Identifying Selective Fibroblast Stress in Alopecia Areata Through Single-Cell Transcriptomics

Abstract

Alopecia areata is the most common subset of the broader alopecia tree, where autoimmune attacks on hair follicles create patchy bald spots. Previous literature, collected from Google Scholar, has confirmed that alopecia areata remolds the scalp environment, but there is little on the specific cellular mechanisms that enable it. In this report, we aim to use single-cell analysis to find a specific cellular group that can help explain this mysterious disease. We found a UCSF sample set consisting of six alopecia areata patients and two healthy controls, using their single-cell RNA data. Central findings of this study include the discovery of an anomalous fibroblast state in Cluster 24, which had a stress signature that exceeded any of its neighboring clusters. This was bolstered by defining expressed genes using the Gene Ontology database, predicted ligand-receptor interactions, and the amplified display of target genes (JUN, GADD45B, HSP1A1, FOS). We ultimately posited that a distinctive fibroblast state, that does not neatly fit into traditional classifications, likely contributes to the overregulation of the scalp that engenders hair loss. These signs were also evenly distributed across our six affected samples, ruling out the possibility of one sample dominating the results. This information contributes to the understanding of the pathophysiology of alopecia areata and emphasizes a need to explore single-cell analyses on more extreme forms of alopecia.

Keywords: Alopecia areata, data science, fibroblasts, scRNA-seq, clustering, hair loss, statistics, scalp, Gene Ontology

Introduction

Alopecia areata is not a disorder that targets individual follicles; rather, it influences the structure that maintains the scalp. Seeing that hair growth is reliant on smooth communication between different cell groups¹, bald spots are often caused by disrupted signaling between keratinocytes (cells in the epidermis that produce keratin) and other cells, breaking down the pathways that keep follicles healthy². Therefore, investigating alopecia areata requires thinking beyond the follicles themselves and examining the environment that surrounds them.

More recent studies paint the follicle as something always transmitting and receiving information from its neighboring cells. This means that dermal fibroblasts are not restricted to producing collagen, instead of dictating actions at the tissue level. Recent analyses revealed that fibroblasts have an assorted population with separate cell groupings, each bringing a different value to the table³. More importantly, they remold their own genes when encountering changes in inflammation, environment, or age⁴. All in all, modern research asserts that fibroblasts are important factors in tissue reformation due to their ability to constantly evolve based on the signals they receive.

This all reaffirms how fibroblasts go above and beyond just supporting tissue. These cells help engineer the underlying structure that follicles are rooted in using collagen, while also secreting cytokines necessary for basic cellular processes⁵. It is important to note that there are two types of fibroblasts: papillary and reticular. The former resides closest to the epidermis, officiating follicular activity, whereas the latter contributes more towards the extracellular matrix (ECM)⁶. This shows how fibroblasts have different functionalities based on their relative location in the skin.

Interestingly, a fibroblast does not have to change locations to change behaviors. The behavior can change if the environment changes. Studies on different fibrotic skin diseases have seen fibroblasts develop a form in which they have increased inflammatory signalling, which developed from long-term exposure to stress⁷. On the flip side, these very same cells can also morph into an obsessive healing unit when tissue is damaged⁸. Seen alongside each other, these probes prove that fibroblast function can and will reorient itself provided certain circumstances.

Fibroblast behavior has also demonstrated its influence on autoimmune diseases like alopecia areata. Deep dives into these syndromes show that some fibroblasts direct immune behavior, contrary to previous views⁹. Further, a spatially resolved atlas shows that fibroblast subtypes surprisingly pop up in reaction to disease¹⁰; these can be tracked by checking chemokine production and stubborn inflammatory indicators¹¹. Overall, this shows that fibroblast transitioning is not intertwined with any specific disease but likely just a normal reaction to recurring stress.

Uncertainty remains with associating these stressed fibroblasts with alopecic scalp¹². However, if this shape can be proven, it could help answer some questions surrounding the relationship between follicle health and changes in the dermis. As prefaced earlier, stress can bring new abilities of fibroblasts to light. Specifically, they express different genes, some responsible for responding to oxidative stress and physical strains, others that also prioritize cell survival¹³. In the long run, though, this would probably disrupt the ECM, impeding hair growth¹⁴.

