extracted hair fibers retain high-resolution of the original image. (F) Length measurements of P21 murine fur from digitally separated fibers (N = 6 (3M/3F), n = 865) match length measurements gathered by hand (N = 3, n = 494). (G) Set of features that can be quantified from digitally separated fibers.

To resolve these issues, we developed a computer vision methodology capable of identifying and measuring hair fibers from image data (Figure 1C). Our approach provides the unique ability to recognize overlapping fibers, streamlining the sample preparation process (Figure 1D). Due to the substantial computational resources necessary to process large, high-resolution images, we utilized a sliding window approach for the parallel processing of small regions of the image (Figure 1D) (Lee et al., 2017; Viola & Jones, 2001). After fiber segments were detected in each region, hairs were reassembled while retaining high resolution (Figure 1E). We manually and computationally assessed fur samples collected from postnatal day 21 (P21) and found that computationally derived length measurements were similar to manual measurements (Figure 1F). In addition to length metrics, once extracted, the entirety of the fiber can be quantified including its thickness and color (Figure 1G). Overall, our deep phenomics pipeline simplifies the processing of high-quality samples at a single-fiber resolution, facilitating quantitative phenotyping and making hair analysis feasible at scale.

Deep hair phenomics recapitulates a canonical understanding of juvenile mouse fur

The conventional approach to quantifying and assessing mouse fur is to classify them based on four distinct hair types: guard, awl, auchene, and zigzag, which can be identified by number of medulla cell columns and number of bends (Figure 2A) (Sundberg, 1994). To investigate the accuracy of our deep hair phenomics pipeline, we collected independent hair samples from P21 mice and compared our computational approach (‘A.I.’) to the current standard method (‘Human’) (Figure 2B–C). The ‘human’ data was collected by manually counting the number of each hair type in a clump of fibers under a microscope (N = 3, n = 494). The ‘A.I.’ data was collected by manually sorting the digitally extracted fibers into categories of each hair type (N = 6, n = 865). We found that ‘Human’ measurements of fiber length closely matched the ‘A.I.’ measurements for each hair type, supporting the accuracy of this technique (Figure 2B). Additionally, we found no significant difference in hair type proportion between ‘Human’ and ‘A.I.’ analyses, demonstrating the ability of our tool to identify fibers without bias towards or against any particular hair type (Figure 2C). Our measurements showed that on average, zigzag hairs were the thinnest and shortest of the hair types (7.93μm wide, 6.29mm long), while awl hairs were the thickest (12.9μm wide) hair and guard hairs were the longest (8.75mm long), recapitulating the field’s understanding of canonical hair types (Figure 2D) (Duverger & Morasso, 2009; Schlake, 2007; Sundberg, 1994).

Figure 2. Integration of A.I. Quantifications with Manual Hair Typing to Generate Single Hair Phenome (shPhenome).

Figure 2.

Open in a new tab

(A) Examples of the four canonical hair types found in mouse fur: guard (G), awl (AW), auchene (AU), and zigzag (ZZ) hairs. (B) Manual measurements of hair lengths (N = 3, n = 494) sorted by type compared to measurements from computationally extracted hair fibers (N = 6 (3M/3F), n = 865). (C) Hair type proportion distribution from manually counted compared to hair type from manually classified computationally extracted hairs. (D) Matrix plot of normalized computationally derived measurements of Area, Length, Width, Average RGB, RGB Standard Deviation, and Dark:Light Ratio sorted by hair type. (E) Single hair phenome (shPhenome) is created using UMAP dimensional reduction to organize extracted hair fibers based on all metrics captured. Each dot represents the summation of measurements for each individual hair fiber (N = 6 (3M/3F), n = 865). (F) Feature Plots of quantitative metrics (Area, Perimeter (Length), A/P Ratio (Width), RGB Color). (G) Plotting of manually determined hair types. (H) Proportion of each hair type in each Leiden cluster. (I) Visualization of hair fibers organized by size recapitulates the hair continuum first described by F.W. Dry in 1926 using modern techniques (Dry, 1926).

To capture the fiber’s physical properties, our tool quantifies fiber area, perimeter (length), area/perimeter ratio (width), average color (Red (R), Green (G), Blue (B)), color variability, and a ratio of dark:light pixels. We used our pipeline to generate a dataset consisting of 865 digitized fibers from P21 murine fur. Analysis and visualization of the multivariate single hair quantifications was performed using the package, Scanpy (Wolf et al., 2018). We utilized Uniform Manifold Approximation and Projection (UMAP) to plot the individual hair fibers based on their physical features in two dimensions, generating the P21 single hair phenome (shPhenome) (Figure 2E). Visualization of quantitative measurements through feature plots underscores that area and area/perimeter ratio were heavily weighted in dimensional reduction, resulting in plots that follow a gradient of hair size (Figure 2F). This continuous stretch of hairs indicates high levels of fiber heterogeneity, reiterating that in contrast to the canonical view of distinct types, hair fibers in murine skin exist in a continuum as previously published by Dry et al. 1926 (Dry, 1926).

Upon overlaying canonical hair types, the plot revealed sections enriched in each of the hair types (Figure 2G). To better visualize how hair types varied across the continuum, we integrated the manually typed hair data with computationally derived Leiden clusters. The largest clusters in terms of area (1 & 2) consisted primarily of the longest and widest hairs, awl and guard (Figure 2H). This was followed by the medium-sized auchene hairs, which were primarily found in clusters 2 and 3 (Figure 2H). Lastly, a small proportion of long zigzag hairs were found in cluster 3, but were the majority of clusters 4, 5, 6, and 7 (Figure 2H). Zigzag hairs not only saturated most clusters, but also had the highest degree of variability (Figure 2H). The single fiber resolution of the quantifications captures a much greater level of heterogeneity within fibers, allowing for appreciation of the full extent of variability in hair’s physical features. Notably, identified phenotypes can be probed visually alongside quantifications (Figure 2I). Deep hair phenomics empowers the field to explore known hair phenotypes in greater detail and paves the way for novel discoveries in hair biology.

Disruption of the adolescent hormonal cascade amplifies hair type proportion shift during murine maturation

During periods of maturation, animals experience dramatic shifts in physical appearance, such as the molting of feathers or growth of terminal/androgenic hair (Sperling, 1991; Yu et al., 2004). Previous studies have demonstrated that successive hair cycles produce different hair types (Chi et al., 2013), however, the mechanisms behind this phenomenon have not been elucidated. To confirm our model’s ability to detect fluctuations in the proportion of hair types, we collected dorsal fur to generate the P21 and P57 shPhenomes consisting of 865 and 545 hairs, respectively. Within the UMAP itself, it is apparent that there is a reduction in the smaller, thinner hairs and a gain of longer, thicker hairs (Figure 3A). Between these time points, there is nearly a 15% loss in the proportion of zigzag hairs matched by an increase in the thicker hair types, specifically in the awl and auchene hairs (Figure 3B). The average fiber of each hair type also increases in average length and width between these timepoints (Figure 3C). Altogether our findings are in agreement with the hair shifts found in previous studies (Chi et al., 2013).

Figure 3. shPhenomics reveals a correlation between the murine hair type shift and the adolescent hormonal cascade.

Figure 3.

