Team:Fudan/Applied Design

SwordS
Antigen density dependent tri-response for therapeutic applications



Abstract


Antigen density on tumor cells’ surface is heterogeneous. Current cellular immunotherapy only targets cells with high expression of specific tumor antigen. While this approach can improve the precision of recognition, it loses the opportunity to strategically treat tumor cells with different surface antigen densities. We are the first to propose a cellular immunotherapy platform, SwordS (SynNotch-Stripe system), that is capable of generating non-monotonic therapeutic responses to one tumor antigen with different surface densities. We demonstrated our concept with the following experiments. In combination with a SynNotch module, which can recognize the antigen, and a Stripe module, which can sort intracellular signal, our engineered cells could generate antigen density dependent tri-response against the target cells with different densities of surface GFP, as well as membrane GPC3 (the tumor antigen of hepatocellular carcinoma). Both experimental data and mathematical simulation show that our platform not only reduces the on-target/off-tumor effects of the original SynNotch system but also provides a combinational therapeutic solution to treat tumors in various states. We believe that SwordS is a promising platform for the next generation of cellular immunotherapy.



Antigen density heterogeneity and limited treatment
two obstacles in improving therapeutic effect and applicability of immunotherapy


Target antigen density heterogeneity

The expression of antigen on individual cells within a given tumor is different or heterogeneous. A solid tumor mass consists of numerous tumor cells. In these tumor cells, some may express relatively fewer tumor antigens, others may express relatively more tumor antigens. Meanwhile, a given tumor antigen is not only expressed on malignant cells, but may also be expressed on normal cells at a low level (Figure 1A). Thus, normal cells expressing a low level of tumor antigens subsequently should not be targeted, otherwise would cause complications. Careful control of the on-target/off-tumor effect is critical to the success of immunotherapy (Figure 1B).




Limited treatment cannot suit all cases

Carcinogenesis is a gradual progress driven by the accumulation of mutations. Tumor cells are highly heterogeneous in their surface tumor antigen expression(1), thus immune resistance(2), sensitivity to the treatment and so on. Meanwhile, efficient recognition by immunotherapy, one of the fundamental challenges for solid tumors, is still on the way comparing with exciting results shown in treating hematological cancers(3). Currently, most existing immunotherapies exhaust in trying multiple methods to improve recognition(4-6), without considerating tumor heterogeneity. They focus on finding an ideal tumor antigen as the target and hope to generate an effective therapeutic response – monotonic response. We believe these conventional one-size-fits-all immunotherapies cannot adapt itself to all complex disease situations in various types of tumors (Figure 1B).




Read more: Current methods to improve recognition and their imperfections


Current methods to improve recognition are not perfect
Emerging methods to improve recognizing precision, like dual recognition (6,7) and tunable sensitivity (5), have been proved capable of eliminating specific tumor cells. However,these methods cannot completely solve the problem.

Specific tumor antigens for dual recognition is hard to find
The first step for immunotherapy is to select a highly specific tumor antigen as the target. However, qualified candidates for tumor antigens are rare in most cases. Taking HCC (hepatocellular carcinoma) as an example, although HCC-associated antigens, like EpCAM(8), NY-ESO-1(9), and GPC3(10), are potential targets of cellular immunotherapies for advanced HCC. However, only GPC3, is wildly accepted as the tumor antigen of HCC owing to its high specificity(6,11,12). Only one tumor antigen for HCC prevents the use of dual recognition.

Tunable sensitivity requires optimized scFv – it is very hard to get
The affinity of single-chain variable fragment (scFv) is essential for recognition. Even though recently high throughput methods have been developed to screen for scFvs with different affinity to the same antigen(13). Specialized knowledge and high expense render it impossible to perform in all laboratories. Only one or two laboratories around the world have the capability to develop scFv with tunable sensitivity.





One important theory of traditional Chinese medicine:
Suit the remedy to the case

Tumors in different states need different treatments. To develop a rational combined therapy, the key question is how to accurately manifest tumors’ condition. Fortunately, with the improved understanding of oncogenesis and emerging therapies, clinical trials of rational combined therapy have become possible(14,15). Through investigation, we found that antigen density heterogeneity, tumor antigen level and expression pattern are associated with disease progression(16). Thus, Antigen density heterogeneity could be used to develop rational combined immunotherapy.

