Precise tumor immune rewiring via synthetic CRISPRa circuits gated by concurrent gain/loss of transcription factors

An active CRISPRa can flexibly program immunostimulatory outputs in tumor cells

Synthetic gene circuits hold considerable potential to enable tumor-specific delivery of immunological interventions¹². We reasoned that the recently established CRISPRa effector platform might be further adapted as key parts for the construction of tumor-targeting immunoregulatory gene circuits. A potent, three-component synergistic activation mediator (SAM) effector was chosen for exploration^27,28. This CRISPRa unit standardly consists of dCas9-VP64 (or unmodified dCas9), targeting sgRNA for activation (sgTGTa), and an artificial co-activator (MS2–p65–HSF1, MPH) specifically recruited by the engineered stem–loops on sgTGTa (Supplementary Fig. 1b). As the fused VP64 moiety on dCas9 was previously shown to be unessential for SAM-mediated gene activation²⁷, the “basic” version of dCas9 was used for the assembly of CRISPRa throughout the present study.

A previously reported tumor-specific promoter of survivin gene (P_Suv)^29,30 or a constitutive SV40 promotor was first placed upstream of the dCas9 coding cassette (Supplementary Fig. 1c, top illustration). The CRISPRa activities associated with these constructs were examined in human and mouse lung tumor cell lines (H1299 and Lewis lung carcinoma [LLC], respectively), along with the primary mouse embryonic fibroblasts (MEFs) as representatives of normal cells. The cells were transfected with plasmids encoding dCas9, MPH, and sgTGTa, together with a reporter construct with designed, 3× target sites upstream of a minimal promoter-led EGFP (see “Methods”). In reference to the activity profile of the SV40 promoter, the P_Suv showed apparently higher activities to drive dCas9 expression in tumors cells than in MEFs (Supplementary Fig. 1c, lower, Flag panel). Importantly, the CRISPRa activities (reported by EGFP levels) in different groups defined by various cell types and dCas9 constructs closely correlated with those for dCas9 expression (Supplementary Fig. 1c, lower). These results provide evidence that controlled dCas9 expression by the use of tumor-specific promoters can enhance the tumor cell selectivity for the CRISPRa outputs.

Next, the effects of P_Suv-CRISPRa on inducing an immunoregulatory output were examined in the H1299 cell line. The endogenous IFNγ (encoded by IFNG), a cytokine strongly involved in natural tumor immune surveillance^31,32, was selected as an endogenous target. Guided by individual sgRNAs corresponding to different positions upstream of IFNG, the CRISPRa complexes (driven by P_Suv-dCas9) led to variable degrees of targeted transcription induction (Supplementary Fig. 1d). In parallel, a downstream marker of IFNγ signaling, i.e., IRF1, showed a similarly variable induction pattern (Supplementary Fig. 1d). Importantly, a neutralizing antibody against IFNγ largely abrogated such secondary IRF1 induction, without affecting the CRISPRa-dependent IFNG activation (Fig. 1b). The introduction of CRISPRa-IFNG in H1299 cells also led to upregulation of cell surface HLA-ABC (Class I), which was likewise abrogated by the IFNγ-blocking antibody (Fig. 1c). These results demonstrated that an active CRISPRa SAM effector could effectively program the tumor cells for induction of endogenous IFNγ, which in turn triggered autocrine/paracrine signaling.

CRISPRa may be easily adapted for enhanced efficiency via pooled sgRNAs targeting the same gene, or for multiplexed targeting via combining more than one gene-specific sgRNAs^27,33,34,35. Consistently, we found that a combination of two sgRNAs targeting IFNG could lead to a more robust downstream effect of HLA up-regulation (Fig. 1d). Moreover, a combination of activating sgRNAs for IFNG and for a T cell chemokine, i.e., CCL21 (Supplementary Fig. 1e), could lead to simultaneous induction of both genes (Fig. 1e). Collectively, our data so far point to the potential use of CRISPRa SAM in a tumor-conditional therapeutic gene circuit, for effective and versatile induction of endogenous immune regulators.

