Background/Purpose: Psoriasis is a chronic IL-23/IL-17-driven inflammatory disease. Yet, despite highly effective targeted biologics, lesions commonly recur after treatment withdrawal, and systemic manifestations such as psoriatic arthritis remain incompletely prevented. We hypothesized that durable resolution is not simply suppression of a single inflammatory pathway, but a continuous, coordinated molecular healing process in which disease programs are progressively lost, and tissue repair programs are restored – features that discrete clinical or timepoint-based comparisons may miss. To this end, we developed PsoriAxis, a computational transcriptomic scoring framework that quantifies where individual skin samples lie along a disease-to-resolution axis. 

Methods: We integrated public, longitudinal psoriasis skin bulk transcriptomic datasets, including pre/post anti-IL-17/IL-17R cohorts, using TimeAx-based alignment to construct a shared pseudotime healing axis. Each sample received a PsoriAxis score quantifying molecular response. We evaluated tissue-state ordering, projection of independent datasets, associations with clinical response, and hidden molecular response states. Data-driven co-expression modules were grouped by kinetic behavior to map healing cascades at the molecular level (Figure 1).

Results: Across discovery and independent datasets, PsoriAxis ordered baseline lesional, treated lesional, and healthy-like samples along an expected disease-to-resolution continuum. Early PsoriAxis position stratified subsequent PASI75 response, with future non-responders showing lower molecular healing scores across treatments. PsoriAxis also revealed hidden response states within clinical categories: responders ranged from healthy-like molecular resolution to residual disease-proximal states, while non-responders ranged from molecular non-progression to partial healing (Figure 2).  Thus, axis position provided a continuous estimate of residual, molecular disease beyond binary clinical response labels.

Co-expression analysis identified 240 bootstrap-stable modules grouped into nine kinetic archetypes, recovering known psoriasis biology while highlighting underexplored programs whose coordination normalized, persisted, or increased along the axis. These kinetics separated early-collapsing inflammatory coordination from persistent residual disease-associated structure and later healing-associated programs involving barrier, lipid, metabolic, basal-attachment, stromal, and macrophage-niche restoration. By defining which programs remain active at each trajectory position, PsoriAxis prioritizes residual disease programs for future durable-response studies (Figure 3).

Conclusion: PsoriAxis defines a continuous molecular healing axis that generalizes across psoriasis transcriptomic datasets, stratifies later clinical response, and reveals hidden response states beyond PASI. By distinguishing clinical improvement from molecular resolution, this framework maps residual disease and repair programs along the healing trajectory, providing a potential basis for future studies of durable response across chronic inflammatory diseases.

Figure 1. PsoriAxis defines a continuous molecular healing axis in psoriasis.

(A) PsoriAxis was constructed from public longitudinal psoriasis skin bulk transcriptomic datasets comprising 530 samples from two cohorts [PMID: 31883845; PMID: 36690571], including anti-IL-17/IL-17R pre/post-treatment lesional and non-lesional samples collected across baseline, week 4, week 12, and week 52. Using our TimeAx framework [Frishberg et al., Nature Communications, 2023], transcriptomes were aligned to generate a continuous PsoriAxis score that places each sample along a disease-to-resolution axis. (B) PsoriAxis orders samples by molecular healing state rather than by patient, tissue label, or treatment timepoint, spanning disease-like/lesional, improving, and healthy-like/resolved states. This enables residual molecular disease to be defined by trajectory position, including samples that may appear clinically improved but remain disease-proximal. (C) Gene programs vary dynamically along PsoriAxis pseudotime, capturing coordinated loss of inflammatory programs, persistence of residual disease signals, and emergence of tissue repair/restoration programs during molecular healing.

Figure 2. PsoriAxis captures molecular healing states and stratifies subsequent clinical response.

(A) In the brodalumab discovery dataset GSE117468, PsoriAxis ordered lesional baseline, treated lesional, and non-lesional samples along the expected disease-to-resolution continuum. Paired Wilcoxon signed-rank tests with Benjamini-Hochberg correction showed significant shifts from lesional baseline to week 4 and week 12 (both adjusted p<0.001) and strong separation between baseline lesional and baseline non-lesional skin (adjusted p<0.001). Baseline non-lesional and week 12 non-lesional samples were not significantly different, although week 12 non-lesional samples trended toward higher PsoriAxis scores.

