Engineering RNA Agonists For Systemic Immunity Without Systemic Toxicity

A conversation with Art Krieg, MD, Founder, President, and Acting CEO/CSO, Zola Therapeutics

As cancer immunotherapy continues to evolve, one of the field’s most persistent challenges remains how to activate powerful anti-tumor immune responses without triggering the systemic inflammatory toxicities that have limited many promising approaches. Continuing our conversation with Art Krieg, MD, founder, president, and acting CEO/CSO of Zola Therapeutics (see Part 1 of our conversation here), Krieg believes the answer lies in a deeper understanding of innate immune biology and the careful design of RNA/DNA hybrid mimics of viral RNA and DNA that are detected by innate immune receptors that evolved for the purpose of selectively promoting antiviral immunity.

Many immune agonists have demonstrated impressive biology but have been limited by systemic inflammatory toxicity. What lessons have the field learned?

One key lesson is that mouse models do not predict human safety or efficacy. I can’t think of a single immune agonist that has failed to show anti-tumor activity in mice, yet only a few have shown acceptable safety while also achieving response rates above the low single digits. Unfortunately, this has led many investors and pharmas to abandon the field, which I believe is drawing the wrong conclusion from the negative results.

I believe the correct lesson to draw from earlier failures is that we needed to go back to the basic science of these immune receptors and to study the species differences between rodents and larger animals, especially primates. We now recognize that in mice, a wide range of different inflammatory signals stimulate CD8+ T cells that can shrink tumors. In humans, many of the same signals actually can suppress these responses. A key immune factor that promotes CD8+ T cell responses in humans is IFN-a. So, one of the lessons we have adopted is to perform our early drug screening and selection using human immune cells, especially those derived from patients with cancer, using stimulation of IFN-a-secretion as a surrogate assay for the desired pattern of immune activation. We select against induction of inflammatory cytokines associated with toxicity and reported to be pro-tumorogenic in humans, such as IL-6, TNF-a, and IL-1b. As we went through candidate selection and screening in vitro and then in vivo, we were surprised and delighted to find that it was possible to induce unprecedented levels of IFN-a secretion without inducing a significant inflammatory response.

The conventional wisdom had been that these responses are inextricably linked — a new learning for us is how to disassociate them, providing greater efficacy without the toxicity historically associated with immune agonists.

Your platform combines TLR7, TLR8, and TLR9 biology. What advantages does this multi-receptor approach provide compared with targeting a single innate immune pathway?

TLR7, TLR8, and TLR9 are expressed in distinct populations of immune cells, each of which separately plays a role in activating immune responses leading to the desired CD8+ T cells. These receptors evolved for the purpose of detecting viruses, and CD8+ T cells evolved for the purpose of killing viral-infected cells without killing healthy cells. These TLRs are not expressed by CD8+ T themselves — they rely on separate populations of immune cells, especially dendritic cells, to be “instructed” what to do. If we think of the CD8+ T cells as the assassins of the immune system, because of their unique ability to selectively kill individual cells in our own body that have become infected or malignant, then the various populations of dendritic cells (DCs) serve as the generals, directing this response. Populations of conventional DC (cDC1 and cDC2) that express TLR7 or TLR8 are critical for “presenting” viral or tumor-associated antigens to the CD8+ T cells and directing them to kill any cell expressing these antigens. cDCs are supported closely by a subset of plasmacytoid DC (pDCs) that express TLR7 and/or TLR9 and serve as factories to produce massive quantities of IFN-a that amplify the desired CD8+ T cells and create an immune milieu conducive to their function, both locally and systemically. Finally, many populations of myeloid cells expressing TLR7 or TLR8 can regulate these responses positively or negatively. Tumors recruit such myeloid cells and program myeloid cells into immunosuppressive states (e.g., myeloid-derived suppressor cells or MDSCs) that promote tumor growth — RNA agonists for TLR7/8 can reprogram these cells into an antiviral gene program, supporting CD8+ T cell function and anti-tumor activity.

These immune suppressive myeloid cells do not express TLR9, and different subpopulations express TLR7 or TLR8. In our earlier clinical trials of a TLR9 agonist, vidutolimod, our biomarker studies revealed that patients with low baseline signatures for immunosuppressive myeloid cells generally responded to treatment, but patients with high signatures for myeloid cells did not. By engaging TLR7/8 in cDC and myeloid cells, and TLR7/9 in pDC, Z-007 is designed to overcome the major resistance mechanism that we identified in the clinical development of vidutolimod. These findings helped us understand that in order to increase the success rate for cancer immunotherapy, it is important to coordinately activate all of these disparate immune cell populations with an agonist inducing the gene expression programs that evolved to drive antiviral responses. We are harnessing these programs to fight tumors, essentially “tricking” the immune cells into attacking tumor antigens as if they were viral antigens. We believe that a TLR7/8/9 agonist can achieve this, if we deliver it into the tumor-associated immune cells.

