AI discovery of TLR agonist-driven phenotypes reveals unique features of peripheral cells from healthy donors and ART-suppressed people living with HIV.

Abstract

Background: Selective and potent Toll-like receptor (TLR) agonists are currently under evaluation in preclinical models and clinical studies to understand how the innate immune system can be harnessed for therapeutic potential. These molecules are designed to modulate innate and adaptive immune responses, making them promising therapeutic candidates for treating diseases such as cancer or chronic viral infections. Much is known about the expression and signaling of TLRs which varies based on cell type, cellular localization, and tissue distribution. However, the downstream effects of different TLR agonists on cellular populations and phenotypes are not well understood. This study aimed to investigate the impact of TLR pathway stimulation on peripheral blood mononuclear cell (PBMC) cultures from people living with HIV (PLWH) and healthy donors.

Methods: The effects of TLR4, TLR7, TLR7/8, TLR8 and TLR9 agonists were evaluated on cytokine production, cell population frequencies, and morphological characteristics of PBMC cultures over time. Changes in the proportions of different cell populations in blood and morphological features were assessed using high-content imaging and analyzed using an AI-driven approach.

Results: TLR4 and TLR8 agonists promoted a compositional shift and accumulation of small round (lymphocyte-like) PBMCs, whereas TLR9 agonists led to an accumulation of large round (myeloid-like) PBMCs. A related increase was observed in markers of cell death, most prominently with TLR4 and TLR8 agonists. All TLR agonists were shown to promote some features associated with cellular migration. Furthermore, a comparison of TLR agonist responses in healthy and HIV-positive PBMCs revealed pronounced differences in cytokine/chemokine responses and morphological cellular features. Most notably, higher actin contraction and nuclear fragmentation was observed in response to TLR4, TLR7, TLR7/8 and TLR9 agonists for antiretroviral therapy (ART)-suppressed PLWH versus healthy PBMCs.

Conclusions: These data suggest that machine learning, combined with cell imaging and cytokine quantification, can be used to better understand the cytological and soluble immune responses following treatments with immunomodulatory agents in vitro. In addition, comparisons of these responses between disease states are possible with the appropriate patient samples.