Construction of a preliminary dual-input, AND–NOT logic circuit

Few promoters are proven to be stringently tumor-specific³⁶. Therefore, the mere use of a “tumor-enhanced” promoter for dCas9 expression (as adopted in the experiments above) is most likely not sufficient to stringently ensure a tumor-specific output by the CRISPRa effector. For an improved cell specificity in output induction, we next sought to incorporate the CRISPRa effector into more sophisticated, multi-input Boolean logic circuits^4,5.

For a CRISPRa device, or likewise a CRISPRi target-repressive device^22,27,37, their individual dCas9 and sgRNA components may be placed under different TF-responsive promoters (“P1” and “P2”) to enable flexible input sensing (in a singular or combinatorial format), and logic processing (Supplementary Fig. 2a). To enable Pol II-dependent sgRNA expression, a design of ribozyme-based auto-processing cassette was adopted (Fig. 2a)³⁸. Using a Cas9/sgRNA-mediated cleavage reporter assay in 293T cells³⁹, we validated that a Pol II-type, CMV promoter-driven sgRNA had an evident targeting activity, although at a level lower than that by conventional Pol III-type (U6) sgRNA construct. Such partially reduced activities associated with the Pol II sgRNA construct are consistent with previous reports^38,40.

Subsequently, we tested (in 293T cells) the use of Pol II promoter-driven sgRNA to shape CRISPRa/i outputs. Here, a constitutive (i.e., CMV) and a signal-dependent promoter were chosen to respectively direct the sgRNA and dCas9 moiety, or vice versa. As for the signal-dependent promoter, we selected an IFN-sensitive response element (ISRE, 5×)⁴¹, which senses activation of an ISGF3 TF complex by type I interferon (IFN).

Specifically, in a first CRISPRi-based circuit, a CMV promoter (or a U6 promoter, as a positive control) drove the constitutive expression of a targeting sgRNA (for target inhibition, sgTGTi), while an ISRE promoter directed the expression of the KRAB repressor domain-fused dCas9 [dCas9-KRAB] (Supplementary Fig. 2b, left illustration). Alternatively, a second circuit instead incorporated an ISRE-controlled sgTGTi, together with a CMV-dCas9-KRAB (Supplementary Fig. 2c, top illustration). A repressible EGFP reporter (in a destabilized form, dEGFP⁴²) with a unique sgTGTi-targeted sequence near the transcription start site was co-transfected for indication of CRISPRi activities. We found that both circuits were capable of programming IFN-dependent repression of dEGFP, although the outputs appeared less robust than those engaged in the U6-sgTGTi positive control group (Supplementary Fig. 2b, c). These results validated significant activities of sgTGTi driven by two Pol II promoters (CMV and ISRE), and established a platform for dynamic CRISPRi outputs via induced expression of either dCas9-KRAB or sgTGTi.

In the ensuing experiment, we constructed an IFN-signaled CRISPRa activating device (Fig. 2b, upper illustration). This device further adopted a design of co-regulation, with individual ISRE promoters driving both dCas9 and sgTGTa (besides a constitutive MPH). A CRISPRa-dependent EGFP reporter, which we used earlier (see Supplementary Fig. 1c), was monitored as circuit output. Expectedly, such a circuit programmed the cells to transduce an IFN signal into CRISPRa activation [EGFP upregulation] (Fig. 2b).

The above circuits with promoter-controlled CRISPRa/i actuators can produce outputs according to increases (gain) in TF activities. However, as TF inactivation events (loss) are prevalent in tumors, a different class of circuits whose outputs are triggered by loss of TF (e.g., p53) activities would be particularly suitable for specific tumor targeting. This functionality requires an inhibitory module⁴³, whose absence allows output production (de-repression), forming a NOT-type logic control. Given the multiplexable targeting feature of dCas9, we hypothesized that the above-validated, Pol II-driven sgRNA expression system may be further harnessed for constructing a NOT logic gate. For instance, in a system with a P1 driving both a dCas9-effector and a sgTGT, the presence of an sgRNA (P2-dependent) against their promoters (with off-switching activity, sgOFF), would concomitantly engage the dCas9 moiety to form a NOT logic gate (Supplementary Fig. 2d). Note that the “sgOFF” is used as a generalized nomenclature in this work.