The concept of a "synthetic viral mimic" is intriguing. Can you explain what that means biologically?

Viral particles contain proteins, lipids, RNA, and DNA (in the case of DNA viruses and retroviruses). These typically enter immune and other cells through endosomes, where they must be rapidly identified as of viral origin to trigger appropriate defenses. These viral structures are highly variable except for two critical features of the viral RNA and DNA that distinguish these from our own nucleic acids: 1) viral RNAs contain uridine, whereas the most abundant RNAs in our cells, rRNA and tRNA, have substitutions of pseudouridine for uridine; and 2) viral DNA contains unmethylated CpG motifs, whereas our own genomic DNA is ~80% CpG methylated. Evolution of multicellular organisms required protecting host genome integrity against DNA viruses and retroviruses seeking to copy their own genetic material into our cells. This general problem was solved by evolving TLR7 and TLR8, which have similar structures but independently recognize uridine-containing RNAs, and by evolving TLR9, which senses unmethylated CpG DNA in viral DNA without being activated by host DNA. Thus, we can make a synthetic viral mimic that is designed to activate these receptors efficiently, triggering the evolutionary pathways that evolved to rapidly induce the most powerful CD8+ T cells possible, without the need to actually understand all of the downstream steps involved. We have a minimal set of signals that tells the immune system what to do without triggering inflammation.

Delivery remains one of the biggest challenges in RNA therapeutics. How does lipid nanoparticle delivery influence both efficacy and safety in your approach?

Delivery is indeed the single biggest challenge in RNA therapeutics. One huge advantage we have over most other RNA therapeutics is that our TLR targets are in the endosomes. All RNA therapeutics enter cells via endosomes. Endosomal escape is one of the major rate-limiting steps for ASO, siRNA, mRNA, and crRNA therapeutics that act in the cytoplasm, with no more than 1%-5% of the cargo that is taken up reaching the functional compartment. All of our cargo that enters the endosomes is in our functional compartment, with no need for endosomal escape. Furthermore, the TLR-expressing immune cells we want to target have lipid uptake receptors on their surface, so specific targeting of the LNPs to other cell surface receptors is not required.

Following IV administration, the LNPs circulate throughout the body before rapid clearance into our target immune cells (as well as hepatocytes, endothelial, and other cells that do not express the TLRs, and so do not respond). Tumor-associated immune cells have increased lipid receptor expression and uptake, providing a kind of natural targeting for Z-007. This makes it possible for us to use tiny doses of our LNPs for therapy – in pet dogs with spontaneous tumors we have seen regression with IV Z-007 doses below 0.005 mg/kg (total dose 0.05 to 0.1 mg) weekly, in contrast to Onpattro, the siRNA therapeutic delivered using an LNP at a dose of 0.3 mg/kg every three weeks. Obviously, therapeutic index is much higher when your drug requires less than 1% of the LNP dose of other therapeutics.

In my last company we used a virus-like particle (VLP) for delivery of a CpG DNA TLR9 agonist. Although this was effective for local intratumoral injection, VLP delivery systems have the common drawback that they are highly immunogenic, inducing antibody responses that can reduce or eliminate the desired activity, which prevents systemic dosing. Switching our formulation to a non-immunogenic LNP avoids those issues, enabling IV dosing as well as other routes.  

Biomarkers appear to play an important role in your development strategy. Which pharmacodynamic signals are you monitoring most closely during development?

At a high level, we monitor two categories of biomarkers: markers present at baseline that are associated with the probability of response and response biomarkers that are induced (or repressed) in patients responding to therapy. Baseline biomarkers of greatest interest for us are transcriptional signatures or gene expression programs present in the untreated tumors. Studies involving thousands of patients with cancer whose tumors were profiled using RNA-seq have reported that certain baseline transcriptional signatures are associated with response across most or all tumor types. For example, signatures for type I and type II IFN activation are commonly associated with both response to immune checkpoint inhibitors (ICI) and survival. We have selected Z-007 in part for its ability to induce these genetic signatures in human tumor-associated immune cells ex vivo and in vivo in mouse tumor models and in healthy monkeys.

