RNAi Screening and Immunotherapy Target Discovery – Truly Translational Research Is Hard

We are constantly told that target ‘X’ is a novel checkpoint inhibitor, and then target ‘Y’, and don’t forget ‘Z’! We hold such claims to a high standard of validation, but press releases often outrun the science. So let’s look at some recent papers in this light.

RNAi screening has been around for several decades and yet like its’ distant cousin, transcript profiling (Txp), it is a technology that only rarely produces actionable, translational, data. Large scale screening of transcription whether using positive (Txp) or negative (RNAi) readouts can, in the wrong hands, produce a lot of false hits. Two recent papers approach this problem carefully. These are both RNAi based screens that use constrained pools and built in redundancy to limit false positives.

Last year an elegantly crafted paper appeared in Nature. The paper was a local tour-de-force with contributions from Dana Farber, the Koch and Whitehead Institutes at MIT, the Broad Institute (MIT and Harvard), and the Novartis Inc Biomedical Research and Genomics Institutes.

Here is the reference:

In vivo Discovery of Immunotherapy Targets in the Tumor Microenvironment. 2014. Nature 506: 52–57. doi:10.1038/nature12988.

Why highlight this paper, nearly a year later? We’ll get to that in a bit. First let’s take a look at the data. The paper poses a provocative question: can novel regulatory switches controlling T cell function in immunosuppressive tumors be identified in vivo?

The background to the work is laid out in the introduction, summarized here:

1) Cytotoxic T cells play a central role in immune-mediated control of cancer because they specifically detect and eliminate cancer cells following TCR-mediated recognition of tumor-derived peptides bound to MHC.

2) Infiltration of both the tumor center and the invasive tumor margin by CD8+ cytotoxic T cells correlates with a favorable prognosis regardless of the extent of tumor invasion and local lymph node involvement.

3) In the majority of patient tumors this anti-tumor immune defense is blocked by immunosuppressive cell populations recruited to the tumor microenvironment, including regulatory T cells, immature myeloid cell populations and tumor-associated macrophages.

4) Highly complex interactions among a variety of different cell types – e.g. tumor cells, immune cells and stromal cells – in the tumor microenvironment contribute to clinical outcome.

5) Such complex cellular interactions are best modeled in vivo.

To tackle this question the authors devised a screen using in vivo pooled short-hairpin RNA (shRNA), shRNAs are artificial RNA molecules containing a tight hairpin turn that silences target gene expression via RNA interference (RNAi). A lentiviral vector is used to target the construct to genomic DNA so that it is integrated.

Two screens were devised. One focused on genes over-expressed in anergic or exhausted T cells. 255 genes were represented by an average of 5 independent shRNAs each. This screen was further split into 2 different pools. The second screen comprised shRNAs targeting kinases and phosphatases, again with an average of 5 shRNAs per target. Thus these pools were both constrained in size and armed with built-in redundancy.

The technique takes advantage of the extensive proliferative capacity of T cells following TCR triggering of the TCR by a tumor-associated antigen. shRNAs capable of restoring CD8 T cell function by targeting negative regulators may be uncovered because they will be over-represented in the T cell compartment upon expansion. The shRNAs within a pool can be quantified by deep sequencing of the shRNA cassette from tumors and secondary lymphoid organs as compared to control tissues.

The relatively simple experimental design was to collect naive T cells from the mice that matched the mice used for the melanoma modeling. The T cells were pretreated for two days with IL-7 and IL-15 prior to lentiviral vector-mediated infection with shRNA pools. The transduced T cells were injected into B6 mice bearing day 14 B16-Ova tumors. After 7 days, T cells were purified from tumors and secondary lymphoid organs (spleen, tumor-draining and irrelevant lymph nodes) for isolation of genomic DNA, followed by PCR amplification of the shRNA cassette. The representation of shRNAs was then quantified in different tissues by Illumina sequencing.

The analysis focused on genes whose shRNAs were over-represented in tumor samples but not spleen, a secondary lymphoid organ, or other organs. Substantial T cell accumulation in tumors was observed for a number of shRNAs, despite the immunosuppressive environment. These shRNAs represent putative novel immune modulatory pathways active in tumor setting.

