T-cell receptor engineering of primary NK-cells to therapeutically target tumours and tumour immune evasion

TCR-engineered cells can be powerful tools in the treatment of malignancies. However tumor-resistance by HLA-class I downregulation can negatively-impact the success of any TCR-mediated cell therapy. Allogeneic NK-cells have demonstrated ecacy and safety against malignancies without inducing GvHD highlighting the feasibility for an “off the shelf” cellular therapeutic. Furthermore, primary NK-cells, sourced from the peripheral blood or umbilical-cord blood, can target tumours using a broad array of intrinsic activation mechanisms. Here, we developed a unique NK:TCR cell therapy from peripheral blood derived NK-cells, which enhances NK-ecacy against tumours through additional TCR-mediated lysis. Upon loss of HLA-class I, NK:TCR will lyse the resistant tumour cells in an NK-mediated manner. Our NK:TCR technology is the rst cellular therapy which incorporates TCR-based targeting of tumours and the associated TCR immune-evasion.


Introduction
In unmodi ed T-cells, the T-cell receptor (TCR) confers speci city and engages antigen which has been processed and presented as peptide by HLA molecules. The effectiveness of TCR-mediated adoptive cell therapy (ACT) has been demonstrated by donor-lymphocyte infusion (DLI) 1-3 , infusions of tumour in ltrating lymphocytes (TILs) 4 or virus-speci c T-cells 5 and TCR-engineered T-cells [6][7][8] . With the careful selection of tumour-speci c targets, TCR-engineered cells have the potential to be extremely effective against malignancies [9][10][11][12] . Despite this, overcoming tumour immune-evasion strategies remains a challenge for all T-cell based immune therapies. One such strategy is HLA-class I loss, observed in patients after TIL therapy whereby unresponsive patients had dysfunctional β-2-microglobulin (B2M), required for stable HLA-class I expression 13 . Furthermore, analysis of relapsed patients following immune checkpoint blockade(ICB) revealed acquired defects in antigen-presentation, including B2M mutations, are contributing factors to immune resistance 14,15 . Increasing TCR-mediated immune pressure on tumours, through ICB and T-cell therapies, can therefore promote HLA-class I loss as a tumour immune-evasion mechanism through immunoediting 16 .
Recently, NK-cells have gained interest in ACT and thanks to advances in ex-vivo NK expansion protocols su cient numbers can be achieved for infusion 17,18 . Once activated, NK-cells share similar effector functions with T-cells, including production of cytotoxic granules and in ammatory cytokines, however activation is independent of antigen. Instead, activation relies on a balance between activating and inhibitory signals from germ-line encoded receptors 19 . To prevent non-speci c activation, NK-cells express killer-cell immunoglobulin-like receptors (KIRS) as well as the NKG2A/CD94 heterodimer which provide inhibitory signals upon engagement with HLA-class I molecules on healthy cells. This control mechanism allows NK-cells to become activated by malignant cells lacking HLA-class I expression due to absence of inhibitory signals 20 . This anti-tumour effector mechanism has been exploited in NK-cell based ACT through KIR-mismatching. KIR-mismatched NK-cells have been bene cial therapeutically and has correlated with survival advantages in AML patients receiving T-cell depleted allogeneic stem cell transplant (alloSCT) 21 . Multiple studies have since demonstrated the e cacy of allogeneic NK-ACT as a standalone therapy for the treatment of haematological and solid malignancies [22][23][24][25][26][27] . Unlike T-cells, allogeneic NK-cells do not cause graft-versus-host-disease (GvHD) highlighting the potential for an "off the shelf" cell product allowing broader patient applicability. Enhancement of NK effector function is now a focus and genetically engineering NK-cells to improve e cacy, persistence and homing is part of ongoing research [28][29][30][31] .