Open in a new tab

(A) P21 (N = 6 (3M/3F), n = 865) and P57 (N = 6 (3M/3F), n = 545) hair sample collections generate shPhenomes for each timepoint. (B) Hair type proportions between P21 and P57. (C) Matrix plot of normalized hair fiber quantifications across conditions. (D) Schematic of gonadectomy surgery at P25 with hair sample collection at P60. (E) Change in fur appearance after gonadectomy. (F) Feature plots and shPhenomes of mice that underwent sham, gonadectomy, and ovariectomy surgeries (N = 12, n = 1155). (G) shPhenomes of Male GDX (N = 3, n = 263) and Male Sham (N = 3, n = 200) with corresponding hair type proportions and quantifications (statistics in Supplemental Figure 2). (H) shPhenomes of Female GDX (N = 3, n = 342) and Female Sham (N = 3, n = 250) with corresponding hair type proportions and quantifications (statistics in Supplemental Figure 2).

The hair type shift coincides with the time in which mice are undergoing sexual maturation, suggesting its potential correlation with the adolescent hormonal cascade. To test whether this hair type shift is associated with hormonal changes during this time point, we performed gonadectomy surgeries in both male and female mice at P25 (Figure 3D). This surgery removes the reproductive organs, effectively disrupting the onset of puberty and sexual maturity (Iakovleva et al., 2020; Karlsson et al., 2015; Kercmar et al., 2014). Five weeks later, at P60, we found that the fur coat of gonadectomized mice appeared visually distinct from their sham counterparts with reduced luster, increased frizziness, and occasionally premature graying (Figure 3E).

Overall, we analyzed 1155 hair fibers across the four conditions: Male Gonadectomy (GDX), Female Ovariectomy (OVX), Male Sham, and Female Sham (Figure 3F). After both male gonadectomy and female ovariectomy, the overall fur composition increased in average fiber length and width. Our analysis shows that the male gonadectomy dramatically increases the proportion of the longer hair types from 35% to 48% (Figure 3G). After female ovariectomy, there is a slight reduction in the proportion of zigzag hairs with small increases in the proportion of guard, awl, and auchene hairs (Figure 3H). As our tool captures the composition of the fur present at time of collection, it is not clear whether these shifts are due to an amplification of hair type switching from zigzag to awl and auchene hairs or rather the shedding of primarily zigzag hair fibers. Future studies may elucidate the mechanisms that drive the shift in hair type proportions resultant of gonadectomy and ovariectomy. These results indicate that the ablation of the adolescent hormonal cascade does not obstruct the switch in hair follicle type proportions from P21 to P57, but instead amplifies it.

Ablation of dermal Lef1 expression during development delays hair follicle cycling and augments hair type determination

The growth of the first fur coat in murine skin is completed at P21, which establishes the first phase of telogen (Müller-Röver et al., 2001). Importantly, multiple genetic mutations that result in the misregulation of functional proteins within the WNT, BMP, TGF, SHH, Notch signaling pathways have been shown to influence the regulation of hair follicle morphogenesis and cycling (Beaudoin et al., 2005; Hibino & Nishiyama, 2004, 2004; Kulessa et al., 2000; Merrill et al., 2001; O. Mori et al., 1996; Myung et al., 2013; Phan et al., 2020; Powell et al., 1998; Tsuji et al., 2003; Zhou et al., 1995). Ablation of Lef1 in the dermis using Twist2-Cre (Lef1-cKO) (Figure 4A) (Phan et al., 2020) perturbs hair follicle cycling resulting in the fur coat visibly changing to be ‘subtly wiry’ at P21 which is then reversed by P57 (Figure 4B). Immunohistochemical and histological analyses revealed that the hair follicle cycle is delayed, as Lef1-cKO skin is late to catagen and telogen (Figure 4C–D). In addition, we observed that during neonatal development, Lef1-cKO skin has a delay in visible pigmentation further indicating a delay in hair cycle initiation and/or progression (Supplemental Figure 3A). As a control we found that immunohistochemical staining for Lef1 shows a reduction in its expression within the dermal papilla during development and maturation (Figure 4C).

Figure 4. shPhenomics detect shifts in hair cycle and hair type determination in dermal Lef1cKO.

Figure 4.

Open in a new tab

(A) Transgenic construct for knockout of dermal Lef1 expression. (B) Lef1cKO mice fur coat phenotype at P21. (C) Dermal Lef1cKO reduces Lef1 expression in the dermal papilla at P2, P21, and P57 (scale bar = 50um). (D) Effect of Lef1 cKO on hair follicle cycling during development. (E) P21 WT (N = 6 (3M/3F), n = 865), P21 cKO (N = 6 (3M/3F), n = 1156), P57 WT (N = 6 (3M/3F), n = 545), and P57 cKO (N = 6 (3M/3F), n = 970) hair samples were collected for generating a shPhenome. (F) Feature plots of area and area/perimeter ratio. (G) Matrix plot of normalized hair fiber quantifications across conditions. (H) shPhenome depicting distribution of hair types. (I) Effect of Lef1 cKO on hair type proportions (statistics in Supplemental Figure 3B–D). (J) Shifts in fiber clustering that result from Lef1 cKO.

To quantify and examine the impact of the hair cycle delay on the shPhenome of Lef1-cKO mouse fur, we collected a total of 3536 hairs fibers from P21/P57 WT and P21/P57 cKO mice (Figure 4E–F). Aggregate quantifications of each timepoint, which can be viewed in a heatmap, revealed that Lef1-cKO mouse fur has increased average hair fiber width (9.37μm versus 9.08μm in WT) and reduced average length (6.63mm versus 6.78mm in WT) (Figure 4G). At P57 there is a reduction in both average hair length and width (10.9μm versus 11.6μm wide and 7.42mm versus 8.16mm long) (Figure 4G). When the hair follicles were individually typed and plotted at P21 (Figure 4H–J), there is a significant reduction in the proportion of awl and auchene hairs, matched with an increase in zigzag hairs, but guard hairs remained unaffected (Figure 4I). By P57, the hair type switch aids in the recovery of the deficit of longer hair types, but interestingly has a significantly greater proportion of auchene hair fibers indicating that in the absence of dermal Lef1, hair follicle type programming can still be modulated, but not perfectly recovered, by the adolescent maturation. These results demonstrate the role of dermal Lef1 in murine hair follicle development, growth, and maintenance.

The fur coat dynamically changes throughout the murine lifespan.

Due to the strong correlation between the hair’s appearance and aging-related decline in function, “coat condition” has been utilized as an important measure for biological aging (Yang et al., 2023). Current methods for measuring the coat’s condition rely on visual scoring of loss of hair and pigmentation but lack the capacity to detect subtle phenotypic shifts (Whitehead et al., 2014). Murine dorsal fur has some variability in appearance throughout aging, but it is difficult to quantify why the coats are visually distinct (Figure 5A). We collected and measured a total of 4342 hair fiber samples from wild type mice from ages P21, P57, 3 months old, 6 months old, 1 year old, and 1.5 years old (Figure 5B). Aggregate quantifications revealed that the average hair fiber increases in length (6.78mm at P21 to 8.94mm at 1.5 year old) and width (9.1μm at P21 to 12.8μm at 1.5 year old) throughout aging (Figure 5C). As 1.5 year old mice are still relatively healthy in appearance and behavior, our results may not fully reflect the effects of chronic aging (Pettan-Brewer & Treuting, 2011; Sundberg et al., 2011). Nonetheless, our data reveals that as mice age, their hair fibers from all types (guard/awl/auchene/zigzag) become thicker and longer.

Figure 5. shPhenomics identifies aging phenotypes across the murine lifespan.

