IPC Councilor April Armstrong, MD, MPH Keck School of Medicine, University of Southern California Studio, California, United States The Eugene M. Farber lecture honors an investigator whose work is relevant to expanding insights into the pathophysiology and treatment of psoriasis and cutaneous autoimmune disease. This year’s recipient, IPC Councilor Dr. April Armstrong, delivered a presentation titled “Getting Clear: Psoriasis Advancements and Beyond,” specifically focused on therapeutic advances, artificial intelligence, and health care delivery. Therapeutic advances Despite the existing approved therapies, many psoriasis patients worldwide remain under- or untreated.1 In 2017, the National Psoriasis Foundation established treatment targets for clinical practice, with goals of very low BSA.2,3 Though advances in pathophysiology knowledge drive therapeutic advancement, therapeutic trials also improve our understanding of vital pathophysiologic pathways.4,5 As a result, new superior psoriasis therapies regularly become available. A promising novel, topical aryl hydrocarbon receptor modulating agent, tapinarof, decreases Th17 and Th2 cytokines, increases antioxidant activity, and repairs skin barrier.6-8 An emerging oral therapy, the highly selective Tyk2 inhibitor deucravacitinib, modulates the JAK-STAT pathway and decreases IL-12, IL-23, and Type-1 IFN levels.8-12 The only pegylated anti-TNFα biologic, certolizumab, is safe in pregnancy as it does not cross the placenta.13 As a class, IL-17 inhibitors are highly efficacious for both psoriasis and psoriatic arthritis (PsA). Similarly, IL-23 inhibitors have robust efficacy and durability but require infrequent injections.13-15 Artificial Intelligence (A.I.) As technology advances, we may begin to leverage A.I. to better phenotype patients for individualized therapy. As current remarkable A.I. programs far outperform humans in areas such as complex games, the question arises: can A.I. shape our knowledge and therapies for psoriasis and other dermatologic diseases? Defined, A.I. is a computer system able to effectively perform tasks that would typically require human intelligence (sensing, reasoning, acting, adapting). ‘Machine learning’ in A.I. uses algorithms whose performance improves as they are exposed to more data over time, and its subset, ‘deep learning,’ utilizes multilayered neural networks to learn from vast amounts of data.16 Machine learning can develop phenotypic clusters with potential therapeutic and prognostic significance without being explicitly programmed.17 An example of machine learning’s utility in medicine can be seen in a Rheumatology secukinumab study where the investigators identified distinct clusters of patients with PsA based on baseline articular, entheseal, and cutaneous disease manifestations. Machine learning detected 13 different phenotypic clusters of patients with psoriatic disease, reporting mean PASI sub scores and % of patients with tender joints across these clusters. Without pre-specified hypotheses, machine learning can discover these clusters based on the data alone.18,19 Health Care Delivery Though the COVID-19 pandemic thrust telemedicine to the forefront, teledermatology advancements preceded the pandemic. A 2018 RCT of nearly 300 patients by Dr. Armstrong studied face-to-face care versus asynchronous teledermatology, where patients and primary care physicians provided history and still photos with subsequent dermatologist assessment, education, and management. Her group found that the online collaborative model effectively improved clinical outcomes (PASI and BSA responses) for psoriasis patients as in-person care and even slightly outperformed in-person care in the patient’s global assessment.20 Though telehealth models can improve psoriasis care, whether they are sustainable and scalable will depend on reimbursement, technology, medicolegal considerations, special body site evaluation, and workforce considerations. To that end, the IPC is working to provide guidance for clinicians on the use of teledermatology.