We hypothesized that the combination of intrinsic, anti-tumour effector functions of NK-cells with TCRengineering (NK-TCR) would create a novel therapeutic strategy to avoid TCR-associated immune resistance. Recently, TCR expression in an FDA-approved NK-cell line, NK-92, demonstrated TCR-mediated e cacy can be achieved in non T-cells 32,33 . Despite the advantages of a cell line as an "off the shelf" cell product, NK-92 must be irradiated prior to infusion to prevent tumorigenesis which affects the in vivo e cacy 34 . Furthermore, NK-92 cells lack expression of KIRS and have low expression of NKG2A which, without further modi cation, would limit e cacy against malignant cells with HLA-class I loss. 35 Described in this study is a two-step retroviral (RV) transduction protocol to allow functional TCR expression in primary NK-cells derived from the peripheral blood of healthy donors. As a model of e cacy, we focus on the expression of a promising TCR targeting BOB1, a transcription factor highly expressed in all healthy and malignant B-cell lineages, including multiple myeloma 11 . We aimed to demonstrate NK-cells expressing the BOB1-speci c TCR enhances NK effector functions through additional antigen-speci c activation whilst retaining NK-mediated effector functions which can be engaged upon HLA-class I loss as an immune-evasion strategy.

Materials And Methods
Genetic modi cation of NK-cells.

Results
Generation of TCR expressing NK-cells following a two-step retroviral production protocol.
Here, we genetically engineered NK-cells using retrovirus (RV) to generate a nal cell product containing TCR expressing NK-cells (NK-TCR) ( transgenic TCR (tgTCR) was murinised (mTCR) to allow distinction between any contaminating T-cells present in the culture which may in uence functional data. Transduction e ciencies were consistent resulting in expression of mTCRβ (BOB1-TCR 34 ± 13%)(supplemental Fig. 1B), which were enriched before a nal stimulation and further expansion. To test reproducibility, NK-BOB1 cell products were generated from 4 donors and repeated twice for each donor. High expression of CD8β, mTCRβ and CD3ε was repeatedly observed in NK-TCR cell products and NK-TCR speci city was con rmed using peptide-MHC tetramers (Fig. 1D-G). All NK:BOB1 remained negative for human TCRαβ indicating any residual Tcells present at isolation did not expand in vitro ( Fig. 1E and G). Following this stepwise method, pure NK-TCR was generated from multiple healthy donors, within 21 days and total fold expansion of 4385 ± 2026 SEM was observed (supplemental Fig. 2). Furthermore, this protocol could be easily adapted to express different clinically relevant TCRs such as CMV-speci c or PRAME-speci c TCRs (Supplemental gure S3 and S4A-E) NK-TCR express a diverse array of receptors required for NK-cell activation.
The phenotype of nal NK-TCR cell products revealed high frequency expression of activation receptors CD2, DNAM1, CD16, NKG2D (Supplemental gure S5). Natural cytotoxicity receptors (NCRs) NKp30 and NKp46 were also highly expressed whereas NKp44 was expressed on only 29 ± 3.8% (Mean ± SD) of NK:BOB1 (Supplemental gure S5). Additionally, high frequency expression of inhibitory receptor NKG2A was repeatedly observed in NK:BOB1 (93.5 ± 3.3% Mean ± SD) and the activating receptor NKG2C was found on a small fraction of cells (12.8 ± 4.8% Mean ± SD). KIR expression was similarly expressed on a small percentage of cells indicating the expansion protocol did not select for a particular KIR expressing population (Supplemental gure S5). As a negative control, TCR negative NK-cells (NK:TCRneg) were expanded in parallel to NK:BOB1 from the same healthy donors. The phenotype of NK:TCRneg did not signi cantly differ except for an increased frequency of NKp44 expressing cells (54.3 ± 2.2% Mean ± SD) suggesting they were more activated. This expression pro le suggested NK-TCR can be activated via many pathways which would permit anti-tumour effector functions.