Figure 5.

Open in a new tab

(A) Mouse fur coats are subtly distinct across aging. (B) Feature plot of fiber area. (C) Normalized aggregate quantifications across aging. (D) Leiden cluster analysis ordered by increasing hair size. (E) Hair samples from P21 (N = 6 (3M/3F), n = 865), P57 (N = 6 (3M/3F), n = 545), 3 month (N = 6 (3M/3F), n = 727), 6 month (N = 6 (3M/3F), n = 768), 1 year (N = 6 (3M/3F), n = 746), and 1.5 year (N = 6 (3M/3F), n = 690) old mice were collected to generate an aging shPhenome. (F) Hair type proportions across the murine lifespan. (G) Leiden cluster proportions across the murine lifespan. (H) Parallel plotting of juvenile, mature, and aged timepoints. (I) Proportions of hairs from isolated timepoints across the Leiden clusters. (J) Quantification and visualization of each hair type across aging.

We investigated how hair type proportions change after P57 and found that by 1 year of age they gradually reverted to a distribution similar to that of P21 fur (Figure 5F). However, by 18 months of age, there is a significant increase in the proportion of thicker hair types (Figure 5F). Parallel plotting of juvenile, mature, and aged timepoints showed that the coat composition shifts from primarily being composed of short zigzag hairs to long awl/auchene hairs (Figure 5H–I). The reduction of zigzag hair type suggests an aging-related bias towards the thicker hair types, particularly awl and auchene (Figure 5H–I). We then grouped the hairs and found that each hair type gradually increases fiber length and width (Figure 5J). Interestingly, there were slight differences in the magnitude of shifts in hair type proportions between sexes. Typically, shifts were more dramatic in male mice relative to female mice (Supplemental Figure 4). Color measurements were gathered, however, our cohort consisted of mice from mixed C3H/HeJ (agouti) and C57BL/6J (black) background limiting our analysis of shifts in pigmentation throughout aging. Additionally, our method only measures fur collected from a singular region (lower dorsum) of the mouse which may not capture the entirety of phenotypic effects on spatially distinct regions of fur nor the impacts of the unsynchronized hair cycle on fiber measurements. Quantification of shifts in hair fiber physical features, outside of type alone, further corroborates the advantage of our technique and its potential to detect subtle phenotypes throughout the murine lifespan.

Discussion

The fields of dermatology and trichology have long been unable to freely explore hair phenotypes due to the time and labor constraints of ‘splitting hairs’. Our deep phenomics tool offers the unique ability to rapidly extract information from digital images of hair at a single fiber resolution. Furthermore, our AI approach can be easily adapted to new imaging methods, modifications in sample preparation, and even recognition of other fiber-like objects. As a result of the high throughput nature of our workflow, we can better understand how hair fiber heterogeneity is altered under various conditions.

Modernizing Hair Type Classification to Capture Fiber Heterogeneity

Currently, the field recognizes four distinct hair types: guard, awl, auchene, and zigzag hairs (Sundberg, 1994). While these classifications are effective for distinguishing fibers based on macroscopic differences, they do not capture microscopic differences such as thickness or color variability. A landmark study by Francis William (F.W.) Dry first described the hair “continuum” that not only varied in hair type, but also in physical properties (Dry, 1926). For instance, shifts in the hair fiber length indicate altered regulation of fiber growth initiation, rate, and termination. Similarly, notable shifts in hair color indicate variable pigment deposition by follicular melanocytes. However, the characterization of the processes that govern these hair fiber features remains obscure as a result of the time and labor-intensive nature of analysis. Due to these limitations, most recorded phenotypes in the literature were only explored when visibly apparent, potentially overlooking subtle phenotypes. Our data corroborates long standing knowledge regarding hair fiber types and rekindles a century old idea via modern techniques.

The ability to extract individual hair fibers alongside their quantitative metrics paves the way for classifying hairs into additional subcategories that encapsulate the entirety of their diversity. Additional methods of quantification, such as counting of medulla columns, may further improve our ability to distinguish hair fiber subtypes. For instance, the class of awl hairs describes fibers that are within the range of two to four columns of medulla cells but do not distinguish between those with variable traits (Supplemental Figure 5A–B). In addition to the heterogeneity of hair fibers of the same class, we have also noted the identification of hairs that share traits with multiple of the canonical classes, such as zigzag hairs without bends (Supplemental Figure 5C). Further exploration of these phenotypes may uncover novel mechanisms involved in hair fiber growth and type differentiation.

Implications of Deep Hair Phenomics in Human & Veterinary Medicine

The fully functional hair follicle requires the coordination of a menagerie of distinct tissue types and cellular lineages including keratinocytes, fibroblasts, adipocytes, nerves, and vasculature (Blanpain & Fuchs, 2006; Morris et al., 2004; Ohyama et al., 2006; Schmidt & Horsley, 2012; Trempus et al., 2003; Winkelmann, 1959; Xiao et al., 2013). Consequently, the hair fiber provides a window into the conditions of an organism’s health, including stress levels, cardiovascular health, and metabolic status (Lin et al., 2016; Raul et al., 2004). In the context of hormone biology, we have shown that there is a characteristic shift in the fur composition of mice after inhibition of the adolescent hormonal cascade. Similarly, through genetic ablation of the expression of dermal Lef1, we have identified measurable shifts in hair follicle development and fiber production. The entirety of the mechanisms that drive these shifts in hair growth and type determination is currently unknown. It has been previously demonstrated that changes in hair matrix keratinocyte function result in changes to hair fiber physical properties (Chen et al., 2008; Cui et al., 2003; Danilenko et al., 1995; Sharov et al., 2006; Sotiropoulou et al., 2013; Zhang et al., 2021). Additionally, mesenchymal interactions between the dermal papillae and hair matrix keratinocytes can affect their function and hair fiber output (Botchkarev & Kishimoto, 2003; Clavel et al., 2012; Driskell et al., 2011; Millar, 2002; Morgan, 2014; Rendl et al., 2005; Sennett & Rendl, 2012). Various signaling pathways, including Wnt, Shh, Eda, Bmp, and Igf, not only influence the hair cycle and rate of fiber growth, but also play significant roles in the determination of hair follicle type (Botchkarev et al., 2002; Cui et al., 2010; Ellis et al., 2003; Weger & Schlake, 2005; Zhou et al., 1995). Future studies may further reveal the causal cell types and signaling cascades that drive shifts in rate of fiber growth, timing of hair cycle stages, and determination of hair follicle type.

While we have specifically highlighted the effects of endocrinology, development, and aging on measurable changes to the hair, there are a multitude of conditions, from lupus to Alzheimer’s dementia, with underexplored hair phenotypes. Additionally, hair sample collection is non-invasive, easy to perform, and storable under variable conditions, making it an ideal tissue source for diagnostic purposes. As fibers are retained over long periods of time, they provide the unique ability to perform retrospective analyses of conditions over weeks to months. By continuing to improve our understanding of how hair fiber growth is altered under non-homeostatic conditions, fiber analysis may be an effective future diagnostic method for human and veterinary medicine.

Methods

Mice

WT mice for aging experiments were derived from a C57BL6/CBA background. We generated dermal fibroblast specific Twist2-Cre Lef1-cKO mice as previously described (Phan et al., 2020). Mice for OVX/GDX experiments were derived from a WT C57BL6 background. All experiments were performed in accordance with approved Washington State University (WSU) Institutional Animal Care and Use Committee (IACUC) protocols and conformed to principles outlined by the NIH Guide for the Care and Use of Laboratory Animals.