In this paper the authors highlight two results with the highest degree of specific enrichment in T cell populations isolated from tumors: Cblb (an E3 ubiquitin ligase that induces T cell receptor internalization) and Ppp2r2d (a phosphatase, not previously studied in T cells). The most enriched shRNAs from this study are shown here (Figure 2 in the paper):

 screen 1

 Note that the statistical significance shown (*: P<0.05; **: P<0.01) is in comparison to control LacZ, and the first 7-10 hits are probably statistically indistinguishable although I didn’t check. Regardless the phosphatase Ppp2r2d sits to the far left on the graph, with a whopping 16x increase in T cells isolated from tumors versus those isolated from spleen.

The authors proceed to confirm the Ppp2r2d hit and explore its’ cellular mechanism of action. This phosphatase pathway controls T cell proliferation and cytokine expression, and its’ blockade by shRNA leads to the accumulation of tumor responsive IFNg-secreting CD8+ T cells. PP2A proteins are a family of phosphatase complexes with catalytc, scaffolding and regulatory subunits. Ppp2r2d is a regulatory subunit that functions by controlling cellular localization and substrate. Specifically, Ppp2r2d directs PP2A to Cdk1 substrates during cell division to inhibit mitotic entry and induce mitotic exit. This is therefore a mechanism underlying the control of T cell proliferation. PP2A also regulates cell apoptosis via BAD-induced cell death and PP2A phosphatases have been shown to interact with the cytoplasmic domains of CD28, CTLA-4 and the NF-κB regulator Carma1. It is not known which regulatory subunits are required for these activities although one might hypothesize a role for Ppp2r2d.

Other interesting hits include Arhgap5, a RHO GTPase previously shown to be important in T cell responses after TCR engagement; SMAD2, a signaling molecule downstream of the highly immunosuppressive TGF-b receptor, and Cbl-b, an E3 ligase that controls TCR responsiveness. Cbl-b is the subject of a recent review (Front Oncol. 2015, 11:58. doi: 10.3389/fonc.2015.00058). In T cells, Cbl-b negatively regulates activation signals through TCR or pattern recognition receptors, and its’ activity is regulated in response to TGF-b signaling (we might refer back to SMAD2 here). cblb-gene-deficient mice spontaneously develop autoimmunity and are highly susceptible to experimental autoimmunity and gene association studies have linked Cbl-b with several human autoimmune diseases. On the other hand cblb knockout CD8(+) T cells are hyper-responsive to TCR and CD28 stimulation and are in part protected against immunosuppression induced by TGF-β, at least in vitro. In vivo, cblb-gene deficiency contributes to tumor rejection due to highly active CD8 T cell and NK cell activity. The role of cbl-b in NK cell anti-tumor responses has received considerable attention. In a paper by Penninger and colleagues it was shown that the innate immune TAM tyrosine kinase receptors (Tyro3, Axl and Mer) are ubiquitylation substrates for Cbl-b. Targeting cblb allowed increased NK cell activation mediated by these receptors, and has led to increased appreciation of the TAMs as potential targets in immunotherapy. Finally, and this is a consistent theme in the immune checkpoint space, Cbl-b activity has also been associated with anti-viral immunity and T cell exhaustion due to chronic activation.

While there is also interest in targeting Cbl-b as a strategy to enhance anti-cancer immunity, we note that Cbl-b is expressed in all leukocyte subsets and regulates multiple signaling pathways in T cells, NK cells, B cells, myeloid cells and dendritic cells. Thus targeting this widely expressed protein may require specific targeting techniques. Apeiron targets Cbl-b in PBMCs using siRNA electroporation – essentially boosting the immune competence of these cells ex vivo, and then putting them back in the patient. There are several POC Cbl-b inhibitor compounds in academic labs whose use thus far appears limited to rodent studies.

Lots of pathways present themselves as interesting drug development targets. A classic example of such a pathway would include a cell surface signaling receptor, the receptor tyrosine kinase, and the downstream intracellular tyrosine kinases. We have lots of examples: VEGFR, EGFR and the like. Prosecution of pathogenic pathways keeps expanding however, to include ever more diverse targets. This brings us to the protein phosphatases, and the reason we’ve returned to the nearly year-old Nature paper highlighted in Part 1.

I’ve highlighted the kinases as an exemplary drug-targeting pathway in part because their mechanism of action is the exact opposite of the phosphatases. Kinases add phosphate groups to specific segments of proteins that are intracellular; phosphatases strip these same phosphate groups away. Since phosphorylation is a critical component of cell signaling cascades leading to gene transcription, regulation of the process is critical to maintain cellular homeostasis and also allow rapid cellular responses. Thus in T cells phosphorylation of ZAP70 immediately upon TCR engagement is a canonical first step in T cell activation. Dephosphorylation of ZAP70 is accomplished by the low molecular weight phosphatase among many other mechanisms. Of note downstream mediators of TCR signaling are independently regulated. As just one example, CTLA4 utilizes PP2A and the SHP phosphatases to inhibit AKT phosphorylation while PD-1 acts via Src homology-2 (SH2) domain-containing phosphatases to block PI3-K activation, among other targets. Targeting phosphatases specifically has been very difficult – these proteins perform diverse functions and are downstream of many different signals.