NK-TCR elicit potent HLA-dependent, antigen-speci c cytotoxicity against tumour targets Functional capabilities of our generated NK-TCR cell products were investigated on day 21-24 without further stimulation. We validated the NK-mediated cytotoxic potential of NK:BOB1 and NK:TCRneg against HLA-class I negative K562 cell line which was always equally potent and therefore no advantage was observed when BOB1-TCR was expressed (Fig. 2). To explore antigen-speci c killing we designed a panel of EBV-LCLs, 2 endogenously expressing the BOB1 T-cell epitope in HLA-B*07:02 and 2 negative for HLA-B*07:02. NK:TCRneg demonstrated low levels of killing against all 4 EBV-LCLs representing the background NK-mediated cytotoxicity ( Fig. 2A). In contrast, NK:BOB1 demonstrated increased killing of HLA-B*07:02 + but not HLA-B*07:02-EBV-LCLs indicative of TCR-mediated lysis ( Fig. 2A). We con rmed antigen-dependence using HLA-B*07:02 + broblasts which were negative for BOB1 antigen and did not induce TCR-mediated cytotoxicity (Fig. 2B). The expression of TCR on NK:BOB1 also increased cytotoxicity against HLA-B*07:02 + cell lines representing B-cell acute lymphoblastic leukemia (B-ALL) and multiple myeloma (MM) which were previously insensitive to NK-mediated cytotoxicity (Fig. 2C). To expand on this further NK:BOB1 was investigated for its ability to lyse HLA-B*07:02 + leukapheresis samples from patients with B-cell chronic lymphocytic leukaemia (B-CLL) or B-ALL. All samples contained > 70% malignant cells and variable levels of NK-mediated cytotoxicity was demonstrated by NK:TCRneg against each of the malignancies (Fig. 2D and E). As before, increased cytotoxicity was demonstrated by NK:BOB1 and the bene t of NK-TCR was more pronounced when the primary malignancy did not induce an NK-mediated response ( Fig. 2D and E). This TCR-mediated cytotoxicity was consistently demonstrated by NK-BOB1 cell products from multiple donors (Fig. 2F). Furthermore, antigen-speci c killing was not restricted to the BOB1-TCR, and NK-TCR expressing PRAME-speci c TCR (NK:PRAME) or CMV-speci c TCR (NK:CMV) were also able to speci cally lyse antigen-expressing target cells (Supplemental gures S4F and S6).
NK-TCR enabled improved e cacy in a pre-clinical in vivo model of multiple myeloma.
Next, we investigated if TCR-mediated activation of NK:BOB1 would improve e cacy of NK-cell therapeutics in vivo. NSG mice were injected with HLA-B*07:02 + multiple myeloma cell line UM9, previously shown to have low NK-mediated activity (Fig. 2). NK:BOB1 and NK:TCRneg were generated from a KIR-matched healthy donor and co-injected with UM9 in the presence of IL15. E cacy was measured by tumour outgrowth and NK:BOB1 demonstrated delayed tumour outgrowth and increased overall survival compared to NK:TCRneg and untreated mice (Fig. 3). These data validated the in vitro results and demonstrated TCR expression in NK-TCR permits an additional activation pathway that ultimately improves NK-cell therapy e cacy in vivo.

The cytotoxicity of NK-TCR is comparable to CD8 T-cells expressing the same TCR.
To further understand the potency of NK-TCR we compared cytotoxicity relative to CD8 T-cells (CD8T) expressing the same TCR. To remove the in uence of the endogenous TCR on tgTCR expression, the endogenous TCRαβ chains of CD8T were deleted using CRISPR/Cas9, as described previously 37 . Despite the presence of a residual human TCRαβ + cell population co-expressing mTCRβ in CD8T:BOB1 (Fig. 4A), the overall frequency and expression of mTCRβ + cells was comparable between NK:BOB1 and CD8T:BOB1 (Fig. 4B). This was not observed in comparison to CD8T without endogenous TCRαβ knockout (Supplemental gure S7A and B). Upon investigation of antigen-speci c cytotoxicity, NK:BOB1 consistently demonstrated a ratio-dependent, higher overall cytotoxicity against HLA-B*07:02 + B-cell targets endogenously expressing BOB1 than did CD8T:BOB1 (Fig. 4C). However, TCR-dependent killing of NK:BOB1 and T:BOB1 was calculated to be comparable and suggested the enhanced NK:BOB1 killing was the accumulative effect of both TCR and NK mediated activation (Fig. 4E). This effect was also demonstrated when compared to CD8T without endogenous TCR knock-out (Supplemental gure S7C-E). Interestingly, peptide-titration revealed CD8T:BOB1 were cytolytic towards targets cells presenting low concentrations of exogenously loaded antigen, whereas NK:BOB1 was more effective at higher concentrations (Fig. 4D). This was more apparent for NK:CMV as the antigenic-peptide is more readily exogenously loaded onto target cells (supplemental gure S8). Unique dual-activation of NK-TCR may therefore offer an advantage over other TCR-engineered cell therapeutics by enhancing cytotoxic potency when antigen is highly expressed.