IPC Councilor Nicole Ward, PhD Case Western Reserve University Cleveland, Ohio, United States Though psoriatic arthritis (PsA) occurs in up to 30% of psoriasis patients, we currently cannot predict who will develop PsA until they begin to show irreversible joint damage. There is a significant overlap of differentially expressed genes (DEGs) amongst different animal models of psoriasis with human psoriasis.1,2 To date, 13 psoriasis mouse models develop PsA-like phenotype, and in most models, the skin inflammation predates the joint disease. Dr. Ward discussed her lab’s goal to model personalized medicine and drug responsiveness by comparing mouse phenotypes with psoriasis patient endotypes. Specific to this lecture, Dr. Ward’s lab examined whether mouse models can predict which psoriasis patients will develop PsA. Her lab has developed three mouse models (KC-Tie2, IL-17C, and KLK6) that develop characteristic psoriasiform skin changes and respond to therapeutics. However, only the KLK6 model displays a similar phenotype to PsA patients (dactylitis, osteopenia, kyphosis, degenerative sacroiliac changes) with both peripheral and axial disease. KLK6 mouse skin also shows increases in key disease signature cytokines (IL-22, IL-17A, etc.) and transcriptionally shows tight overlap with human disease.3-7 Using RNA-Seq, Dr. Ward is attempting to identify novel human disease biomarkers using transcriptomics and bioinformatics. Her lab compared genes for the KLK6 mice against the two other mouse models, lesional and non-lesional skin in cutaneous-only patients (PsC) versus those with PsA, lesional modulation with infliximab over ten weeks, and blood of PsC vs. PsA patients.7 In KLK6 mice, they identified 19 transcripts, of which 11 were increased and eight decreased. While some had been previously associated with psoriasis, others had not. Their current goal is to generate a list of putative PsA Biomarkers to test in a well-defined patient cohort via comparison of blood and skin from healthy controls, PsC, PSA, osteoarthritis, rheumatoid arthritis, and other inflammatory diseases. This will allow testing for biomarker success that only occurs in PsA patients. Once a potential biomarker is identified, prospective testing in a new patient cohort will test success at predictability could develop a PsA biomarker. The vision is to ultimately develop a diagnostic test for a biomarker to predict which patients will develop PsA, with goals for personalized medicine.7 Then, artificial intelligence and machine learning could potentially identify cause and effect, correlation, mechanisms of action and comorbidities, drug responders, and more.
Vinod Chandran, MD PhD University of Toronto Toronto, Ontario, Canada Psoriasis and psoriatic arthritis (PsA) are heterogenous diseases.1 Though there is overlap in many genetic pathways (IL-17, IL-23, etc.), PsA synovium gene expression differs from that of skin.2 Given the heterogeneity of psoriatic disease, biomarker research is becoming increasingly employed. Biomarkers are objectively measured and evaluated characteristics that can serve as indicators of normal biologic processes, pathogenic processes, or pharmacologic responses to therapies.3-5 Prediction, a time-denominated inference about an unknown present or future state based on available information, is vital in biomarkers. Three common types of medical predictions include risk, diagnosis, and prognosis. Optimal predictive diagnostic tests are significant, accurate, generalizable, and easily deployed.5-7 For psoriatic disease, biomarkers that can predict risk, assess activity, and monitor therapeutic response could revolutionize diagnosis and treatment. Machine learning is a budding tool in biomarker research. Patrick et al. show you can predict PsA with high confidence in at least 10% of patients, based on GWAS in several cohorts across 200 markers. A challenge is that when precision was high, the sensitivity was low, but this could be corrected by combining various methods.8 Various methods are being utilized in biomarker discovery. Single-cell genome studies identified that CXCL10 is a possible vital biomarker to predict PsA development. Patients who developed PsA had overexpression of baseline synovium and serum CXCL10 levels that decreased over time, whereas those with the cutaneous-limited disease did not.9-11 Proteomics are also being utilized. Cretu et al. identified many markers that are overexpressed in PsA skin and synovial fluid when comparing skin biopsies and synovial fluid.12 A challenge with serum proteomics is correcting for high dynamic range, which is targeted and can be expensive.13 There is also a recent focus on metabolomic fingerprinting. Dr. Chandran’s group discovered apparent differences when comparing psoriasis to PSA serum metabolites, but disease activity may explain much of the differences for these potential diagnostic markers.14 Response to treatment is clearly of interest in biomarker research. Using genotype markers in a large cohort, Dand et al. found that adalimumab had a better response in HLA-C*06:02 negative patients than ustekinumab, especially if they had PsA. The opposite was also true.15 Rahmati et al. observed clear gene expression profiles between IL-17 responders and non-responders using transcriptomic markers. They found that Rho-GTPase pathway was important and strongly signaled specific in IL-17, but not in TNF-α, response.16 In a study using immune cell phenotyping, Miyagawa et al. found that with strategic biologic treatment based on phenotypic differences in helper T cells (vs. usual practice), ACR20 response is significantly improved (85% vs. 55%).17 In summary, biomarkers have great potential to improve defined clinical outcomes in psoriasis. Though initial studies are promising, many clinically translational challenges still exist, and research remains in its infancy.
