Cutaneous graft-versus-host disease (cGVHD) is the leading cause of morbidity and mortality post-allogeneic hematopoietic cell transplantation (HCT). There is general uncertainty about the utility and safety of skin biopsy for diagnosing cGVHD in the early post-transplant period. In collaboration with UW Medicine and Mayo Clinic, I conducted a retrospective analysis of skin biopsies performed within 100 days post-HCT between 2000 and 2014. 740 biopsies from 602 patients (356 male, 256 female) were included in analysis. 87.1% (n=525) were white. The age range was 19-75y, with a mean of 50.3y. 456 (61.6%) skin biopsies were performed on inpatients, and 284 (38.3%) on outpatients. Only 8 (1.3%) patients had documented biopsy complications. The primary complication was excessive bleeding, which resolved after application of pressure bandage. Under the guidance of dermatopathologist Dr. Michi Shinohara, I conducted analyses to explore demographic and hematological features that may influence patient care or increase risk for biopsy complications. For example, on average, Hispanic patients received biopsies 8 days later following rash onset compared to their non-Hispanic counterparts. I am currently looking into clinical approach differences between Hispanic and non-Hispanic patients, such as complication documentation/assessment variability and presence of additional consultations. With respect to blood features, the mean neutrophil count was 0.228 K/µL and 2.98 K/µL and the mean platelet count was 18.7 K/µL and 100.6 K/µL for patients with and without biopsy complications, respectively. The complication rate for patients with either extremely low neutrophils (<0.11 K/µL) or platelets (<20 K/µL) was 5.3%. We conclude that skin biopsies performed in the immediate post-HCT period have a very low serious complication rate, even in patients with low cell counts. When skin biopsies are otherwise medically indicated in this patient population, concern regarding skin biopsy safety should not deter performance of this procedure in this patient population.

Merkel cell carcinoma (MCC) is an aggressive skin cancer with high risk of metastasis. Recent developments of PD-1/PD-L1 immunotherapy significantly improved treatment outcomes of metastatic MCC. While approximately half of patients respond to immunotherapy, the other half of the patients do not benefit. Thus, it has become increasingly important to identify factors that potentially impact immunotherapy response. Multiple studies in melanoma have demonstrated that patients who are older or have a higher body mass index (BMI) show better immunotherapy response. The purpose of this study is to explore the effects of age and/or BMI in MCC patients to immunotherapy response. I helped create a cohort of 183 patients who had undergone immunotherapy initially identified in a longitudinal single center registry. I then helped collect information about age and BMI at the start of immunotherapy along with other clinical features. Treatment response, disease-specific and overall survival were analyzed in 183 patients using cox regression and natural cubic spline models. During this process, I assisted in distinguishing the number of splines we would like to use for our analysis, as well as choosing the best model to accurately represent data for different variables. After adjusting for age, sex, and stage, BMI did not have significant impact on overall survival (p=1.0), objective response (p=0.5), or disease progression (p=0.8) while on immunotherapy. A nonlinear relationship between age and immunotherapy response was observed and showed potentially worse response to treatment in older patients. However, this was not statistically significant (p<0.1). Unlike prior studies in melanoma, we found that BMI does not have a significant impact on immunotherapy in MCC. Older age may have a negative impact on the response. This warrants additional research into the difference between melanoma and MCC to further elucidate mechanism of actions.

The epidermis is a multi-layered tissue at the body surface made of cells called keratinocytes that protect humans from infection and dehydration. Keratinocytes undergo a unique program of differentiation while moving upwards in the epidermal tissue. During the final stage of maturation, each keratinocyte must eliminate its organelles and nuclei to allow flattening of the cells to form the uppermost layers that provide a water-tight seal for the body. To understand this crucial biological process, I aimed to train an online artificial intelligence (AI)-based image analysis algorithm called Biodock AI to detect key morphological features of differentiating keratinocytes and of engineered epidermal tissue. I annotated specific cellular and tissue structures in a set of microscopy images to train a new supervised AI model to recognize these features. In brightfield histological images of engineered human skin, I used an area selection tool to identify the boundaries of the epidermis to assess tissue thickness. To train our pipeline to recognize a common pathological feature, I next labeled nuclei that were improperly retained in the cornified layers of various drug-treated tissues. After the algorithm was trained by Biodock AI, our pipeline successfully replicated our tissue outlines and identified retained nuclei. Zooming to the subcellular level, I next labeled organelle features within fluorescence microscopy images of keratinocytes. I selected endoplasmic reticulum (ER) fragments that had broken off the tubular network during differentiation. Our AI-trained image analysis pipeline successfully identified ER fragments with high concordance with those annotated by a lab member. In summary, I successfully trained and implemented AI-based image analysis pipelines to detect tissue and sub-cellular features that characterize the process of epidermal differentiation. Our results demonstrate the potential of AI algorithms to accelerate imaging-based research to understand cellular differentiation and tissue pathology while mitigating bias and error from human investigators.