To test this principle, a circuit was first assembled with 5× ISRE-driven dCas9-KRAB and sgTGTi, together with a Pol III promoter-driven sgOFF (U6-sg#1-i) targeting a sequence within the ISRE promoter. Here, the sg#1-i would engage dCas9-KRAB into a dual-target repression module (“2× OFF”) against itself and sgTGTi. For comparison purposes, in an alternative circuit featuring a constitutive dCas9-KRAB (CMV) and a regulated ISRE-sgTGTi, we implemented a single-target inhibitory format (“1× OFF”), where the sg#1-i was poised to suppress only the sgTGTi component (Supplementary Fig. 2e, top illustration). With the 2× OFF format, the U6-driven sg#1-i notably impeded IFN-dependent induction of dCas9-KRAB compared to a non-targeting sgRNA (against a luciferase-derived sequence, sgLuc), which reflected the direct actions by this inhibitory module. Concomitantly, IFN-triggered synthetic output, i.e., dEGFP down-regulation, was abrogated by sg#1-i (Supplementary Fig. 2e, lower right). On the other hand, with the 1× OFF format where the U6-sg#1-i would only suppress sgTGTi, IFN-triggered dEGFP down-regulation was restored to a lesser extent (Supplementary Fig. 2e, lower left). These results suggested a framework for a sgOFF-dependent control of CRISPRi function, where a design of co-inhibition against its multiple components could lead to a more efficient control.

Unlike the U6-sg#1-i used above, a Pol II-dependent CMV-sg#1-i appeared significantly less effective in our testing attempts to suppress 5× ISRE-dependent dCas9-KRAB levels (Supplementary Fig. 2f), which was likely to reflect a combination of its dampened expression via a Pol II system and certain unfavorable features in targeting position/sequence. Subsequently, to sensitize the CRISPRi components to repression by sgOFF, we inserted artificial sgOFF target sequences (each in a 2× format) near the core promoters in both ISRE-led dCas9-KRAB and sgTGTi constructs (Supplementary Fig. 2g, top illustration). Two parallel CRISPRi actuator modules respectively featuring a different set of such artificial “cis-elements” were constructed, and were assembled together with their corresponding CMV-driven sgOFFs (sg#2-i and sg#3-i). The results showed that only one of the two CMV-driven sgOFFs (sg#3-i) visibly blunted IFN-induced dCas9-KRAB and the associated changes in dEGFP (Supplementary Fig. 2g, lower). Such variable effects by Pol II-driven sgOFFs also confirmed the importance of target optimization for their improved performance.

A CRISPRa SAM-containing circuit was also designed to incorporate a similar sgOFF-dependent inhibitory module, taking advantage of the fact that an unmodified dCas9 would serve as a common component for simultaneous gene activation (programmed with a stem–loop-modified sgTGTa) and inhibitory circuit control (with a normally scaffolded sgOFF)^37,44. Following target optimization (Supplementary Fig. 2h), an effective CMV-driven sgOFF, i.e., sg#4-i, was assembled with the triplicate CRISPRa components which were individually led by ISRE promoters [featuring a common proximal target sequence] (Fig. 2c, left illustration). The results showed that IFN/CRISPRa-dependent EGFP reporter induction was suppressed by CMV-sg#4-i for more than 50% (compared to sgCon-i), consistent with the pattern of dCas9 expression (Fig. 2c, d).

To test the control of the sgOFF inhibitory module by a physiological signal, an NF-κB-responsive element (5× NκRE) promoter was used to drive sg#4-i. This promoter responds to NF-κB, a TF commonly activated by inflammatory signals such as TNFα. In cells co-transfected with all circuit components (ISRE-CRISPRa/sgTGTa/MPH and NκRE-sg#4-i), IFN-triggered EGFP output was evidently reduced by the co-addition of TNFα, in a pattern correlating with that of dCas9 (Fig. 2e, f). Overall, these results presented a preliminary design of a P1-CRISPRa/i effector- and P2-sgOFF-incorporated circuit to sense a TF1⁺/TF2⁻ state (AND–NOT logic). In the remainder of this study, we focused on the CRISPRa SAM-featured circuits, due to their versatility and effectiveness in directly driving cells toward immune rewiring (see Fig. 1).