Gonadectomy Surgical Procedure

Mice were gonadectomized at P25 and fur samples were collected at P60. The ovaries or testes were removed as previously described in Klappenbach, 2023 (Klappenbach et al., 2023). Sham surgeries were performed identically to gonadectomies except that the gonads were not excised.

Histological Analysis

Tissue samples were fixed in 4% PFA overnight and paraffin embedded. Samples were sectioned at 10 μm and stained using standard hematoxylin and eosin (H&E) staining protocols. Slides were imaged using a Nikon DS-Fi3 color camera through the Nikon Scope software.

Immunofluorescence Analysis

Tissue samples were fixed in 4% PFA for 30 minutes, embedded in Tissue Plus O.C.T Compound, and stored at −80°C. The staining and sectioning procedures were described in Salz and Driskell, 2017 (Salz & Driskell, 2017). Tissue was stained with the following primary antibodies: human-ITGA6 rat (1:200, BD Biosciences) and human LEF1 rabbit (1:200, Cell Signaling). Secondary antibodies used were AF555 anti-rat (1:1000, Fisher), AF647 anti-rabbit (1:1000, Fisher). Slides were imaged using a Leica SP8 confocal microscope through the LAS X software.

Manual Hair Measurements

All manually measured hairs were traced by hand and then recorded through FIJI’s built-in measurement function (Schindelin et al., 2012; Schneider et al., 2012). Hair types of both manually viewed and digitally extracted hairs were recorded based on previously described characteristics of the hair types: zigzag hairs were identified as having a single medulla column and multiple kinks, auchene hairs were identified as having two to four medulla columns and a single bend, awl hairs were identified as having three to five medulla columns, and guard hairs were identified as having two medulla columns (Duverger & Morasso, 2009). Hair type classification of digitally extracted hairs was performed with randomized order of hair fibers and blind to genotype, treatment, and quantifications. Hair cycle staging was performed through histological analysis in accordance with previously described characteristics (Müller-Röver et al., 2001). For statistical analyses, Chi-squared was used to first evaluate if total proportions between conditions were significantly different and two-sample z-test for proportions was used to determine which particular hair types significantly differed.

Hair Sample Preparation

Fur was collected from the lower dorsum of the mouse at each of the indicated timepoints. For each biological replicate, we prepared 5 microscope slides with approximately 20-40 hairs per slide. Each slide was pre-prepared with 5μm tissue sections and lightly stained with eosin to aid in the automated focusing of the objective when imaging.

Hair Sample Imaging

All images were taken with the Aperio GT450 owned and operated by the Washington Animal Disease Diagnostic Laboratory (WADDL). All samples were imaged with a 40x objective and saved as an Aperio Scanscope Virtual Slide (SVS) file. Conversion to PNG was performed using the application, reaConverter 7 Pro.

Model Training and Accuracy

To reduce the computational load of image processing, we have implemented a sliding window approach in which the image is broken up into smaller overlapping “windows” or tiles. Each of these tiles is processed individually to identify regions of hair fibers in each segment of the image. Once processed, the spatial information for each detected hair is saved. Initially, each individual tile is scanned for potential duplicates or fragmented masks to reduce the computational load of stitching. A stitching algorithm is then used to identify the neighboring tiles where the fiber continues. Areas of sufficient overlap were then merged to generate an image of a full-length hair.

To train our model, we generated 100,000 images of 1024x1024 images of hair fibers in various orientations. To generate images, we annotated 100 unique hair fibers with various orientations and curvature. These hair fibers were then randomly rotated and resized. A random 1024x1024 cutout of a random hair was then placed on one of 50 random background images. Each image had a random number (up to 15) of hairs placed in the image. Afterwards, a variable blur and/or color filter was added to the image to mimic variable focus of the camera or imaging artifacts. Together, these image augmentations permitted synthetic generation of unique images of hair fibers that closely resemble real images and were extremely diverse. As previously mentioned, we utilized the OBBDetection package to recognize the hair fibers. This library is typically used in conjunction with images annotated in the Dataset for Object deTection in Aerial Images (DOTA) format to support oriented bounding boxes. We have adapted this to integrate with datasets annotated in the COCO style annotation format. Mask R-CNN was used as the general framework for object instance segmentation (He et al., 2018). The resnet152 framework was used as the backbone network resulting in highly accurate feature extraction from the images (He et al., 2015). This model was trained for 1000 epochs, achieving an accuracy metric of 99.7%. To further test the efficacy of the model, we visualized the predicted masks on sample images of varying complexity (Supplemental Figure 6). After training, the model was able to recognize and extract both isolated and overlapping hair fibers. Regions of the hair that are out of focus when imaged typically leads to fragmentation of the fiber which may be resolved by future improvements in model training and optimization of imaging protocols.

All model training and synthetic data generation was performed on the WSU high-performance computing cluster, Kamiak. Model training was performed on a NVIDIA A100 GPU. Inference was performed in parallel on NVIDIA K80 GPUs. CUDA version 10.5.105 and CUDNN version 7.5.1 were utilized for performing computations through a GPU. OBBDetection was installed according to https://github.com/jbwang1997/OBBDetection. Modifications were made for training on images of size 1024x1024 with COCO format annotations. The residual network ResNet152 was used for feature extraction. Warm-up was performed at a learning rate of 0.01 for 500 iterations. Training was performed for 1000 epochs by stochastic gradient descent (SGD) optimization with a learning rate of 0.005.

Hair Fiber Quantification

Our tool quantifies fiber area, perimeter (length), area/perimeter ratio (width), average color (Red, Green, Blue), color variability, and a ratio of dark:light pixels. Fiber area and perimeter were determined by OpenCV’s contourArea and arcLength functions. We utilized the perimeter value as an approximation of hair length as fibers were hundreds of times longer than they were wide (Supplemental Figure 7A–B). To calculate an approximate width value, we utilized the area/perimeter ratio of the fiber (Supplemental Figure 7C). RGB color values were determined by the average and standard deviation of the color value of each pixel within the hair fiber mask. Average RGB values were calculated without exclusion of the overlapping regions as they do not significantly impact average color values and increase time of computation (Supplemental Figure 8). Dark/light ratio is determined by binarizing a grayscale image of the hair fiber where any pixel value greater than 200 is considered ‘light’ and any value less than 200 is ‘dark’.

Digital Hair Fiber Analysis

To analyze this multivariate data, we utilized the single-cell RNA sequencing analysis package, Scanpy (Wolf et al., 2018), for ease of integration with dimensional reduction and clustering algorithms. Filtering was performed to remove fragments and debris picked up by the algorithm.

Figure Generation

Digital assets were adapted from Biorender.com.

Supplementary Material

1

Acknowledgements

All confocal microscopy was performed on equipment housed and maintained by the Franceschi Microscopy and Imaging Center at WSU. Large datasets were processed with Washington State University High Performance Computing Cluster (HPCC), Kamiak.

Funding Sources

RRD was supported by NIH R01 AR078743. SMT and QP were supported by the Poncin Fellowship. QMP and JM were supported by NIH T32GM008336.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Conflict of Interest Statement

The authors have submitted and filed a provisional patent protecting the computer vision technology used in this manuscript.

Data Availability Statement

No new large genomic datasets were created in this manuscript. The code implemented can be found at: https://github.com/DriskellLab/Makkar-et-al.-2024.