Newer evidence is overturning this notion, placing fibroblasts as a keystone component in the disease microenvironment. Historically, alopecia areata researchers have focused on the autoimmune side; more current studies have revealed fibroblast subpopulations which potentially participate in single-cell signalling. Lan et al. (2026) recently published an array of fibroblasts expressing macrophage migration inhibitory factor (MIF) in an alopecia areata patch; its calculated ligand-receptor inferences support the idea of fibroblasts conversing with dendritic cells¹⁵. Cutting-edge therapeutic solutions for alopecia areata targeting fibroblasts are promising, evidencing a link between fibroblasts and disease advancement¹⁶. These studies have simply nicked the iceberg to understand how fibroblasts contribute to immune defense. However, a problem arises in the inability to concretely define these groups.

The goal of this study is to see whether alopecia areata scalp fibroblasts have a stress-activated fibroblast cluster that falls through ordinary classifications. We hypothesized that an unbiased sorting of data would produce clusters of fibroblasts that had a bloated stress signature compared to their peers. Our approach was based on finding if the two variables are correlated, detailing the characteristics that made the cluster unique, and finally verifying the quality of the results.

Methods

As for the data, we utilized scalp samples from a prior study of UCSF through the NCBI Gene Expression Omnibus (GSE233906). From it, we only took 6 of the alopecia areata samples and two of the healthy scalp samples, excluding the mice samples.

Sample Index	GEO Accession (GSM)	Species	Donor ID	Disease Status	Source	Lesion  Status	Cells Retained	Exclusion
1	GSM7432360	Homo sapiens	AA Patient 1	Alopecia Areata	Scalp Skin	Lesional	~1,450	No
2	GSM7432361	Homo sapiens	AA Patient 2	Alopecia Areata	Scalp Skin	Lesional	~1,120	No
3	GSM7432362	Homo sapiens	AA Patient 3	Alopecia Areata	Scalp Skin	Lesional	~1,890	No
4	GSM7432363	Homo sapiens	AA Patient 4	Alopecia Areata	Scalp Skin	Lesional	~980	No
5	GSM7432364	Homo sapiens	AA Patient 5	Alopecia Areata	Scalp Skin	Lesional	~1,310	No
6	GSM7432365	Homo sapiens	AA Patient 6	Alopecia Areata	Scalp Skin	Lesional	~1,640	No
7	GSM7432366	Homo sapiens	Control Patient 7	Healthy Normal	Scalp Skin	Non- lesional	~1,550	No (Retained as control)
8	GSM7432367	Homo sapiens	Control Patient 8	Healthy Normal	Scalp Skin	Non-lesional	~1,720	No (Retained as control)
9	GSM7432368	Mus musculus	UG Biological Replicate 1	No disease	Dorsal Skin	Non-lesional	Did Not Calculate	Yes
10	GSM7432369	Mus musculus	UG Biological Replicate 2	No disease	Dorsal Skin	Non-lesional	Did Not Calculate	Yes
11	GSM7432370	Mus musculus	AA Biological Replicate 1	Total hair loss	Dorsal Skin	Lesional	Did Not Calculate	Yes
12	GSM7432371	Mus musculus	AA Biological Replicate 2	Total hair loss	Dorsal Skin	Lesional	Did Not Calculate	Yes
13	GSM7432372	Mus musculus	AA Biological Replicate 3	Total hair loss	Dorsal Skin	Lesional	Did Not Calculate	Yes
14	GSM7432373	Mus musculus	UG Biological Replicate 3	No disease	Dorsal Skin	Non-lesional	Did Not Calculate	Yes

The intended use case for this data was only on sorting lymphocyte datasets. To counter this, our programming pipeline had an ingrained filter pooling together only the structural and skin cells. Part of this filter included checking the quality of the cells we were letting in. 75% of our cells ranged between 1500 and 4500 unique genes, so we only allowed cells with 200 to 6000 unique genes to remove doublets. Additionally, we required at least 500 unique molecular identifiers (UMI) to remove roaming background mRNA. In the same vein, when a cell dies, the cytoplasm leaks out, but the mitochondria stay intact. Therefore, we set a mitochondrial gene percentage cutoff of 5% to keep only healthy cells in the pipeline. The dermis contains a plethora of other cells beyond fibroblasts, so to prevent contamination, we filtered only cells that showed signs of established fibroblast markers. We did not aggregate by cell to prevent pseudo-replication, a common error in single-cell analyses, from overstating our results. Later in the process, when analyzing the stress signature, we ensured that each donor had near equivalent levels to rule out anomalies.

