Cyclic peptides in the beyond rule of 5 space are increasingly gaining traction as a therapeutic platform, and their astonishing passive permeability and ability to bind difficult or “undruggable” targets is continually studied, developed, and exploited as potent lead drug compounds. RNA, as a cellular information carrier and gene regulator, has been implicated in many human diseases. The therapeutic potential of targeting RNA with drug-like molecules is being increasingly recognized and explored, but the physicochemical properties required to selectively target RNA from a membrane-permeable, therapeutically relevant paradigm remain unprobed, especially important considering existing efforts have yielded compounds that are largely antithetical to membrane permeability. We investigated the possibility of lipophilic, cell-permeable cyclic peptides binding RNA with high affinity and specificity. Chapter 1 describes the design and deployment of two choice cyclic peptides against a library of randomized RNAs to yield an RNA aptamer to each cyclic peptide that binds with high affinity and specificity. Sequence level apo and holo chemical probing efforts reveal crucial differences between the two complexes.

In a collaborative effort with the Theodore Holman lab (UC Santa Cruz), chapter 2 describes attempts towards the medicinal chemistry optimization of SLUG01, a derivative of a ML355, a previously reported human 12-Lipoxygenase inhibitor.

As large language models have greatly increased in capability in recent years, it becomes increasingly important to improve our ability to exert control over their outputs. In this thesis, I discuss several such control schemes I have developed, ranging from pure inference-time control to finetuning-based alignment methods. I will first discuss highly general methods that apply to unstructured natural language generation, including both an inference-time control scheme called FUDGE as well as a reinforcement-learning based finetuning approach called RLCD. I will next discuss more specialized methods that can be used for control in more structured domains such as molecule design, program synthesis, and semantic parsing. Finally, I will show how many of these ideas can be used in conjunction with structured planning via prompting to extend our control to much longer outputs—in the range of thousands of words—in an automatic story generation application.

Learning and memory-related behaviors in mammals depend on prototypical GPCR signaling and subsequent changes in gene expression. Immediate early gene (IEG) expression is necessary for memory and is induced by increased nuclear cyclic adenosine monophosphate (cAMP) signaling. However, the downstream mechanisms mediating this process have remained unclear. Here, we demonstrate in detail a surprising mechanism by which the G-protein coupled receptor (GPCR) β2 Adrenergic Receptor (β2AR) under stimulation indirectly facilitates nuclear cAMP signaling via sequestration of a phosphodiesterase (PDE)4D5/β-arrestin complex. By combining compartment localized cAMP biosensors, Förster resonance energy transfer (FRET)-based live cell imaging, immunofluorescent imaging, and transgenic mice, we found the following:• Deletion of G-protein receptor kinase (GRK) phosphorylation sites on β2AR (GRKΔ) blocked agonist-induced receptor internalization to the endosomal compartment. • This loss of GRK/arrestin mediated internalization also reduced nuclear cAMP in hippocampal neurons. • Loss of nuclear cAMP signaling impaired nuclear PKA-mediated immediate early genes (IEGs) expression in neurons. • In animals, loss of GRK phosphorylation resulted impairment of long-term memory in a Morris water maze, a deficit in learning induced IEG expression, but intact working memory. • In wildtype neurons, β2AR stimulation promoted internalization of the receptor to the endosome, and β-arrestin-dependent recruitment of cAMP-degrading PDE4D5 to the internalized receptor. • Intriguingly, inhibition of β-arrestin-PDE4D recruitment alone prevented β2AR-dependent increases in nuclear cAMP signaling in neurons and PDE4D5 recruitment. • Furthermore, direct PDE4 inhibition was sufficient to rescue the β2AR-dependent nuclear cAMP signal in vitro, and in vivo ameliorated the long-term memory deficits of GRKΔ mice. • PDE4D5 but not other PDE isoforms were found enriched in the nucleus of neurons and stimulation of β2AR causes the movement of PDE4D5 out of the nucleus into the cytosol. • Additionally, Stimulation with other GPCR agonists also caused ligand dependent trafficking of PDE4D5, suggesting the proposed mechanism may have implications for other Gs-coupled • Receptors. Additionally, we found removal of GRK phosphorylation sites on β2AR at serines 355/356 or knockout of β2AR resulted in aberrant anxiety response in the elevated plus Maze (EPM). • Furthermore, we show that knockout of β1AR results in a loss of both anxiety response in the EPM and acute stress response in a force swim test (FST). This work therefore indicates learning and memory relies critically upon the endosomal, GRK-phosphorylated β2AR sequestration of a β-arrestin/PDE4D complex, which indirectly facilitates nuclear cAMP signaling by effectively removing a PDE4D blockade of cAMP signaling from the nucleus. Therefore, these data constitute a novel and major mechanism by which learning and memory-related behaviors in mammals are regulated by GPCR signaling, at the endosome as well as suggest regulation of stress or memory controlled by different β-adrenergic subtype. These data also suggest differential βAR subtype regulation of stress response.