Multifunctional, antigen-speci c effector functions are elicited by NK-TCR.
To understand the functional capabilities of NK:BOB1 further, we analysed degranulation and cytokine production of NK:BOB1 and CD8T:BOB1 cell products. All cell products were capable of degranulation, as measured by CD107a, and production of TNFα and IFN-γ in ammatory cytokines except for one CD8T donor that produced low levels of cytokine (Supplemental gure S9). NK-mediated responses were demonstrated by NK:BOB1 and NK:TCRneg which, unlike CD8T, degranulated and produced TNFα and IFN-γ in response to K562 (Fig. 5). Antigen-speci c degranulation and cytokine production was also demonstrated by NK:BOB1 and CD8T:BOB1 following stimulation with HLA-B*07:02 + B-cell targets (Fig. 5). Furthermore, these ndings were not limited to the BOB1-TCR and NK:CMV and CD8T:CMV also showed antigen-speci c responses (supplemental gure S10). Interestingly, CD8T:BOB1 demonstrated signi cantly increased degranulation against multiple myeloma cell line UM9 (Fig. 5A). This trend was also observed for cytokine production, and suggests the effector functions of NK-TCR may be in uenced by the target cell itself ( Fig. 5B and C). Nonetheless, these data demonstrate NK-TCR can evoke multifunctional, antigen-speci c effector functions.
NK-mediated effector functions of NK-TCR can be engaged upon HLA-class I loss on tumour targets.
Finally, we modelled HLA-class I loss in tumours by generating HLA-B*07:02 + EBV-LCLs, B-ALL and MM cell lines with β-2-microglobulin knock-out (B2M KO). B2M KO prevented cell surface expression of HLAclass I molecules in all cell lines (Fig. 6A). B2M KO or wild-type cells were then co-cultured with NK:BOB1, NK:TCRneg, CD8T:BOB1 and CD8T:MOCK. In this model of HLA-class I loss, CD8T:BOB1 killed the wildtype cells but was unable to kill the B2M KO cells due to absence of TCR-speci c epitope at the cell surface ( Fig. 6B and C). In contrast, NK:BOB1 and NK:TCRneg demonstrated equally potent killing of B2M KO cells due to activation of NK-mediated cytotoxicity ( Fig. 6B and C). As demonstrated previously, only NK:BOB1 was able to kill wild-type cells representing TCR-mediated killing. Next, we compared BOB1-TCR e cacy in CD8T and NK-cells in vivo using the same UM9 multiple myeloma model previously described (Fig. 3A). KIR-matched NK-cells and CD8T were isolated from the same healthy donor and as a negative control, the CMV-TCR was also expressed in both NK-cell and CD8T in parallel to the BOB1-TCR (Fig. 7A).
CD8T:BOB1 cells are capable of controlling high tumour burdens in xenograft models 11 , therefore we doubled the number of tumour cells infused to further understand the capabilities of NK:BOB1 in comparison to CD8T:BOB1. NK:BOB1 and CD8T:BOB1 treated mice demonstrated reduced tumour outgrowth of wildtype UM9 compared to untreated mice, whilst NK:CMV and CD8T:CMV treated mice did not (Fig. 7B). Furthermore, no signi cant difference was found between NK:BOB1 and CD8T:BOB1 treated mice until day 26 post infusion, demonstrating comparable TCR-mediated control of the tumour. However, this effect did not persist in NK:BOB1 treated mice as demonstrated by increased tumour outgrowth whilst CD8T:BOB1 maintained control of the tumour (Fig. 7B). In parallel to this, mice were similarly infused with a 50/50 mix of UM9 wildtype and UM9 B2M KO cells alongside NK-TCR and CD8T (Fig. 7A). In contrast to UM9 wildtype mice, NK:CMV signi cantly reduced tumour outgrowth compared to untreated mice, demonstrative of NK-mediated targeting of HLA-class I negative tumour cells (Fig. 7C). Furthermore, NK:BOB1 demonstrated an enhanced effect as both HLA-class I negative and positive cells can be targeted (Fig. 7C). CD8T:BOB1, and not CD8T:CMV, also signi cantly controlled tumour outgrowth, however CD8T:BOB1 was no longer able to maintain control of tumour outgrowth as only half the tumour expressed HLA-class I (Fig. 7C). These data demonstrate the NK-mediated effector functions of NK-TCR can also be engaged in the absence of TCR-speci c epitope following loss of HLA-class I and has potential to prevent outgrowth of TCR-resistant malignancies.