Our programming language of choice was R (v4.5.1), using packages like Seurat (v5.3.0), SingleR (v2.10.0), celldex (v.1.18.0), nichenetr (v2.0.3) and clusterProfiler (v4.16.0). We integrated all the data together in one pool using canonical correlation analysis (CCA) across 30 dimensions, paired with figure wielded 20 principal components (PCs). For clustering, we used the Louvain algorithm, a greedy method that optimizes modularity, to identify communities in our dataset.

We first sorted all cells against a R dataset that sorted them into groups (e.g., fibroblast, keratinocyte, endothelial). For example, if a cell showed signs of DCM or LUM, our code would match it up with the fibroblast batch. This same process was also used to establish subdivisions, like determining which fibroblasts were papillary, reticular, or dermal. Only then did we begin grouping them into clusters and generate a stress score.

We looked for any overwhelming expression from any one of the 8 donor records to double-check the veracity of our results. We also did the Kruskal-Wallis test with Dunn’s post-hoc test, under the guidance of my statistics professor Kelvin Leeds. Both of these works towards measuring if and how each cluster is different. UCSF’s patients were identified randomly with no collection of personal information, ruling out any potential error in reporting data.

Results

We drew a line between Cluster 24 and the a-priori score, which was preset to prevent circular reasoning. The heatmap is sorted in descending order from left to right, where you can see Cluster 24 being the farthest to the left. More importantly, it wasn’t just one box on the heatmap that drove this. It was an even, cohesive response instead of Group 10, for example, with their strong expression of HMOX1. The genes highlighted in the heatmap for G24, like JUN or GADD45B, are prominent enough to warrant further investigation.

The volcano plot placed a stamp of approval on our notion. It showed that G24 exuded tissue remodeling genes while suppressing predicted genes scientists nowadays affiliate with papillary or reticular fibroblasts. Doing run-of-the-mill tasks like tissue repair while also battling stress at the same time shows how unsung the contributions of fibroblasts are. This special character they adopt may even have unknown abilities that only come with the particular genetic makeup they have.

Breaking down the GO results, which had actin filament organization and unfolded protein response at the top, revealed that the surrounding environment has a root cause of mutations giving fibroblast more work to deal with. The genes and their significance came from the clusterProfiler library, a R package we prefaced earlier.

Our box-and-whisker plot shows that it was only clearly demonstrated in Cluster 24. This rules out that it was a tissue-wide apocalypse that induced high stress signaling in all cells. In Figure 4, we can see there are 3 separate box-and-whisker plots per cluster. While the priori stress score is most important, adding a random mix of genes as well can help buttress that our results are not made from transcripts that were just floating in the background. Obviously, the housekeeping control (made of genes needed for basic biological function) has the highest module score since those would be most prevalent. However, it does not deter our results since if they are all around the same level, it shows that the spike in expression is concentrated in genes that truly matter.

Nevertheless, it would be prudent to forget to check the quality of the cell population. Samples taken as per the techniques of UCSF’s study often leave a lot of immune cells and cell fragments that need to be filtered out. To make sure that we were not left with pointless leftovers, we calculated two separate panels: one to look for fibroblast markers in G24, and another to differentiate itself from cluster that are similar in genetic makeup.

While we now understand more about Cluster 24, we have yet to explore its external engagement. Our tests predicted secretions of TGFB1 and CXCL12 to interact with receptors like TGFBR2 and CXCR4. This can be read as G24 having potential ties with the immune system, which could perhaps explain the autoimmune nature of this disease. This can explain how the cluster actually participates in disease rather than being a far-flung onlooker.

All donors showed a near-equal ratio of expression across most tests. This can eliminate the possibility of one contaminated or anomalous sample influencing our results. We chose to do our tests aggregating on a donor level instead of a cell level to make our correlation more meaningful. Ultimately, this reinforces that our findings are not only meaningful but reproducible.