(B) Among week 12 lesional samples from GSE117468, PsoriAxis scores stratified PASI75 response depth. PASI75 responders showed higher PsoriAxis scores than non-responders by Wilcoxon rank-sum test (median Δ=0.398; adjusted p<0.001), with PASI90+ samples tending toward more resolved molecular states. Notably, clinical responders spanned a broad range of PsoriAxis scores, indicating variable depth of molecular response beyond the PASI75 label; non-responders also separated into molecularly distinct groups, although sample size was limited.

(C) Independent projection of the GSE136757 TYK2/JAK1 inhibitor (PF-06700841) dataset, together with healthy reference samples from GSE32407, reproduced the expected molecular healing continuum. Placebo-treated lesions remained baseline-like (Wilcoxon rank-sum adjusted p=0.288), whereas active-treatment lesions shifted significantly by week 2 and week 4 (both adjusted p<0.001). Baseline non-lesional samples aligned with healthy reference skin (adjusted p=0.262), supporting the healthy-like anchor.

(D) Early lesional PsoriAxis scores stratified later PASI75 response across treatment datasets. In the brodalumab cohort, week 4 PsoriAxis scores stratified week 12 PASI75 response across all doses (median Δ=0.256; Wilcoxon rank-sum p=0.0011; AUC=0.866) and in the 210 mg Q2W subgroup (median Δ=0.215; adjusted p=0.039; AUC=0.889). In the TYK2/JAK1 inhibitor PF-06700841 cohort, week 2 PsoriAxis scores showed a supportive trend for stratifying week 4 PASI75 response (median Δ=0.432; p=0.064; AUC=0.889), limited by small sample size.

Figure 3. PsoriAxis resolves coordinated molecular cascades during psoriasis healing.

(A) Heatmap of 240 bootstrap-stable data-driven co-expression modules clustered by connectivity dynamics across PsoriAxis sliding windows, from baseline lesional to late healthy-like/non-lesional states, with healthy reference samples shown separately. Ward.D2 clustering identified nine kinetic clusters.

(B) Representative cluster centroids illustrate three major kinetic behaviors: Cluster 5, an early-collapsing disease-associated pattern; Cluster 4, an early-high pattern with partial resolution and residual coordination; and Cluster 8, a broad healing-increasing pattern that rises during late repair/restoration. Colored trajectories show observed cluster centroids, light grey trajectories show size-matched random-module centroids, and detached points at right show healthy reference connectivity.

(C) Representative data-driven modules and curated prior-knowledge signatures are plotted as connectivity above matched random background across PsoriAxis pseudotime, with healthy reference values shown separately at right. Solid lines denote data-driven modules and dotted lines denote prior-knowledge signatures. The selected modules span three biological phases: early-resolving inflammatory effector programs, including acute IL-17/IL-1β/IL-20 myeloid cytokine (M94/C5) and CXCL2/CXCL6 neutrophil chemotaxis (M617/C5); a partially resolving type-I IFN antiviral ISG program (M20/C4) that remains coordinated through intermediate healing states; and a cornified-envelope keratinocyte barrier program (M19/C8) that progressively restores during late healing. The alignment between data-driven modules and literature-based signatures supports the biological validity of PsoriAxis and shows that the trajectory orders known psoriasis biology into a molecular healing cascade.

(D) Over-representation analysis of the partially resolving type-I IFN antiviral ISG module (M20/C4) highlights interferon-α/β signaling and canonical ISG programs, supporting its annotation as a residual immune-coordination module.

Unlike early-collapsing downstream inflammatory effector modules, M20/C4 approaches healthy-like coordination only late along PsoriAxis, suggesting that clinical response may suppress downstream inflammation before fully resolving upstream immune-priming or residual coordination programs. Together, these analyses illustrate how PsoriAxis maps programs lost early during treatment, residual programs that remain along the trajectory, and healing-associated modules restored during molecular resolution, providing a basis for future studies of durable response.

ACR Meeting Abstracts - https://acrabstracts.org/abstract/psoriaxis-reveals-continuous-molecular-healing-and-residual-disease-states-in-psoriasis/