Discussion
The data presented in this study demonstrates the enhancement of NK effector function by genetically engineering primary NK-cells to express a TCR. High expression of TCR was repeatedly observed in modi ed NK-cells and the protocol can be used to express different TCRs targeting many tumour types.
NK-TCR permits an alternative NK-activation mechanism against malignancies which are insensitive to NK-mediated attack. To date, autologous NK-ACT has not shown clinical-e cacy and inhibition by self-HLA is likely to contribute to its ineffectiveness 39,40 . Equipping autologous NK-cells with tumour-speci c TCR may therefore override inhibition signals and boost therapeutic e cacy. Furthermore, the cumulative effect of NK-mediated and TCR-mediated activation increased overall cytotoxicity against tumour targets when compared with tgTCR expression in T-cells (Fig. 4C). This increased potency could be bene cial in clinical contexts in which NK-cells are already associated with anti-tumour effects, such as alloSCT. Graft verse leukaemia (GvL) effects are described in patients transplanted with alloreactive KIR mis-matched NK-cells [21][22][23]41 . Our data infers that tgTCR expression in infused NK-cells could enhance this effect. Another application could be early infusion of allogeneic NK-cells expressing virus-speci c and tumourreactive TCR following T-cell depleted alloSCT. This would potentially allow preservation of GvL effects and protection against harmful viral re-activities in the absence of T-cells, without the risk of GvHD 5 . Our data does indicate NK-TCR may not be as potent against low antigen expression when compared to CD8T expressing the same TCR ( Fig. 4D and Supplemental gure S8). For this reason, BOB1 is an ideal target for NK-TCR for it is highly expressed in most B-cell malignancies 11 , nevertheless antigen expression must be carefully considered for alternative TCR-speci cities.
To our knowledge this is the rst study demonstrating TCR expression in primary NK-cells derived from the peripheral blood (PB) of healthy donors. To date, functional TCR expression in non-T-cells has been restricted to the NK-cell line, NK-92 32,33 , largely owing to low transduction e ciencies observed in primary NK-cells. As demonstrated in this study, improvements to NK-expansion protocols have now enabled e cient retroviral transduction of primary NK-cells (supplemental gure S1). PB offers a readily available source of mature NK-cells with a potent cytotoxic pro le, leading to their use in multiple clinical studies 42 . In allogeneic settings, alternative NK sources such as cord blood (CB) or haematopoietic stem cells have also been used due to the reduced risk of T-cells being present in the graft 28,43 . In this study T-cells present after NK-isolation did not expand upon stimulation and were absent from nal NK-TCR cell products (Fig. 1). Although not extensively studied, we have demonstrated functional TCR could also be expressed in CB-NK suggesting our protocol is not limited to PB-NK (data not shown). Primary NK-cells can offer certain advantages for cellular therapy when compared to NK-92 cell lines. Firstly, it is necessary to irradiate NK-92 cells prior to patient infusion to circumvent tumorigenesis, which can negatively impact on therapeutic persistence and e cacy and often requires multiple doses 34 . In contrast, it was recently demonstrated a single dose of CB-NK expressing CD19-IL15 CAR achieved complete responses in patients with CLL and NHL and the NK-cells persisted at least 12 months after infusion 44 . Secondly, although NK-92 express a wide array of activation receptors they lack expression of CD16, KIRS and have low expression of NKG2A which can limit the tumour targeting potential compared to a primary NK population. Here, we demonstrate NK-TCR derived from PB have high expression of CD16 which can therefore be activated by monoclonal antibody and induce antibody-dependent-cellularcytotoxcity (ADCC) in combinational therapies (Supplemental gure S5). Furthermore, NK-TCR has a diverse array of KIRS, and high expression of NKG2A, required for NK-cell activation in the absence of HLA-class I on tumour cells (Supplemental gure S5). Therefore, our NK-TCR approach using primary NKcells, is unique in its ability to target tumours speci cally as well as the associated immune resistance mechanisms following increased immune pressure.