Here, we report the “SwordS”, SynNotch-Stripe system, which can spontaneously generate non-monotonic responses by targeting different surface tumor antigen density of tumor cells at different states (Figure 2).




SwordS enables antigen density dependent
tri-response for therapeutic applications

SwordS consists of two main modules, SynNotch and Stripe, and one supportive module, SynTF-SynPro (Figure 3).




Module 1: SynNotch

SynNotch, an engineered transmembrane receptor, bridges intra- and extra-cellular information.

Synthetic Notch (SynNotch)(17) consists of three parts, the synthetic extracellular recognition domain (SynECD, e.g.scFv), the core transmembrane domain of wild Notch receptor(18), and the synthetic intracellular transcriptional domain (SynICD, e.g.SynTF). When the SynECD binds to its targeting surface antigen, induced cleavages take place on the core transmembrane domain of SynNotch, releasing the SynICD. The SynICD would be transported into the nucleus and activate the transcription of its corresponding promoter (Figure 4).


SynNotch is an ideal platform for customized antigen sensing behavior.

SynNotch provides us an exciting platform because its SynECD and SynICD are both customizable. SynECD can be designed based on currently available scFvs for different tumors such as α-GPC3 for HCC(11, 19), α-Her2 for glioma(4), α-CD19for acute lymphoid leukemia(20), etc. SynICD will trigger customized output after SynECD recognition.

Module 2: Stripe

Stripe consists of 3 interactional circuits.

SynPro S, an activating-form promoter triggered by the existence of initial signal (S), is the start point of the whole module. Meanwhile, activation strength of Promoter S (Pro S) is positively correlated with the concentration of S.SynTF X1/2, SynTF Y, are two orthogonal silencing-form synthetic transcription factor (SynTF) and their corresponding silencing-form synthetic promoter (SynPro) are SynPro X and SynPro Y. (SynTF X1 and SynTF X2 are the same transcription factor on two different open reading frames.) Importantly, the inhibition threshold of SynTF X must be much higher than SynTF Y. A, B, C are three response factors (Figure 5A).

When S’s concentration is none or low. The expression level of circuit ① is low. The concentration of SynTF X1 and Y are not sufficient to inhibit SynPro X or SynPro Y. In this condition, the expression level of Circuit ② is normal.Thus, SynTF X2, the product of circuit ②, is in high concentration to sufficiently inhibit SynPro X and inhibit the expression of circuit ③ (Figure 5B). Only A (in green) is produced.

When S’s concentration is medium. The expression level of ① circuit is mediam. Produced SynTF X1 is still not sufficient to inhibit high-threshold SynProX. However, SynTF Y produced by circuit ① can significantly inhibit low-threshold SynPro Y to express circuit ②. SynPro X cannot be inhibited by either SynTF X1 (not sufficient) or SynTF X2 (inhibited) (Figure 5C). Only B (in red) is produced.

When S’s concentration is high. The expression level of ① circuit is high. Both SynTF X1 and Y are in high concentration. Expression of circuit ② and ③ are inhibited (Figure 5D). C (in blue) is massively produced. Please note that when S’s concentration is low or medium, C is being produced. However, these expressions can be neglected comparing to the dominantly expressed A (in a low concentration of S) or B (in a medium concentration of S).


Stripe achieves combined immunotherapy through an adjustable signal sorting module.

Stripe can spontaneously generate tri-response depending on the intensity of its input: a low-intensity plateau, a hump-like signal peak responding to a medium intensity, and a high-intensity plateau (Figure 6). In a tumor therapy oriented project, we can separately designate the three outputs as no therapeutic factor expression, therapeutic factor I expression, and therapeutic factor II expression. That’s to say, if the input intensity is medium, the output tends to be therapeutic factor I. If the input shifts towards high intensity, the output shifts to therapeutic factor II, and if the input shifts towards low intensity (most likely it is the basal expression in normal cells), output nothing. Thus, normal cells with a low density of surface antigen could be ignored, reducing the off-target effect. Meanwhile, the expression of therapeutic factor I or II can be adjusted via the signal sorting module to suit different tumor conditions.