References

  1. Agramunt J, Parke B, Mena S, Ubels V, Jimenez F, Williams G, Rhodes AD, Limbu S, Hexter M, Knight L, Hashemi P, Kozlov AS, & Higgins CA (2023). Mechanical stimulation of human hair follicle outer root sheath cultures activates adjacent sensory neurons. Science Advances, 9(43), eadh3273. 10.1126/sciadv.adh3273 [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Almashali M, Alekrish K, & Shadid A (2023). The Use of 2D:4D Digit Ratio as a Predictor of Androgenetic Alopecia: A Review. Dermatology Practical & Conceptual, 13(4), e2023237. 10.5826/dpc.1304a237 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Archila J, Manzanera A, & Martínez F (2022). A multimodal Parkinson quantification by fusing eye and gait motion patterns, using covariance descriptors, from non-invasive computer vision. Computer Methods and Programs in Biomedicine, 215, 106607. 10.1016/j.cmpb.2021.106607 [DOI] [PubMed] [Google Scholar]
  4. Arteta C, Lempitsky V, Noble JA, & Zisserman A (2013). Learning to Detect Partially Overlapping Instances. 3230–3237. 10.1109/CVPR.2013.415 [DOI] [Google Scholar]
  5. Bae H, & An Y-K (2023). Computer vision-based statistical crack quantification for concrete structures.Measurement, 211, 112632. 10.1016/j.measurement.2023.112632 [DOI] [Google Scholar]
  6. Beaudoin GMJ, Sisk JM, Coulombe PA, & Thompson CC (2005). Hairless triggers reactivation of hair growth by promoting Wnt signaling. Proceedings of the National Academy of Sciences, 102(41), 14653–14658. 10.1073/pnas.0507609102 [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Blanpain C, & Fuchs E (2006). Epidermal Stem Cells of the Skin. Annual Review of Cell and Developmental Biology, 22, 339–373. 10.l146/annurev.cellbio.22010305,104357 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Botchkarev VA, Botchkareva NV, Sharov AA, Funa K, Huber O, & Gilchrest BA (2002). Modulation of BMP signaling by noggin is required for induction of the secondary (nontylotrich) hair follicles. The Journal of Investigative Dermatology, 118(1), 3–10. 10.1046/j.1523-1747.2002.01645.x [DOI] [PubMed] [Google Scholar]
  9. Botchkarev VA, & Kishimoto J (2003). Molecular Control of Epithelial–Mesenchymal Interactions During Hair Follicle Cycling. Journal of Investigative Dermatology Symposium Proceedings, 8(1), 46–55. 10.1046/j.1523-1747.2003.12171.x [DOI] [PubMed] [Google Scholar]
  10. Chai J, Zeng H, Li A, & Ngai EWT (2021). Deep learning in computer vision: A critical review of emerging techniques and application scenarios. Machine Learning with Applications, 6, 100134. 10.1016/j.mlwa.2021.100134 [DOI] [Google Scholar]
  11. Chen J, Jaeger K, Den Z, Koch PJ, Sundberg JP, & Roop DR (2008). Mice expressing a mutant Krt75 (K6hf) allele develop hair and nail defects resembling pachyonychia congenita. The Journal of Investigative Dermatology, 128(2), 270–279. 10.1038/sj.jid.5701038 [DOI] [PubMed] [Google Scholar]
  12. Chi W, Wu E, & Morgan BA (2013). Dermal papilla cell number specifies hair size, shape and cycling and its reduction causes follicular decline. Development (Cambridge, England), 140(8), 1676–1683. 10.1242/dev.090662 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Clavel C, Grisanti L, Zemla R, Rezza A, Barros R, Sennett R, Mazloom A, Chung C-Y, Cai X, Cai C-L, Pevny L,Nicolis S, Ma’ayan A, & Rendl M. (2012). Sox2 in the dermal papilla niche controls hair growth by fine-tuning Bmp signaling in differentiating hair shaft progenitors. Developmental Cell, 23(5), 981–994. 10.1016/j.devcel.2012.10.013 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Cui C-Y, Durmowicz M, Ottolenghi C, Hashimoto T, Griggs B, Srivastava AK, & Schlessinger D (2003). Inducible mEDA-A1 transgene mediates sebaceous gland hyperplasia and differential formation of two types of mouse hair follicles. Human Molecular Genetics, 12(22), 2931–2940. 10.1093/hmg/ddg325 [DOI] [PubMed] [Google Scholar]
  15. Cui C-Y, Kunisada M, Piao Y, Childress V, Ko MSH, & Schlessinger D. (2010). Dkk4 and Eda Regulate Distinctive Developmental Mechanisms for Subtypes of Mouse Hair. PLOS ONE, 5(4), e10009. 10.1371/journal.pone.0010009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Danilenko DM, Ring BD, Yanagihara D, Benson W, Wiemann B, Starnes CO, & Pierce GF (1995). Keratinocyte growth factor is an important endogenous mediator of hair follicle growth, development, and differentiation. Normalization of the nu/nu follicular differentiation defect and amelioration of chemotherapy-induced alopecia. The American Journal of Pathology, 147(1), 145–154. [PMC free article] [PubMed] [Google Scholar]
  17. Di Fraia M, Tieghi L, Magri F, Caro G, Michelini S, Pellacani G, & Rossi A (2023). A Machine Learning Algorithm Applied to Trichoscopy for Androgenic Alopecia Staging and Severity Assessment. Dermatology Practical & Conceptual, 13(3), e2023136. 10.5826/dpc.1303al36 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Driskell RR, Giangreco A, Jensen KB, Mulder KW, & Watt FM (2009). Sox2-positive dermal papilla cells specify hair follicle type in mammalian epidermis. Development (Cambridge, England), 136(16), 2815–2823. 10.1242/dev.038620 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Driskell RR, Clavel C, Rendl M, & Watt FM (2011). Hair follicle dermal papilla cells at a glance. Journal of Cell Science, 124(8), 1179–1182. 10.1242/jcs.082446 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Dry FW (1926). The coat of the mouse (Mus musculus). Journal of Genetics, 16(3), 287–340. 10.1007/BF02983004 [DOI] [Google Scholar]
  21. Duverger O, & Morasso MI (2009). Epidermal Patterning and Induction of Different Hair Types during Mouse Embryonic Development. Birth Defects Research. Part C, Embryo Today : Reviews, 87(3), 263–272. 10.1002/bdrc.20158 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Ellis T, Smyth I, Riley E, Bowles J, Adolphe C, Rothnagel JA, Wicking C, & Wainwright BJ (2003). Overexpression of Sonic Hedgehog suppresses embryonic hair follicle morphogenesis. Developmental Biology, 263(2), 203–215. 10.1016/S0012-1606(03)00394-4 [DOI] [PubMed] [Google Scholar]
