(P 229) EXPRESSION OF TRANSPOSABLE ELEMENTS AND IKZF1 PREDICTS IMMUNE INFILTRATES IN SARCOMAS AND CORRELATES WITH OUTCOMES AFTER IMMUNE CHECKPOINT INHIBITION

Objective: In sarcomas, response rates to immune checkpoint inhibition (ICI) are overall low, variable, and difficult to predict. The types and degree of baseline immune infiltrates correlate with ICI response, but the underlying biology driving differences in tumor immune populations is not well understood. One potential tumor-intrinsic mechanism is epigenetic de-repression of transposable elements (TEs), which are derived from antient viral sequences embedded in our genomes. Aberrant expression of TEs can activate inflammatory signaling and could therefore influence tumor immune infiltrates. We previously disclosed preliminary findings from a transcriptomic analysis of 67 pre-treatment biopsies of sarcomas in patients treated on ICI trials. We found that expression of TEs and a transcription factor and epigenetic regulator, IKZF1, predicted immune-hot status and correlated with inflammatory pathway expression. Here, we update these findings to include analysis of additional cohorts, sarcoma subtype-specific data, on-treatment biopsies, and additional clinical outcome measures

Methods: We retrospectively analyzed RNA-seq data from our initial cohort of 67 sarcoma patients representing 12 subtypes who were treated on ICI clinical trials (pembrolizumab plus talimogene laherparepvec, nivolumab plus bempegaldesleukin, and pembrolizumab plus epacadostat) including 46 paired pre- and on-treatment biopsies. In addition, we analyzed transcriptomic data collected by the TCGA for a cohort of 190 sarcoma patients representing 5 subtypes (available via dbGAP) and from the SARC028 trial (pembrolizumab) (obtained through agreement with SARC), which was available for 30 patients representing 7 subtypes. MCP counter was used to describe immune infiltrates, after correcting for batch effects, which were then used to cluster patients into distinct immune-hot and immune-cold groups using a hierarchical clustering of principal components approach. TE and gene expression were quantified using the REdiscoverTE package. Lasso logistic regression including variables of TE and epigenetic regulator expression, sarcoma subtype, batch, and tertiary lymphoid structures (TLS; defined by an RNA-seq score), which is a known predictor of ICI response, was performed to identify predictors of immune infiltrates. A TE score was derived from the expression of TEs that significantly contribute to the regression model. Survival outcomes were analyzed using a Cox regression that included tumor subtype and tumor size (where available) as variables.

Results: An initial analysis revealed that immune infiltrates correlated with improved overall response rate in the immune hot (30%) compared to the immune-cold group (8.1%) (p=0.02, Fisher’s exact test). An updated model including TLS confirmed TE (r2=0.27) and epigenetic gene expression (r2=0.30) as the best predictors of immune status. High TE score and IKZF1 expression correlated with progression free survival (PFS) (TE score: high [4.4 months] vs. low [1.8 months], p=1.65x10-3; IKZF1: high [5.3 months] vs. low [1.8 months], p=9.28x10-3). Analysis of 46 paired pre- and on-treatment biopsies demonstrated that a shift from immune cold to hot corresponded with increased TE score (p=7.38 x 10-4) and IKZF1 expression (p=1.73 x 10-5). The opposite occurred in tumors that switched from hot to cold; TE score (p=0.010), IKZF1 expression (p=0.05); paired t-test. In 190 sarcoma patients from the TCGA, high expression of TEs and IKZF1 predicted immune hot status (controlling for subtype), expression of immune pathways such as type I interferon (p< 0.001) and antigen processing machinery (p< 0.001) and correlated with overall survival (OS) (TE score, p=1.26 x 10-3; IKZF1, p=4.94 x 10-3). A multivariate analysis within individual subtypes revealed TE score as significantly associated with OS (leiomyosarcoma, LMS; p=0.01, dedifferentiated liposarcoma, DDLPS; p=0.01, undifferentiated pleomorphic sarcoma, UPS; p=0.01). Finally, within a subset of the SARC028 cohort for which RNA-seq data was available, IKZF1 expression predicted immune type (r2=0.36) and was associated with improved PFS (p=0.05), after correcting for subtype. IKFZ1 expression correlated with immune pathways such as cGAS-STING (r2=0.76, p=1.45x10-6) and antigen-processing machinery (r2=0.53, p=2.91x10-3). In the SARC028 cohort, transcriptomic data was only available for three responding patients and the method for RNA-sequencing did not capture sufficient TE transcripts, thus precluding analysis of response and TE expression.

Conclusion: We identify tumor-intrinsic features that contribute to tumor immune phenotypes, focusing on expression of epigenetic regulators and TEs, which are normally epigenetically silenced and can stimulate inflammatory signaling. In an initial heterogeneous cohort of sarcoma patients treated with ICI, we found that TE and IKZF1 expression predicted immune infiltrates, correlated with PFS, expression of inflammatory pathways, and increased in tumors which gained immune infiltrates during ICI treatment. In a second cohort from the TCGA, TE and IKZF1 expression again predicted immune infiltrates when controlling for subtype, correlated with inflammatory signaling pathway expression, and with OS. Within the individual subtypes of LMS, DDLPS, and UPS, TE expression correlated with OS. In a third cohort of patients treated with pembrolizumab on SARC028, IKZF1 predicted immune hot status when accounting for subtype and correlated with PFS and immune signaling; correlation with response and TE expression was not evaluable. Overall, these updated findings support further exploration of TE and IKZF1 expression as potential subtype agnostic predictive and prognostic biomarkers and suggest these factors may influence tumor immune infiltrates. Pharmacologic strategies using epigenetic drugs to de-repress TEs could be explored as a strategy to enhance tumor immune infiltration in sarcomas.