Sample	JUN	GADD45B	HSPA1A	FOS
1	15.2	16.4	15.8	16.1
2	16.8	15.5	17.2	16.4
3	17.1	17.2	16.3	15.9
4	15.9	16.8	16.5	17.5
5	16.5	16.1	17	16.8
6	18.5	18	17.2	17.3

Discussion

We concluded that there is a difference between one cluster and all the other ones within our integrated pipeline. As we gather more information to understand how and why fibroblasts are actually causing hair loss, it is important we understand what this cell population truly does¹⁷. If this type of activity in fibroblasts does indeed remodel the follicular environment, then it would be integral to both research in this sector as well as future therapeutical solutions.

The genes of this group are expressed in a very unique pattern. JUN, FOS, HSPA1A, HSP90AA1, and GADD45B are all elevated (Figure 1). This typically signifies protein folding protection, and other variants of self-care as a cell activates as it tries to fight back against an external force to maintain homeostasis. In this case, the response happens to go hand-in-hand with ECM remodeling. This is yet another piece of evidence striking fibroblasts down to a more adaptive role in the dermal ecosystem.

Since follicles rely on an ECM that must be perfectly balanced, this is especially pertinent to the scalp. When certain transcription factors are removed, it hangs fibroblasts out to dry, making them more prone to irregularly changing the collagen structure, mutations in its rigidity, or even just hyperactivity of certain signaling molecules¹⁸. Overcorrection can also harm follicles. When your body tries to nurture these cells back to normality, sometimes rapid change can adversely effect this super sensitive system.

G24 being on the radar of each donor shows that what we are uncovering is not an artifact (Table 2), proving that is not just an example of the “Texas Sharpshooter” fallacy. It also brings into question whether this tendency is present across other diseases, like in hair loss from systemic lupus erythematosus. However, single cell data does have a decent number of variances between batches, making it harder to pick out its attached biological impact.

One would think that there would be a sizeable chunk of gene transmission that makes Cluster 24 more similar than different to its neighboring cell clusters. Instead, G24 is quite isolated from the rest of them (Figure 5). Fibroblasts do not gently oscillate between levels of stress; they sometimes reach a breaking point; once hit, they are irreversibly detached from its nearby tissue. This concept jells with the signs we have seen with G24, throwing any possibility of external factors majorly influencing our results. These abandoned cells show how alopecia areata makes “pockets” of cells that remain in perpetual despair while others are unaffected, perhaps engendering the spotty hallmarks of the disease.

It was interesting to see how each cluster stratified. Most of them consisted of papillary fibroblasts, understandably since the scalp requires a lot of them for structural support. The clusters statistically closest to G24 were technically papillary, but it also had some mismatches. This difference is intriguing because papillary fibroblasts are known for their role in follicle support and epidermal communication¹⁹. On the flip side, Cluster 24 cells showed increased stress and remodeling signals (Figure 4). Just because this group does not neatly fit in standard definitions of fibroblasts does not mean it should be thrown away. Instead, we can say that G24 cells are synthesizing their lineage with environmental requirements creating a hybrid form.

That theory clicks with modern scientific opinions on the topic. They see fibroblasts less and less as a static cell group but one that is a jack-of-all-trades (quiescent, activated, repair related, stress responsive, etc.)²⁰. It would be more fitting to rank cells on a spectrum, instead of drawing precise lines around cell groups that are constantly evolving based on their whereabouts in subcutaneous tissue. Our results support this, since they cannot accommodate traditional categorizations. Tissue can be ordered under two things: what a cell does and a what state it is in. We argue that the latter is more influential than the former and is glossed over by type-only analysis.