Construction of a customized AND–NOT logic circuit (v1) targeting a malignant state

The AND-NOT logic circuit is particularly relevant for driving tumor-specific output (Fig. 3a), via potentially enabling the detection of simultaneous gain- and loss-of TF activities (e.g., hypoxia-inducible factor 1α [HIF1α] and p53, respectively) characteristic of many tumors^15,16,17,45. We, therefore, sought to construct the first version of HIF1α/p53-sensing, AND–NOT logic circuit using the CRISPRa/sgOFF strategy (v1).

As the H1299 cell line is p53-deficient⁴⁶, it was used to establish the cell system for circuit testing. We validated a hypoxia-response element (HRE, 3x) in this cell line using a HIF1α-activation reagent, i.e., CoCl₂ (Fig. 3b)⁴⁷. We further generated an H1299-derived stable cell line for tetracycline-controlled p53 expression and activity (p53-tet) (Supplementary Fig. 3a, b). A well-known p53-responsive promoter (from the intron of MDM2, P_M2⁴⁸) was also validated in p53-tet H1299 cells (Fig. 3c). Individual HRE promoters were used to drive the expression of all three CRISPRa components (targeting EGFP for induction). Additionally, the core promoters for these units all contain the same sequence targeted by the sgOFF (sg#4-i), which in turn, is part of the inhibitory module controlled by a P_M2 promoter (Fig. 3d, left illustration, the circuit denoted as v1.1 to indicate its stage of development). In p53-tet H1299 cells introduced with such an AND–NOT circuit, a HIF1α input (+CoCl₂) engaged CRISPRa to induce EGFP expression for about fivefold under the p53-deficient condition (−DOX). In contrast, a simultaneous p53 input (+DOX) acted to partially inhibit the CoCl₂-signaled EGFP induction (up to ~50%), consistent with its dampening effect on dCas9-Flag protein (Fig. 3d, right).

Next, we set out to simplify the CRISPRa expression system into a single-promoter construct, as in a previous report³⁸. Such simplification would also conceivably enable a more uniformed, sgOFF-mediated transcriptional suppression of all actuator components (Fig. 3e, upper illustration, circuit v1.2). Herein, a 3× HRE promoter drove a single mRNA precursor, in which dCas9 and MPH were separated by an internal ribosome entry site (IRES), followed by the auto-processing sgTGTa cassette in the 3′ untranslated region. The same P_M2-sg#4-i module was used as the NOT gate. In p53-tet H1299 cells, such a modified circuit (v1.2) wired a CoCl₂/HIF1α input into an about twofold increase of EGFP output in cells with the p53-null state [−DOX] (Fig. 3e). In this context, a simultaneous DOX/p53 signal restored the CoCl₂/CRISPRa-enhanced EGFP down to a level seen in untreated cells, suggesting an effective inhibitory control by P_M2-sg#4-i on the inducible portion of CRISPRa activity (Fig. 3e and Supplementary Fig. 3c). Nevertheless, an apparent background CRISPRa activity resisted inhibition by DOX/p53. Therefore, despite apparently exhibiting some inverse-tuning ability, the NOT logic gating of CRISPRa/i by Pol II-driven sgOFFs presented non-optimal precision in various contexts (Fig. 2d–f, Fig. 3d, e and Supplementary Fig. 2g). A further improvement in the NOT gate performance was warranted.