A recent paper demonstrates the feasibility of targeting specific PP2A regulatory subunits, and this is why we have returned to the shRNA screening paper a year later.

Here is the reference:

Preventing proteostasis diseases by selective inhibition of a phosphatase regulatory subunit. 2015. 348: 239-242. doi: 10.1126/science.aaa4484.

In this paper the authors tackle a different PP2A pathway, and describe a compound identified as Sephin1 (selective inhibitor of a holophosphatase), a small molecule inhibitor of Ppp1r15a. This is a simple molecule indeed:


 While likely lacking in traditional drug-like development properties, the compound was orally available and concentrated specifically in the CNS while disappearing rapidly from plasma circulation – in this case a beneficial profile, as the targeted pathology models result from derangement of the mis-folded protein response in the CNS, a cellular process regulated by PP2A activities. Specifically, Ppp1r15b acts to prolong a beneficial adaptive phospho-signaling pathway that protects cells from otherwise lethal protein misfolding stress. This cellular biology is counter-regulated by the stress-induced regulatory subunit Ppp1r15a. The trick then was to inhibit Ppp1r15a specifically, without inhibiting the related protein Ppp1r15b. Serphin1 safely prevented CNS motor, morphological, and molecular defects of two distinct protein-misfolding diseases in mice, Charcot-Marie-Tooth 1B, and amyotrophic lateral sclerosis. Thus this paper demonstrated that 2 closely related regulatory subunits of phosphatases could be the target of successful proof-of-concept drug discovery.

The second RNAi screening paper leads us to a more tractable target. In this paper, the screening is done in vitro, using a tumor cell line. Such experiments are notorious for producing results that cannot be reproduced or confirmed, so the authors step carefully, building in layers of controls, as we’ll see.

Here is the reference:

A high-throughput RNAi screen for detection of immune-checkpoint molecules that mediate tumor resistance to cytotoxic T lymphocytes. 2015. EMBO Mol Med 7: 450–463. doi: 10.15252/emmm.201404414).

The paper is from the Beckhove lab in Heidelberg – they’ve done a lot of this type of work. The screen utilizes MCF7 breast cancer cell lysis as the read-out. Cell lysis was triggered by either survivin-specific activated T cells or T cell engaging bi-specific antibodies that cross-linked CD3 on activated T cells to EpCAM on tumor cells (CD3 x EpCAM bispecific). MCF7 (luciferase+) cells were transfected with siRNAs then cocultured with CD8+ T cells. The luciferase serves as a marker of tumor cell death. The screen was focused on a library of 520 genes coding for transmembrane and cell surface proteins, with the goal of identifying antibody targets. Controls included scrambled RNAi constructs (negative controls) and PD-L1, Ceacam-6 and Galectin-3 (positive controls). Furthermore, the screen was run 3 times, independently.

Here is a snapshot of the reported results.

 screen 2

 The hits uncovered using this approach included CCR9, a chemokine receptor widely reported to be important in tumor cell resistance and metastasis. The authors trace the activity in the assay from the tumor-specific knockdown of CCR9 to increased activation of STAT signaling in T cells, by an unknown mechanism. This resulted in changes in gene expression in those T cells that correlated with T cell effector functions including increased production of IFN-g, IL-2, and TNF). The authors report they have generated experimental validation for GLIPR1L1 (poorly known) and GHSR (a hypermethylated gene in some cancers) in addition to CCR9. On the flip side, hits such as Frizzled D3 (FZD3) may or may not have translational relevance, as biology relevant to the experimental system is not readily deconstructed. This lab has done similar work in pancreatic cancer and will present results next week on a gene/protein that is unknown to me (TONI1, AACR abstract #245).

These papers present interesting and careful approaches to the use of RNA-targeting screens, a field long bedeviled by false promise. Such techniques, and the application of other gene-targeting methods, may yet uncover novel immune modulation targets. As always it is critical to view such papers and results with care and maintain a healthy degree of skepticism. That said, both of these papers have presented reasonable results suitable for further hypothesis testing.