Biological Process Description	Gene Count	Gene Ratio	p-value (Adj.)
small GTPase-mediated signal transduction	158	0.052	1.1×10−11
ribonucleoprotein complex biogenesis	134	0.0498	1.4×10−11
viral process	125	0.0452	1.7×10−11
actin filament organization	115	0.0448	2.1×10−11
regulation of proteolysis	102	0.0425	2.3×10−11
response to oxidative stress	96	0.0422	2.5×10−11
negative regulation of locomotion	98	0.042	2.6×10−11
regulation of apoptotic signaling pathway	92	0.0415	2.8×10−11
negative regulation of cell motility	88	0.0402	2.9×10−11
cell-substrate adhesion	84	0.0395	3.1×10−11
response to oxygen levels	85	0.037	3.2×10−11
intrinsic apoptotic signaling pathway	82	0.036	3.4×10−11
response to decreased oxygen levels	105	0.0345	4.8×10−11
response to transforming growth factor beta	78	0.0342	3.6×10−11
ribosome biogenesis	80	0.0342	3.7×10−11
protein localization to nucleus	88	0.034	3.9×10−11
endothelial cell migration	75	0.0338	4.1×10−11
response to endoplasmic reticulum stress	88	0.0322	4.2×10−11
viral life cycle	92	0.0318	4.0×10−11
Rho protein signal transduction	22	0.0195	4.5×10−11

Oxidative stress and unfolded protein responses popped up quite high on our GO enrichment, showing how difficult it is for cells to maintain homeostasis (Figure 3, Table 3). These tags only show up when cells are hit with inflammation, radical oxygen compounds, or other random damage²¹. It writes a narrative of a cell leaving its comfort zone while pulling out all the stops to stay functional. This twofoldness is key to evaluating alopecic scalp.

We also predict these fibroblasts messages peripherally. In our inferred ligand-receptor pairs, two interactions caught our eye. TGFBR2-related signaling is proven to impact epithelial cell behavior, changing how they support capillaries²². On the other hand, CXCR4-related signaling ties together tissue remodeling and immune regulation, which aligns with autoimmune disorders in general. While these interactions are predictive, they can spell out how fibroblasts impact follicle growth, even from a distance.

Stress activation should not be loathed. It probably needs to keep the tissue functional even as it faces pressure. Taking that into consideration, it may not be the presence of those cells but perhaps how exuberantly they express themselves. The difference between a structural or destructive fibroblast could stem from how long the stress response persists. This is important because it shows how fibroblasts can go undetected as they shift from a helping hand to an agent of chaos.

Our study does have limitations though. An attribute of single-cell analyses is that we can read RNA signaling but cannot extrapolate how much of a protein gets generated. It also can’t tell you whether the cell group is near blood vessels or inflammatory infiltrates. Even so, our results are still substantial. While no single study format can entirely cover all the different niches of the disease, our research can complement other articles to create a more complete picture.

The mechanism in which the samples were extracted in UCSF’s study can artificially trigger a gene response (e.g. JUN, FOS, etc.). Nevertheless, we remain confident in our conclusion. First, the extraction method was the same in both the control and diseased tissue (Table 1). This may create variance in the exact expression amount, but the discrepancy between the two groups cannot be papered over. Existing publications assert that if such an effect existed, it would be equally dissipated over many clusters instead of spiking in one²³. As the forefront of biotechnology advances, state-of-the-art examination technologies paired with in-situ hybridization could provide more elaborate results in the near future²⁴.

There are steps we can take to overcome these limitations. Spatial transcriptomics could specify where the cell group exists in tissue²⁵. Protein-level validation can check if RNA signals are actually translated to protein production. Furthermore, we can expand beyond alopecia areata to androgenic or traction alopecia to see if a pattern exists. With these confirmatory tests, we can check if G24 would be a viable target for pharmaceuticals.

To sum up, we see alopecia areata as a disease that holds a great deal of gravity, pulling in much more than just follicles. Our project invested in Cluster 24, a fibroblast population, that changes its features based on the biochemical backdrop it faces. New trailblazing therapies have to understand the impact cells further from the follicular niche have on hair loss. Broadening our horizons, we also concluded that cell state is undoubtedly instrumental to disease, sometimes more so than cell type.

Acknowledgments

I would love to thank Professors Kelvin Leeds and Revathi P. Shenoy for empowering me with the skills needed to pursue this project. Shoutout to Sanbomics and Pratik Vangal for inspiring this intellectual pursuit in the first place. Finally, I would like to thank the NHSJS Peer Review and Publication staff for their insightful feedback and all the help they provided along the journey.