Develop Stripe with dynamic modeling

To generate an antigen density-dependent tri-response pattern for SwordS, designing the gene transcription network in Stripe plays a central role. The traditional method to analyze a gene transcription network uses Hill Equation to describe the relationship between each pair of transcription factor and its recognition site. However, since its parameters are substantially statistical, Hill Equation cannot provide enough accuracy and flexibility. We build up a Probabilistic Model and apply it to gene transcription network. Our model reveals the relationship between the inherent randomness and phenotypical properties of biochemical reactions in the cell nucleus. We achieve high accuracy and flexibility when modeling a gene transcription network for Stripe development. To provide further insights, we have created an online software to enable you to design and analyze your own gene transcription network dynamically.




Module S: SynTF & SynPro


In silico modeling indicates that the key criterion for functional Stripe construction is the matching orthogonal SynTF X and SynTF Y that enabled sufficient separation of the low and high thresholds. Although previously published works have reported a synthetic Stripe-like circuit in bacterial(21) or in mammalian(22) cells, a tunable system has not yet been created. Their output responses to signal intensity are relatively fixed, which means they respond to high, moderate, and low levels of input with constant intervals. However, tunable intervals should be built so that the system can be tuned to against tumor antigen with heterogeneity. We created and characterized SynTF-SynPro in response to desired intervals. This set is a powerful toolbox to construct customized signaling sorting characteristics.




SynNotch + Stripe
combining these two sharp blades, with the support of SynTF-SynPro, SwordS is a promising platform for the next generation cellular immunotherapy.






The opportunity of applying SwordS to treat HCC


Liver cancer, including HCC (hepatocellular carcinoma), is an extraordinarily heterogeneous disease because of its diversity. As a result, individualized therapy is greatly needed(23). Antigen expression pattern is related to HCC progression and prognosis. High expression of HCC-associated tumor antigen was associated with better prognosis(16): EpCAM whose positive or negative expression can be an indicator to distinguish HCC subtype(24) and the expression of GPC3 is more frequently observed in moderately or poorly states HCC than in well state one(25). Executing combinational immunotherapy is believed to be a dramatic improvement of treatment to HCC(26, 27). With SwordS and the known tumor antigen GPC3, we propose to release therapeutic factor I when GPC3 is medium and release therapeutic factor II when GPC3 is high.

When the expression of the tumor antigen is high, the acquired immune system is relatively easy to identify tumor cells, a therapeutic factor which can enhance acquired immune system might be beneficial for the treatment. Thus, we propose IL-18 as the designating therapeutic factor II, which is secreted protein and promotes the proliferation of CD8+ T cells(28). When the tumor antigen expression is medium, expressing a therapeutic factor that can enhance the innate immune system and facilitate the recognition by acquired immune system might be better. Thus, we propose IL-12 as the designating therapeutic factor I, which could drive potent innate immune responses to cancer(29). In summary, with SwordS, we are able to differential express either IL-12 or IL-18 depending the surface antigen destiny, to strategically kill tumor cells.

The patent is being prepared.





Demonstration


We did a lot of work to demonstrate the feasibility of our project and preliminarily characterized SynTF-SynPro set.

◆Click here to see experimental data and prediction from dynamic modeling.






Postscript


We believe SwordS is a universal solution to various cancers. When we need to demonstrate our idea in one specific cancer, we picked the one with the highest risk for Chinese population, HCC.

◆Click here to see more information about HCC.








Read more: Reference


1. K. Newick, S. O'Brien, E. Moon, S. M. Albelda, CAR T Cell Therapy for Solid Tumors. Annual Review of Medicine68, 139--152 (2017).

2. C. Bock, T. Lengauer, Managing drug resistance in cancer: lessons from HIV therapy. Nature Reviews Cancer12, 494-501 (2012).

3. W. A. Lim, C. H. June, The Principles of Engineering Immune Cells to Treat Cancer. Cell168, 724--740 (2017).

4. H. G. Caruso et al., Tuning Sensitivity of CAR to EGFR Density Limits Recognition of Normal Tissue While Maintaining Potent Antitumor Activity. Cancer Research75, 3505--3518 (2015).

5. X. Liu et al., Affinity-Tuned ErbB2 or EGFR Chimeric Antigen Receptor T Cells Exhibit an Increased Therapeutic Index against Tumors in Mice. Cancer Research75, 3596--3607 (2015).