The implications of immune pressure is particularly evident in CD19 CAR-T therapy through the appearance of antigen-negative relapses which is a common feature of antibody-based approaches targeting non-essential proteins 45,46 . For TCR-mediated approaches, speci c targeting of essential proteins from intracellular processes can be achieved. However, loss of antigenicity can occur due to HLA-class I loss/downregulation as a result of immunoediting and has been described in patients treated with ICB and TILs 13,14,47 . The duality of NK-TCR means that acquired loss of HLA-class I, as a result of TCR-mediated immune pressure activates innate NK responses. Our data supported this as NK-cells with or without BOB1-TCR expression were cytotoxic against target cells with a β-2-microglobulin knock-out, whilst the wild-type counterpart was insensitive to NK-mediated attack and could only be targeted via the BOB1-TCR (Figs. 6 and 7). In theory, the TCR-mediated and NK-mediated pathways of NK-TCR could provide feedback for each other and prevent a selective growth advantage for tumours with defects in antigen presentation 14 . However, tumours can also acquire speci c HLA-allele loss following immune pressure, as seen in patients who relapsed following HLA-haploidentical SCT 48 . HLA-dependence of NK-TCR may therefore be considered a limitation and as with most TCR-mediated approaches a multi-HLA targeting cell product will be paramount to its broader application.
For any ACT, it is important to reduce toxicity to allow infusion of clinically-effective doses. NK-cells have a reduced cytokine pro le which limits their capacity to cause cytokine release syndrome even when expressing CD19-CAR 44 . We demonstrate NK-TCR have reduced cytokine production in response to antigen compared to NK-mediated activation, suggesting TCR stimulation would not be harmful (Fig. 5), and tgTCR expression in T-cells has not yet shown such toxicities 6,7 . Furthermore, evidence that allogenic NK-cells do not cause graft-versus-host-disease (GvHD), makes them a safer alternative when Tcells are considered too high risk 26,27 . Together, these properties make NK-TCR an interesting candidate for an "off the shelf" TCR-based ACT. Although generating an NK-TCR cell product from primary NK-cells is feasible, this protocol is laborious and translating to the clinic may be challenging, although not impossible. A single-transduction step would simplify the approach, however this method needs to be optimised and so far has resulted in lower tgTCR expression and e cacy (data not shown). Moreover, there are several limitations to overcome to become a standardised clinical approach. For instance, an obstacle for "off the shelf" allogeneic ACT is avoiding elimination by the host immune system and currently allogeneic NK-cells are given to patients who have undergone lymphodepletion regimens 22,44 .
NK-cells themselves are short lived and, although an advantage in terms of toxicity, this may reduce clinical e cacy. It is known that in the absence of IL15 primary NK-cells do not persist 49 and in this study, tumour outgrowth occurred in NK:BOB1 treated mice following IL15 withdrawal (Fig. 7B). This was suggestive of decreased NK-cell persistence and also makes direct comparison with CD8T in vivo di cult. Increasing persistence by inclusion of autonomous IL15 production in NK-TCR would therefore be important to its development. There is also limited clinical success of NK-cell targeting solid malignancies, despite many solid malignancies demonstrating down-regulated HLA-class I 50 . This is often associated with insu cient migration and tra cking of NK-cells into these sites. Modifying NKcells to express molecules that enhance persistence and homing is possible but further NK-TCR manipulation maybe technically challenging. Finally, other than HLA-class I loss, tumours can exhibit many immune evasion strategies 51 including immune-checkpoint inhibition which can directly impact on NK function 52,53 . Therefore, combination therapy with ICB should also be considered in future NK therapeutics 54 . The associated resistance to immune-checkpoint blockade may in turn be overcome by NK-TCR by preventing a selective growth advantage for tumours with defects in antigen presentation.
In conclusion, following our described two-step retroviral transduction protocol, TCR can be feasibly expressed in primary NK-cells and used for ACT. The data presented in this study demonstrate TCRmediated NK responses can enhance NK-cell e cacy against malignancies, particularly when resistant to NK-mediated attack. In addition, NK-TCR can also target HLA-class I loss, an associated immune-evasion strategy of TCR-mediated approaches which makes NK-TCR a unique cellular therapeutic.

Declarations
Competing-interest-statement The authors declare no competing nancial interests.