  23. Esteva A, Chou K, Yeung S, Naik N, Madani A, Mottaghi A, Liu Y, Topol E, Dean J, & Socher R (2021). Deep learning-enabled medical computer vision. Npj Digital Medicine, 4(1), 1–9. 10.1038/s41746-020-00376-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. GAO M, WANG Y, XU H, XU C, YANG X, NIE J, ZHANG Z, LI Z, HOU W, & JIANG Y. (2022). Deep Learning-based Trichoscopic Image Analysis and Quantitative Model for Predicting Basic and Specific Classification in Male Androgenetic Alopecia. Acta Dermato-Venereologica, 102, 564. 10.2340/actadv.vl01.564 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Gokce N, Basgoz N, Kenanoglu S, Akalin H, Ozkul Y, Ergoren MC, Beccari T, Bertelli M, & Dundar M (2022). An overview of the genetic aspects of hair loss and its connection with nutrition. Journal of Preventive Medicine and Hygiene, 63(2 Suppl 3), E228–E238. 10.15167/2421-4248/jpmh2022.63.2S3.2765 [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Grymowicz M, Rudnicka E, Podfigurna A, Napierala P, Smolarczyk R, Smolarczyk K, & Meczekalski B (2020). Hormonal Effects on Hair Follicles. International Journal of Molecular Sciences, 21(15), 5342. 10.3390/ijms21155342 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Hamann W. (1995). Mammalian cutaneous mechanoreceptors. Progress in Biophysics and Molecular Biology, 64(1), 81–104. 10.1016/0079-6107(95)00011-9 [DOI] [PubMed] [Google Scholar]
  28. Hasan R, Juma H, Eid FA, Alaswad HA, Ali WM, & Aladraj FJ (2022). Effects of Hormones and Endocrine Disorders on Hair Growth. Cureus, 14(12), e32726. 10.7759/cureus.32726 [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. He K, Gkioxari G, Dollár P, & Girshick R (2018). Mask R-CNN (arXiv:1703.06870). arXiv. 10.48550/arXiv.1703.06870 [DOI] [PubMed]
  30. He K, Zhang X, Ren S, & Sun J (2015). Deep Residual Learning for Image Recognition (arXiv:1512.03385). arXiv. 10.48550/arXiv.1512.03385 [DOI]
  31. Hibino T, & Nishiyama T (2004). Role of TGF-beta2 in the human hair cycle. Journal of Dermatological Science, 35(1), 9–18. 10.1016/jjdermsci.2003.12.003 [DOI] [PubMed] [Google Scholar]
  32. Ho CH, Sood T, & Zito PM (2024). Androgenetic Alopecia. In StatPearls. StatPearls Publishing. http://www.ncbi.nlm.nih.gov/books/NBK430924/ [PubMed] [Google Scholar]
  33. Iakovleva TV, Kostina NE, Makarova EN, & Bazhan NM (2020). Effect of gonadectomy and estradiol on the expression of insulin signaling cascade genes in female and male mice. Vavilov Journal of Genetics and Breeding, 24(4), 427. 10.18699/VJ20.635 [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Jeong HK, Park C, Henao R, & Kheterpal M (2023). Deep Learning in Dermatology: A Systematic Review of Current Approaches, Outcomes, and Limitations. JID Innovations, 3(1). 10.1016/j.xjidi.2022.100150 [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Karlsson SA, Haziri K, Hansson E, Kettunen P, & Westberg L (2015). Effects of sex and gonadectomy on social investigation and social recognition in mice. BMC Neuroscience, 16(1), 83. 10.1186/sl2868-015-0221-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Kercmar J, Snoj T, Tobet SA, & Majdic G (2014). Gonadectomy prior to puberty decreases normal parental behavior in adult mice. Hormones and Behavior, 66(4), 667–673. 10.1016/j.yhbeh.2014.09.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Kim J, Oh I, Lee YN, Lee JH, Lee YI, Kim J, & Lee JH (2023). Predicting the severity of postoperative scars using artificial intelligence based on images and clinical data. Scientific Reports, 13, 13448. 10.1038/s41598-023-40395-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Klappenbach CM, Wang Q, Jensen AL, Glodosky NC, & Delevich K (2023). Sex and timing of gonadectomy relative to puberty interact to influence weight, body composition, and feeding behaviors in mice. Hormones and Behavior, 151, 105350. 10.1016/j.yhbeh.2023.105350 [DOI] [PubMed] [Google Scholar]
  39. Kulessa H, Turk G, & Hogan BL (2000). Inhibition of Bmp signaling affects growth and differentiation in the anagen hair follicle. The EMBO Journal, 19(24), 6664–6674. 10.1093/emboj/19.24.6664 [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Lasisi T, Smallcombe JW, Kenney WL, Shriver MD, Zydney B, Jablonski NG, & Havenith G (2023). Human scalp hair as a thermoregulatory adaptation. Proceedings of the National Academy of Sciences of the United States of America, 120(24), e2301760120. 10.1073/pnas.2301760120 [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Lee J, Bang J, & Yang S-I (2017). Object detection with sliding window in images including multiple similar objects. 2017 International Conference on Information and Communication Technology Convergence (ICTC), 803–806. 10.1109/ICTC.2017.8190786 [DOI] [Google Scholar]
  42. Li W-Q, Cho E, Han J, Weinstock MA, & Qureshi AA (2016). Male pattern baldness and risk of incident skin cancer in a cohort of men. International Journal of Cancer, 139(12), 2671–2678. 10.1002/ijc.30395 [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Lin RL, Garibyan L, Kimball AB, & Drake LA (2016). Systemic causes of hair loss. Annals of Medicine, 48(6), 393–402. 10.1080/07853890.2016.1180426 [DOI] [PubMed] [Google Scholar]
  44. Liu F, Hamer MA, Heilmann S, Herold C, Moebus S, Hofman A, Uitterlinden AG, Nöthen MM, van Duijn CM, Nijsten TE, & Kayser M (2016). Prediction of male-pattern baldness from genotypes. European Journal of Human Genetics, 24(6), 895–902. 10.1038/ejhg.2015.220 [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Mendelsohn AR, & Larrick JW (2020). The Danger of Being Too Sympathetic: Norepinephrine in Alzheimer’s Disease and Graying of Hair. Rejuvenation Research, 23(1), 68–72. 10.1089/rej.2020.2309 [DOI] [PubMed] [Google Scholar]
  46. Merrill BJ, Gat U, DasGupta R, & Fuchs E (2001). Tcf3 and Lef1 regulate lineage differentiation of multipotent stem cells in skin. Genes & Development, 15(13), 1688–1705. 10.1101/gad.891401 [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Millar SE (2002). Molecular Mechanisms Regulating Hair Follicle Development. Journal of Investigative Dermatology, 118(2), 216–225. 10.1046/j.0022-202x.2001.01670.x [DOI] [PubMed] [Google Scholar]
  48. Morgan BA (2014). The Dermal Papilla: An Instructive Niche for Epithelial Stem and Progenitor Cells in Development and Regeneration of the Hair Follicle. Cold Spring Harbor Perspectives in Medicine, 4(7), a015180. 10.1101/cshperspect.a015180 [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Mori O, Hachisuka H, & Sasai Y (1996). Effects of transforming growth factor beta 1 in the hair cycle. The Journal of Dermatology, 23(2), 89–94. 10.Ill1/j.1346-8138.1996.tb03976.x [DOI] [PubMed] [Google Scholar]