6. Y. Bi et al., Treatment of hepatocellular carcinoma with a GPC3-targeted bispecific T cell engager. Oncotarget8, 52866--52876 (2017).

7. K. T. Roybal et al., Precision Tumor Recognition by T Cells With Combinatorial Antigen-Sensing Circuits. Cell164, 770--779 (2016).

8. T. Yamashita et al., EpCAM-positive hepatocellular carcinoma cells are tumor-initiating cells with stem/progenitor cell features. Gastroenterology136, 1012-1024. e1014 (2009).

9. S. Nakamura et al., Expression and immunogenicity of NY‐ESO‐1 in hepatocellular carcinoma. Journal of gastroenterology and hepatology21, 1281-1285 (2006).

10. X. B. Man et al., Upregulation of Glypican‐3 expression in hepatocellular carcinoma but downregulation in cholangiocarcinoma indicates its differential diagnosis value in primary liver cancers. Liver International25, 962-966 (2005).

11. H. Gao et al., Development of T cells redirected to glypican-3 for the treatment of hepatocellular carcinoma. Clinical Cancer Research20, 6418--6428 (2014).

12. W. Li et al., Redirecting T Cells to Glypican-3 with 4-1BB Zeta Chimeric Antigen Receptors Results in Th1 Polarization and Potent Antitumor Activity. Human gene therapy28, 437--448 (2017).

13. E. Drent et al., A Rational Strategy for Reducing On-Target Off-Tumor Effects of CD38-Chimeric Antigen Receptors by Affinity Optimization. Molecular Therapy, (2017).

14. B. Al-Lazikani, U. Banerji, P. Workman, Combinatorial drug therapy for cancer in the post-genomic era. Nature biotechnology30, 679-692 (2012).

15. P. Sharma, J. P. Allison, Immune checkpoint targeting in cancer therapy: toward combination strategies with curative potential. Cell161, 205-214 (2015).

16. J. Liang et al., Expression pattern of tumour-associated antigens in hepatocellular carcinoma: association with immune infiltration and disease progression. British journal of cancer109, 1031--1039 (2013).

17. L. Morsut et al., Engineering Customized Cell Sensing and Response Behaviors Using Synthetic Notch Receptors. Cell164, 780--791 (2016).

18. S. J. Bray, Notch signalling in context. Nature Reviews Molecular Cell Biology17, 722--735 (2016).

19. Y.-F. Zhang, M. Ho, Humanization of high-affinity antibodies targeting glypican-3 in hepatocellular carcinoma. Scientific reports6, 33878 (2016).

20. S. A. Grupp et al., Chimeric antigen receptor-modified T cells for acute lymphoid leukemia. New England Journal of Medicine368, 1509--1518 (2013).

21. S. Basu, Y. Gerchman, C. H. Collins, F. H. Arnold, R. Weiss, A synthetic multicellular system for programmed pattern formation. Nature434, 1130--1134 (2005).

22. D. Greber, M. Fussenegger, An engineered mammalian band-pass network. Nucleic Acids Research38, e174--e174 (2010).

23. L. Li, H. Wang, Heterogeneity of liver cancer and personalized therapy. Cancer Letters379, 191--197 (2016).

24. T. Yamashita et al., EpCAM and α-fetoprotein expression defines novel prognostic subtypes of hepatocellular carcinoma. Cancer research68, 1451-1461 (2008).

25. H. Shirakawa et al., Glypican‐3 expression is correlated with poor prognosis in hepatocellular carcinoma. Cancer science 100, 1403-1407 (2009).

26. M. Tagliamonte et al., Combinatorial immunotherapy strategies for hepatocellular carcinoma. Current opinion in immunology39, 103-113 (2016).

27. Z. Wang, S.-J. Qiu, S.-L. Ye, Z.-Y. Tang, X. Xiao, Combined IL-12 and GM-CSF gene therapy for murine hepatocellular carcinoma. Cancer gene therapy8, (2001).

28. B. Hu et al., Augmentation of Antitumor Immunity by Human and Mouse CAR T Cells Secreting IL-18. Cell Reports20, 3025-3033 (2017).

29. M. Q. Lacy et al., Phase II study of interleukin-12 for treatment of plateau phase multiple myeloma (E1A96): a trial of the Eastern Cooperative Oncology Group. Leukemia research33, 1485-1489 (2009).