Alex Tsoi, PhD University of Michigan Ann Arbor, Michigan, United States Though many models in high throughput data have advanced biomedicine, data modeling still faces challenges as complex systems may only contribute a modest effect size from each feature.1 In this compelling lecture, Dr. Tsoi focused on three aspects of machine learning in psoriasis modeling: genetically regulated components of gene expression, risk of psoriatic arthritis, and drug responses. Modeling genetically regulated components of gene expression Identifying genome-wide association study (GWAS) signals is easier than interpreting the results. Many psoriasis-associated loci will not change the function of a protein but rather change expression levels for the genes they target. The objective is to identify causal genes from associated loci by modeling the genetically regulated component and then impute other genotyped individuals' transcriptome to be associated with the phenotype of interest. Modeling programs, therefore, prioritize genes that are likely to be causal for the phenotype.2 Once a model is created, you can apply it to independent datasets. With a larger sample size, the predictive performance matches the genetic heritability of the expression profiles with more refined confidence intervals. Dr. Tsoi’s group, for example, used expansive Genotype-Tissue Expression (GTEx) portal RNA-seq to improve predictive performance of their cohort, then used association analysis to reveal genes from 13 existing and five new psoriasis-associated loci.3 Modeling risk of PsA among psoriasis patients In addition to using genetic data to model profiles, potential clinical implications are vital. Dr. Tsoi’s team used machine learning to create a model that predicts which psoriasis patients will develop PsA. They performed a meta-analysis of PsA and cutaneous-limited psoriasis (PsC) vs. controls, then used GWAS to find new genome-wide significant loci for PsA or PsC. Their model showed 98% precision with 33% recall for the top 10% of patients predicted to develop PsA and 83% precision with 55% recall for the top 20%. Importantly, integrating genetic data allowed for dynamic predictions.4 Modeling drug response using genomic data In recent work, Dr. Tsoi’s team used cytokine signatures to characterize different skin types in psoriasis and atopic dermatitis. From existing transcriptome psoriatic lesional skin data, they implemented a self-organizing map to highlight heterogeneity among patients. Using RNA-seq, they found the baseline transcriptome expression levels of uninvolved skin has implication for anti-TNF PASI response.5 In unpublished work (currently under review) they observed impressive prediction results via cytokine-stimulated signatures to model PASI-75 response; for the top 20% of patients predicted to be PASI-75 responders, they displayed 80% accuracy in prediction. Challenges of machine learning Machine learning is not without challenges. The sample size is critical to provide robust information modeling for better parameter estimation and cross-validation. Reproducibility in different settings is crucial, and we need to continue associating molecular data with clinical response. A “multi-omics” (genome, proteome, transcriptome, etc.) integrative approach to refine models will continue to improve machine learning.
Cheryl Cameron, PhD Case Western Reserve University Cleveland, Ohio, United States
George Gondo, MA National Psoriasis Foundation Portland, Oregon, United States
IPC Councilor James T. Elder, MD, PhD University of Michigan, Ann Arbor, Michigan, United States
Wynnis Tom, MD University of California San Diego San Diego, California, United States
Sarah Whitley, MD, PhD University of Pittsburgh Pittsburgh, Pennsylvania, United States
Daisuke Yamada, MD PhD University of California Davis School of Medicine Sacramento, California, United States
Zhenrui Shi, MD PhD University of California Davis Department of Dermatology Sacramento, California, United States
Rahul Raiker, BS West Virginia University School of Medicine Morgantown, West Virginia, United States
Nicholas Theodosakis, MD, PhD Massachusetts General Hospital Department of Dermatology Boston, Massachusetts, United States
Anjana Sevagamoorthy MBBS, MPH University of Pennsylvania Philadelphia, Pennsylvania, United States
Dustin P. DeMeo Case Western Reserve University School of Medicine Cleveland, Ohio, United States
More Events
Latest news New IPC CouncilorJune 16, 2022Dr. Claus Johansen joins the group of IPC experts who are working to advance knowledge and enhance the care psoriasis patients around the world. PsoProtect Registry Data UpdateJune 15, 2022PsoProtect, a global registry for clinicians to report outcomes of COVID-19 in individuals with psoriasis, has updated their preliminary data and now includes the first 1,652 cases. Commentary: Biomarkers of Disease Progression in Psoriasis: Current State-of-the-ArtJune 13, 2022In personalized care for patients, psoriasis biomarkers will be highly relevant. At present, various centers have investigated biomarkers and their clinical significance for psoriasis progression. Read commentary from Andrea Chiricozzi, MD, on three recent publications. Arcutis Biotherapeutics, Inc. Announced Findings From the Skin Insights: Uncovering Psoriasis SurveyJune 02, 2022Arcutis Biotherapeutics, Inc. shared their finding from the Skin Insights: Uncovering Psoriasis Survey. More News