References

1. J. H. Lee, S. Choi. Deciphering the molecular mechanisms of stem cell dynamics in hair follicle regeneration. Experimental & Molecular Medicine. Vol. 56, pg. 110-119, 2024, https://doi.org/10.1038/s12276-023-01151-5. [↩]
2. J. Park, E. K. Jun, D. Son, W. Hong, J. Jang, W. Yun, B. S. Yoon, G. Song, I. Kim, S. You. Overexpression of nanog in amniotic fluid–derived mesenchymal stem cells accelerates dermal papilla cell activity and promotes hair follicle regeneration. Experimental & Molecular Medicine. Vol. 51, pg. 1-15, 2019, https://doi.org/10.1038/s12276-019-0266-7. [↩]
3. C. Philippeos, S. B. Telerman, B. Oulès, A. O. Pisco, T. J. Shaw, R. R. Elgueta, G. Lombardi, R. R. Driskell, M. Soldin, M. D. Lynch, F. M. Watt. Spatial and single-cell transcriptional profiling identifies functionally distinct human dermal fibroblast subpopulations. Journal of Investigative Dermatology. Vol. 138, pg. 811–825, 2018, https://doi.org/10.1016/j.jid.2018.01.016. [↩]
4. L. Solé-Boldo, G. Raddatz, S. Schütz, J. P. Mallm, K. Rippe, A. S. Lonsdorf, M. Rodríguez-Paredes, F. Lyko. Single-cell transcriptomes of the human skin reveal age-related loss of fibroblast priming. Communications Biology. Vol. 3, pg. 188, 2020, https://doi.org/10.1038/s42003-020-0922-4. [↩]
5. J. D. Humphrey, E. R. Dufresne, M. A. Schwartz. Mechanotransduction and extracellular matrix homeostasis. Nature Reviews Molecular Cell Biology. Vol. 15, pg. 802-812, 2014, https://doi.org/10.1038/nrm3896. [↩]
6. H. Pageon, H. Zucchi, D. Asselineau. Distinct and complementary roles of papillary and reticular fibroblasts in skin morphogenesis and homeostasis. European Journal of Dermatology. Vol. 22, pg. 324-332, 2012, https://doi.org/10.1684/ejd.2012.1693. [↩]
7. C.-C. Deng, Y.-F. Hu, D.-H. Zhu, Q. Cheng, J.-J. Gu, Q.-L. Feng, L.-X. Zhang, M. Xu, Z. Rong, S.-L. Ye, F.-Q. Wu. Single-cell RNA-seq reveals fibroblast heterogeneity and increased mesenchymal fibroblasts in human fibrotic skin diseases. Nature Communications. Vol. 12, pg. 3709, 2021, https://doi.org/10.1038/s41467-021-24110-y. [↩]
8. C. F. Guerrero-Juarez, P. H. Dedhia, S. Jin, R. Ruiz-Vega, D. Ma, Y. Liu, K. Yamaga, O. Shestova, D. L. Gay, Z. Yang, K. Kessenbrock, Q. Nie, M. V. Plikus. Single-cell analysis reveals fibroblast heterogeneity and myeloid-derived adipocyte progenitors in murine skin wounds. Nature Communications. Vol. 10, pg. 650, 2019, https://doi.org/10.1038/s41467-018-08247-x. [↩]
9. X. Xu, D. Zhang, P. Zanvit, et al. Anatomically distinct fibroblast subsets determine skin autoimmune patterns. Nature. Vol. 601, pg. 118–124, 2021, https://doi.org/10.1038/s41586-021-04221-8. [↩]
10. K. Liu, Y. Cui, H. Han, et al. Fibroblast atlas: Shared and specific cell types across tissues. Science Advances. Vol. 11, no. 14, Article eado0173, 2025, https://doi.org/10.1126/sciadv.ado0173. [↩]
11. Y. Gao, et al. Cross-tissue human fibroblast atlas reveals myofibroblast subtypes with distinct roles in immune modulation. Cancer Cell. Vol. 42, pg. 1764–1783, 2024, https://doi.org/10.1016/j.ccell.2024.08.020. [↩]
12. N. Khodeneva, M. A. Sugimoto, C. S. A. Davan-Wetton, T. Montero-Melendez. Melanocortin therapies to resolve fibroblast-mediated diseases. Frontiers in Immunology. Vol. 13, pg. 1084394, 2023, https://doi.org/10.3389/fimmu.2022.1084394. [↩]