Once we begin clinical development of Z-007 and we have response data in patients, we will test for associations between the patients’ baseline biomarkers and outcomes, which will enable us to identify populations most likely to benefit from our treatment and potential resistance mechanisms.  

The second broad category of biomarkers of interest for us are those that are induced in patients in response to Z-007 therapy activating TLR7/8/9 in the target immune cells in vivo. In prior studies of TLR agonist therapy in humans, we and others have shown that serum CXCL10 is induced as early as two hours after injection, peaking around 12 to 24 hours, and then returning toward baseline. This is an important marker for us to confirm that our drug is getting to immune cells in vivo and activating the intended pathway. Moreover, in prior studies we found an association between the magnitude of the increase in serum CXCL10 and the probability of tumor shrinkage in both PD-1 blockade resistant melanoma and NSCLC. Therefore, in our initial human clinical trials we have set a target for the level of serum CXCL10 induction that we want to see. Following on from the early induction of serum CXCL10 are subsequent changes in the tumor microenvironment that we will monitor by biopsy to identify resistance mechanisms. In responding patients, we expect to see induction of the transcriptional signatures previously identified in patients responding to checkpoint inhibitors, which are induced by Z-007 ex vivo, and we expect to see activation of dendritic cells, myeloid cells, and infiltration of CD8+ T cells.

What distinguishes productive immune activation from excessive inflammatory activation, and how do you engineer around that balance?

A common misconception in cancer immunotherapy is that any immune stimulation is good, and that the goal should be to make “cold,” or uninflamed tumors “hot,” or inflamed. Although this generally works in mouse tumor models, approaches such as STING agonists that induce inflammation in the tumor rarely lead to patient benefit and in fact have been associated with severe toxicity, including death. Many leading investigators in cancer immunotherapy have reported evidence that classic inflammatory responses leading to cytokines such as IL-6 and IL-1b are actually pro-tumorogenic and therefore not productive.

Important insights have come from large unbiased studies of the biomarkers in patients with diverse tumor types that distinguish responders to checkpoint inhibitors from non-responders and that correlate with patient survival. Several independent research groups have published studies showing that gene expression signatures associated with type I and type II IFN are among the best positive predictors for outcome, suggesting that these can be used as surrogate assays to distinguish productive from nonproductive (inflammatory) immune activation.

A new discovery for us is how to design TLR7/8/9 agonists and the LNP delivery system in such a way as to trigger the productive immune activation (measured primarily by secretion of IFN-a and induction of published gene signatures associated with human clinical response and survival) without the undesired inflammatory response (measured primarily by secretion of IL-6, TNF-a, and IL-1b and induction of inflammatory gene signatures associated with treatment resistance and worse survival in patients with cancer).  

From a manufacturing and CMC perspective, what challenges are unique to developing RNA-based immune agonists compared with other RNA medicines?

From a manufacturing and CMC perspective, we can consider two classes of RNA medicines: those that can be delivered naked and those that require a delivery system to protect them against degradation by serum and tissue nucleases. RNA medicines such as siRNA are synthesized with a fully modified and stabilized backbone, so these can be delivered “naked,” conjugated to ligands for cell surface receptors on hepatocytes or other target cells. This is possible because the RISC complex that mediates RNA interference easily tolerates modifications protecting the siRNA from degradation. These complete modifications are not tolerated by TLRs, ribosomes, or CRISPR, and so our TLR agonists and mRNA and crRNA medicines require native RNA backbones that are sensitive to nucleases and therefore must be protected using LNP, viral, or other delivery systems. Fortunately, our unmodified RNA and DNA is easily synthesized using standard automated processes developed for other synthetic RNA medicines, and decades of LNP development have made LNP formulation relatively routine, especially since we do not need the latest ionizable lipids developed to improve endosomal escape.

About The Expert

Art Krieg, MD, founded Zola Therapeutics in 2023 and serves as CEO. Previously, Krieg founded Checkmate Pharmaceuticals, until its acquisition by Regeneron in 2022; RaNA Therapeutics (subsequently Translate Bio, acquired by Sanofi); and Coley Pharmaceutical Group until its acquisition by Pfizer in 2008. Krieg served as CSO of Pfizer’s Oligonucleotide Therapeutics Unit from 2008 to 2011 and cofounded the first antisense journal, Nucleic Acid Therapeutics and the Oligonucleotide Therapeutics Society, for which he recently served as president. Krieg has published more than 250 scientific papers and is an inventor on over 50 U.S.-issued patents covering oligonucleotide technologies. He is currently an adjunct professor in the UMass Chan Medical School RNA Therapeutics Institute.