  50. Mori S, Tachibana Y, Suzuki M, & Harada Y (2022). Automatic worm detection to solve overlapping problems using a convolutional neural network. Scientific Reports, 12(1), 8521. 10.1038/s41598-022-12576-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Morris RJ, Liu Y, Marles L, Yang Z, Trempus C, Li S, Lin JS, Sawicki JA, & Cotsarelis G (2004). Capturing and profiling adult hair follicle stem cells. Nature Biotechnology, 22(4), 411–417. 10.1038/nbt950 [DOI] [PubMed] [Google Scholar]
  52. Müller-Röver S, Foitzik K, Paus R, Handjiski B, van der Veen C, Eichmüller S, McKay IA, & Stenn KS (2001). A Comprehensive Guide for the Accurate Classification of Murine Hair Follicles in Distinct Hair Cycle Stages. Journal of Investigative Dermatology, 117(1), 3–15. 10.1046/j.0022-202x.2001.01377.x [DOI] [PubMed] [Google Scholar]
  53. Myung PS, Takeo M, Ito M, & Atit RP (2013). Epithelial Wnt Ligand Secretion Is Required for Adult Hair Follicle Growth and Regeneration. Journal of Investigative Dermatology, 133(1), 31–41. 10.1038/jid.2012.230 [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Neau J-P, Godeneche G, Mathis S, & Guillet G (2014). Neurodermatology. Handbook of Clinical Neurology, 121, 1561–1594. 10.1016/B978-0-7020-4088-7.00104-8 [DOI] [PubMed] [Google Scholar]
  55. Ohyama M, Terunuma A, Tock CL, Radonovich MF, Pise-Masison CA, Hopping SB, Brady JN, Udey MC, & Vogel JC (2006). Characterization and isolation of stem cell-enriched human hair follicle bulge cells. The Journal of Clinical Investigation, 116(1), 249–260. 10.1172/JCI26043 [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Pettan-Brewer C, & Treuting PM (2011). Practical pathology of aging mice. Pathobiology of Aging & Age Related Diseases, 1, 10.3402/pba.vli0.7202 [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Phan QM, Fine GM, Salz L, Herrera GG, Wildman B, Driskell IM, & Driskell RR (2020). Lef1 expression in fibroblasts maintains developmental potential in adult skin to regenerate wounds. eLife, 9. 10.7554/eLife.60066 [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Porter PS (1971). The genetics of human hair growth. Birth Defects Original Article Series, 7(8), 69–85. [PubMed] [Google Scholar]
  59. Powell BC, Passmore EA, Nesci A, & Dunn SM (1998). The Notch signalling pathway in hair growth. Mechanisms of Development, 78(1–2), 189–192. 10.1016/s0925-4773(98)00177-4 [DOI] [PubMed] [Google Scholar]
  60. Pun TB, Neupane A, Koech R, & Walsh K (2023). Detection and counting of root-knot nematodes using YOLO models with mosaic augmentation. Biosensors and Bioelectronics: X, 15, 100407. 10.1016/j.biosx.2023.100407 [DOI] [Google Scholar]
  61. Raul J-S, Cirimele V, Ludes B, & Kintz P (2004). Detection of physiological concentrations of cortisol and cortisone in human hair. Clinical Biochemistry, 37(12), 1105–1111. 10.1016/iclinbioche,200402010 [DOI] [PubMed] [Google Scholar]
  62. Rendl M, Lewis L, & Fuchs E (2005). Molecular Dissection of Mesenchymal–Epithelial Interactions in the Hair Follicle. PLOS Biology, 3(11), e331. 10.1371/journal.pbio.0030331 [DOI] [PMC free article] [PubMed] [Google Scholar]
  63. Salz L, & Driskell RR (2017). Horizontal Whole Mount: A Novel Processing and Imaging Protocol for Thick, Three-dimensional Tissue Cross-sections of Skin. Journal of Visualized Experiments: JoVE, 126, 56106. 10.3791/56106 [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Schielein MC, Tizek L, Ziehfreund S, Sommer R, Biedermann T, & Zink A (2020). Stigmatization caused by hair loss – a systematic literature review. JDDG: Journal Der Deutschen Dermatologischen Gesellschaft, 18(12), 1357–1368. 10.llll/ddg.14234 [DOI] [PubMed] [Google Scholar]
  65. Schindelin J, Arganda-Carreras I, Frise E, Kaynig V, Longair M, Pietzsch T, Preibisch S, Rueden C, Saalfeld S, Schmid B, Tinevez J-Y, White DJ, Hartenstein V, Eliceiri K, Tomancak P, & Cardona A (2012). Fiji—An Open Source platform for biological image analysis. Nature Methods, 9(7), 10.1038/nmeth.2019 [DOI] [PMC free article] [PubMed] [Google Scholar]
  66. Schlake T. (2005). FGF signals specifically regulate the structure of hair shaft medulla via IGF-binding protein 5. Development, 132(13), 2981–2990. 10.1242/dev.01873 [DOI] [PubMed] [Google Scholar]
  67. Schlake T. (2007). Determination of hair structure and shape. Seminars in Cell & Developmental Biology, 18(2), 267–273. 10.1016/j.semcdb.2007.01.005 [DOI] [PubMed] [Google Scholar]
  68. Schmidt B, & Horsley V (2012). Unraveling hair follicle-adipocyte communication. Experimental Dermatology, 21(11), 827–830. 10.llll/exd.12001 [DOI] [PMC free article] [PubMed] [Google Scholar]
  69. Schmidt-Ullrich R, & Paus R. (2005). Molecular principles of hair follicle induction and morphogenesis. BioEssays: News and Reviews in Molecular, Cellular and Developmental Biology, 27(3), 247–261. 10.1002/bies.20184 [DOI] [PubMed] [Google Scholar]
  70. Schneider CA, Rasband WS, & Eliceiri KW (2012). NIH Image to ImageJ: 25 years of image analysis. Nature Methods, 9(7), 671–675. 10.1038/nmeth.2Q89 [DOI] [PMC free article] [PubMed] [Google Scholar]
  71. Sennett R, & Rendl M (2012). Mesenchymal–epithelial interactions during hair follicle morphogenesis and cycling. Seminars in Cell & Developmental Biology, 23(8), 917–927. 10.1016/j.semcdb.2012.08.011 [DOI] [PMC free article] [PubMed] [Google Scholar]
  72. Sharov AA, Sharova TY, Mardaryev AN, di Vignano AT, Atoyan R, Weiner L, Yang S, Brissette JL, Dotto GP, & Botchkarev VA (2006). Bone morphogenetic protein signaling regulates the size of hair follicles and modulates the expression of cell cycle-associated genes. Proceedings of the National Academy of Sciences of the United States of America, 103(48), 18166–18171. 10.1073/pnas.06088991Q3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  73. Sotiropoulou PA, Karambelas AE, Debaugnies M, Candi A, Bouwman P, Moers V, Revenco T, Rocha AS, Sekiguchi K, Jonkers J, & Blanpain C (2013). BRCA1 deficiency in skin epidermis leads to selective loss of hair follicle stem cells and their progeny. Genes & Development, 27(1), 39–51. 10.1101/gad.206573.112 [DOI] [PMC free article] [PubMed] [Google Scholar]
  74. Sperling LC (1991). Hair anatomy for the clinician. Journal of the American Academy of Dermatology, 25(1 Pt 1), 1–17. 10.1016/0190-9622(91)70167-z [DOI] [PubMed] [Google Scholar]
  75. Sundberg JP (1994). Handbook of Mouse Mutations with Skin and Hair Abnormalities: Animal Models and Biomedical Tools. CRC Press. [Google Scholar]