13. X. Zhu, et al. A systematic characterization of fibroblast subtypes and heterogeneity. iScience. Vol. 28, no. 12, pg. 113915, 2025, https://doi.org/10.1016/j.isci.2025.113915. [↩]
14. A. J. G. McDonagh, L. Cawood, A. G. Messenger. Expression of extracellular matrix in hair follicle mesenchyme in alopecia areata. British Journal of Dermatology. Vol. 123, pg. 717–724, 1990, https://doi.org/10.1111/j.1365-2133.1990.tb04188.x. [↩]
15. X. Lan, et al. Single-cell landscape of immune remodeling in alopecia areata suggests MIF+ fibroblasts and their potential ligand-receptor crosstalk with dendritic cells. Frontiers in Medicine. Vol. 13, pg. 1849368, 2026, https://doi.org/10.3389/fmed.2026.1849368. [↩]
16. R. B. Jalili, N. Ansari, A. Elahi, et al. Fibroblast cell-based therapy prevents induction of alopecia areata in an experimental model. Cell Transplantation. Vol. 27, pg. 1375–1386, 2018, https://doi.org/10.1177/0963689718773311. [↩]
17. S. An, M. Zheng, I. G. Park, L. Song, J. Kim, M. Noh, J.-H. Sung. CXCL12 drives reversible fibroimmune remodeling in androgenetic alopecia revealed by single-cell RNA sequencing. International Journal of Molecular Sciences. Vol. 26, pg. 6568, 2025, https://doi.org/10.3390/ijms26146568. [↩]
18. S. Quist, J. Quist. Keep quiet—how stress regulates hair follicle stem cells. Signal Transduction and Targeted Therapy. Vol. 6, pg. 364, 2021, https://doi.org/10.1038/s41392-021-00772-4. [↩]
19. B. Liu, A. Li, J. Xu, Y. Cui. Single-cell transcriptional analysis deciphers the inflammatory response of skin-resident stromal cells. Frontiers in Surgery. Vol. 9, pg. 935107, 2022, https://doi.org/10.3389/fsurg.2022.935107. [↩]
20. V. S. LeBleu, E. G. Neilson. Origin and functional heterogeneity of fibroblasts. The FASEB Journal. Vol. 34, pg. 3519-3536, 2020, https://doi.org/10.1096/fj.201903188r. [↩]
21. C. X. C. Santos, L. Y. Tanaka, J. Wosniak Jr., F. R. M. Laurindo. Mechanisms and implications of reactive oxygen species generation during the unfolded protein response: Roles of endoplasmic reticulum oxidoreductases, mitochondrial electron transport, and NADPH oxidase. Antioxidants & Redox Signaling. Vol. 11, pg. 2409–2427, 2009, https://doi.org/10.1089/ars.2009.2625. [↩]
22. M. Rendl, L. E. Lewis, E. Fuchs. Molecular dissection of mesenchymal–epithelial interactions in the hair follicle. PLoS Biology. Vol. 3, pg. e331, 2005, https://doi.org/10.1371/journal.pbio.0030331. [↩]
23. S. C. van den Brink, F. Sage, A. Vértesy, B. Spanjaard, J. D. Peterson-Maduro, J. Baron, C. Robin, A. van Oudenaarden. Single-cell sequencing reveals dissociation-induced gene expression in tissue subpopulations. Nature Methods. Vol. 14, pg. 935–936, 2017, https://doi.org/10.1038/nmeth.4437. [↩]
24. S. K. Longo, M. G. Guo, A. L. Ji, P. A. Khavari. Integrating single-cell and spatial transcriptomics to elucidate intercellular tissue dynamics. Nature Reviews Genetics. Vol. 22, pg. 627–644, 2021, https://doi.org/10.1038/s41576-021-00370-8. [↩]
25. L. Steele, B. Olabi, K. Roberts, P. Mazin, S. Koplev, C. Tudor, B. Rumney, C. Admane, T. Jiang, D. Correa-Gallegos, K. P. Chakala, A. Binkevich, N. H. Gopee, A. V. Predeus, M. Prete, E. Winheim, K. Annusver, A. Forsthuber, L. Francis, M. Haniffa. A single-cell and spatial genomics atlas of human skin fibroblasts reveals shared disease-related fibroblast subtypes across tissues. Nature Immunology. 2025, https://doi.org/10.1038/s41590-025-02267-8. [↩]