  76. Sundberg JP, Berndt A, Sundberg BA, Silva KA, Kennedy V, Bronson R, Yuan R, Paigen B, Harrison D, & Schofield PN (2011). The mouse as a model for understanding chronic diseases of aging: The histopathologic basis of aging in inbred mice. Pathobiology of Aging & Age Related Diseases, 1. 10.3402/pba.vli0.7179 [DOI] [PMC free article] [PubMed] [Google Scholar]
  77. Takeo M, Toyoshima K-E, Fujimoto R, Iga T, Takase M, Ogawa M, & Tsuji T (2023). Cyclical dermal micro-niche switching governs the morphological infradian rhythm of mouse zigzag hair. Nature Communications, 14(1), 4478. 10.1038/s41467-023-39605-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  78. Thom E. (2016). Stress and the Hair Growth Cycle: Cortisol-Induced Hair Growth Disruption. Journal of Drugs in Dermatology: JDD, 15(8), 1001–1004. [PubMed] [Google Scholar]
  79. Toda Y, Okura F, Ito J, Okada S, Kinoshita T, Tsuji H, & Saisho D (2020). Training instance segmentation neural network with synthetic datasets for crop seed phenotyping. Communications Biology, 3(1), 1–12. 10.1038/s42003-020-0905-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  80. Trempus CS, Morris RJ, Bortner CD, Cotsarelis G, Faircloth RS, Reece JM, & Tennant RW (2003). Enrichment for living murine keratinocytes from the hair follicle bulge with the cell surface marker CD34. The Journal of Investigative Dermatology, 120(4), 501–511. 10.1046/j.1523-1747.2003.12088.x [DOI] [PubMed] [Google Scholar]
  81. Tschandl P, Rinner C, Apalla Z, Argenziano G, Codella N, Halpern A, Janda M, Lallas A, Longo C, Malvehy J, Paoli J, Puig S, Rosendahl C, Soyer HP, Zalaudek I, & Kittler H (2020). Human–computer collaboration for skin cancer recognition. Nature Medicine, 26(8), 1229–1234. 10.1038/s41591-020-0942-0 [DOI] [PubMed] [Google Scholar]
  82. Tsuji Y, Denda S, Soma T, Raftery L, Momoi T, & Hibino T (2003). A Potential Suppressor of TGF-β Delays Catagen Progression in Hair Follicles. Journal of Investigative Dermatology Symposium Proceedings, 8(1), 65–68. 10.1046/j.15231747.2003.12173.x [DOI] [PubMed] [Google Scholar]
  83. Ubbens JR, & Stavness I (2017). Deep Plant Phenomics: A Deep Learning Platform for Complex Plant Phenotyping Tasks. Frontiers in Plant Science, 8. 10.3389/fpls.2017.01190 [DOI] [PMC free article] [PubMed] [Google Scholar]
  84. Viola P, & Jones M (2001). Rapid object detection using a boosted cascade of simple features. Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001, 1, I–I. 10.1109/CVPR.2001.990517 [DOI] [Google Scholar]
  85. Wang S, Yatagawa T, Suzuki H, & Ohtake Y (2022). Image Based Measurement of Individual Fiber Lengths for Randomly Oriented Short Fiber Composites. Journal of Nondestructive Evaluation, 41(2), 45. 10.1007/sl0921-022-00876-z [DOI] [Google Scholar]
  86. Weger N, & Schlake T (2005a). Igfbp3 modulates cell proliferation in the hair follicle. The Journal of Investigative Dermatology, 125(4), 847–849. 10.1111/j.0022-202X.2005.23886.x [DOI] [PubMed] [Google Scholar]
  87. Weger N, & Schlake T (2005b). IGF-I Signalling Controls the Hair Growth Cycle and the Differentiation of Hair Shafts. Journal of Investigative Dermatology, 125(5), 873–882. 10.1111/j.0022-202X.2005.23946.X [DOI] [PubMed] [Google Scholar]
  88. Wei ML, Tada M, So A, & Torres R (2024). Artificial intelligence and skin cancer. Frontiers in Medicine, 11. 10.3389/fmed.2024.1331895 [DOI] [PMC free article] [PubMed] [Google Scholar]
  89. Whitehead JC, Hildebrand BA, Sun M, Rockwood MR, Rose RA, Rockwood K, & Howlett SE (2014). A Clinical Frailty Index in Aging Mice: Comparisons With Frailty Index Data in Humans. The Journals of Gerontology Series A: Biological Sciences and Medical Sciences, 69(6), 621–632. 10.1093/gerona/gltl36 [DOI] [PMC free article] [PubMed] [Google Scholar]
  90. Williams R, Westgate GE, Pawlus AD, Sikkink SK, & Thornton MJ (2021). Age-Related Changes in Female Scalp Dermal Sheath and Dermal Fibroblasts: How the Hair Follicle Environment Impacts Hair Aging. Journal of Investigative Dermatology, 141(4, Supplement), 1041–1051. 10.1016/j.jid.2020.ll.009 [DOI] [PubMed] [Google Scholar]
  91. Winkelmann RK (1959). The innervation of a hair follicle. Annals of the New York Academy of Sciences, 83, 400–407. 10.llll/j.1749-6632.1960.tb40915.x [DOI] [PubMed] [Google Scholar]
  92. Wolf FA, Angerer P, & Theis FJ (2018). SCANPY: Large-scale single-cell gene expression data analysis. Genome Biology, 19(1), 15. 10.1186/sl3059-017-1382-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  93. Wolff H, Fischer TW, & Blume-Peytavi U (2016). The Diagnosis and Treatment of Hair and Scalp Diseases. Deutsches Ärzteblatt International, 113(21), 377–386. 10.3238/arztebl.2016.0377 [DOI] [PMC free article] [PubMed] [Google Scholar]
  94. Xiao Y, Woo W-M, Nagao K, Li W, Terunuma A, Mukouyama Y-S, Oro AE, Vogel JC, & Brownell I (2013). Perivascular hair follicle stem cells associate with a venule annulus. The Journal of Investigative Dermatology, 133(10), 2324–2331. 10.1038/jid.2013.167 [DOI] [PMC free article] [PubMed] [Google Scholar]
  95. Yang J-H, Hayano M, Griffin PT, Amorim JA, Bonkowski MS, Apostolides JK, Salfati EL, Blanchette M, Munding EM, Bhakta M, Chew YC, Guo W, Yang X, Maybury-Lewis S, Tian X, Ross JM, Coppotelli G, Meer MV, Rogers-Hammond R, … Sinclair DA (2023). Loss of epigenetic information as a cause of mammalian aging. Cell, 186(2), 305–326.e27. 10.1016/j.cell.2022.12.027 [DOI] [PMC free article] [PubMed] [Google Scholar]
  96. Yu M, Yue Z, Wu P, Wu D-Y, Mayer J-A, Medina M, Widelitz RB, Jiang T-X, & Chuong C-M (2004). The developmental biology of feather follicles. The International Journal of Developmental Biology, 48(0), 181–191. 10.1387/iidb.031776my [DOI] [PMC free article] [PubMed] [Google Scholar]
  97. Zhang C, Wang D, Wang J, Wang L, Qiu W, Kume T, Dowell R, & Yi R (2021). Escape of hair follicle stem cells causes stem cell exhaustion during aging. Nature Aging, 1(10), 889–903. 10.1038/s43587-021-00103-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  98. Zhou P, Byrne C, Jacobs J, & Fuchs E (1995). Lymphoid enhancer factor 1 directs hair follicle patterning and epithelial cell fate. Genes & Development, 9(6), 700–713. 10.1101/gad.9.6.700 [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS2025756-supplement-1.pdf (9.7MB, pdf)

Data Availability Statement

No new large genomic datasets were created in this manuscript. The code implemented can be found at: https://github.com/DriskellLab/Makkar-et-al.-2024.

RESOURCES