Development of a more accurate NOT gate for CRISPRa by employing the anti-CRISPR AcrIIA4

Given the observed imprecision with the sgOFF-dependent NOT gates, we realized that a pitfall of such a design is that both the actuator and its inhibitor are based on dCas9-mediated DNA targeting, which may readily endow a modest potency for the sgOFF module. The conceivably limited expression of sgOFF by a Pol II is likely to present an additional challenge for its effectiveness. Therefore, a more robust inhibitory unit against the CRISPRa would be instrumental for assembling the NOT gate. Along this line, we noted that recent pioneering works had uncovered numerous phage-derived proteins (anti-CRISPRs) that directly antagonize CRISPR–Cas activities⁴⁹. Importantly, one of the anti-CRISPR proteins, i.e., AcrIIA4, was previously established to potently inhibit Cas9-mediated DNA sequence interrogation^24,25,26. Subsequently, we adopted AcrIIA4 for developing another NOT gate for CRISPRa.

We first compared the original AcrIIA4 in bacterial codons with its human codon-optimized counterpart (referred to as ACRmax herein). While both forms effectively reduced the activities of co-transfected CRISPRa, the ACRmax exhibited an even greater inhibitory potency (Supplementary Fig. 4a and Fig. 4a). Subsequently, we constructed a P_M2-ACRmax inhibitory module, which was assembled together with a single-unit P_Suv-driven CRISPRa actuator (Fig. 4b, left illustration). In p53-tet H1299 cells, the EGFP output under the p53-null condition (−DOX) was about four-fold higher than that under the DOX/p53-rescued condition (Fig. 4b, middle and right). Consistent with the mechanism of AcrIIA4 action²⁵, p53/ACRmax-associated regulation of EGFP output showed little correlation to the levels of dCas9. Importantly, compared to earlier experiments with the sgOFF system (see Fig. 3e and Supplementary Fig. 3c), we noticed that a lower dose of DOX (10 ng/ml) here caused a greater degree of output inhibition (Fig. 4b, right). This is not due to the use of different dCas9 constructs, as a control experiment showed that a P_Suv or CoCl₂-activated HRE drove similar levels of dCas9 expression (Supplementary Fig. 4b). Such comparisons indicated a better performance by the P_M2-ACRmax module over its sgOFF-based counterpart. Additional tests were carried out using a p53-tet/H1299-derived sub-line with a virally integrated P_M2-ACRmax construct. Even when a strong CRISPRa (CMV-dCas9/U6-sgTGTa) actuator was introduced by transfection, engagement of the integrated P_M2-ACRmax by a DOX-restored p53 input could effectively suppress the EGFP output (Supplementary Fig. 4c–e).

Such favorable gating performance by P_M2-ACRmax in the p53-tet H1299 cells made us proceed further. The event of p53 inactivation, although not uniformly occurring in tumors, is strongly associated with a higher degree of malignancy⁵⁰. Therefore, it would be particularly desirable if P_M2-ACRmax inhibitory module could effectively sense the basal activity of endogenous p53, leading to accurate identification of p53-inactivated tumor cells (Fig. 4c). A549 is a human lung cancer cell line that has WT p53. The p53 functional status in these cells was additionally validated under the treatment with a p53 stabilizer (Nutlin-3) or a chemotherapeutic drug (cisplatin) (Supplementary Fig. 4f, g). We transfected both A549 and H1299 cells with a CRISPRa actuator (CMV-dCas9/U6-sgTGTa) and the P_M2-ACRmax module. In stark contrast to the unrestricted output pattern in H1299 cells, the CRISPRa-dependent EGFP output was markedly inhibited by P_M2-ACRmax in A549 cells (Fig. 4d). A similar observation was made in another p53-sufficient breast cancer cell line MCF-7 (Supplementary Fig. 4h).

To formally establish that P_M2-ACRmax-mediated inhibition on CRISPRa was responsive to basal p53 signaling in A549 cells, genetic knockout of p53 in these cells was carried out via genome editing (Cas9). Next, six different clones of parental A549 cells or the p53-deficient derivatives were introduced with the CMV-CRISPRa/P_M2-ACRmax circuit. The results indicated that targeted p53 inactivation in A549 cells markedly mitigated P_M2-ACRmax-dependent suppression of CRISPRa output (Fig. 4e). We also conducted similar tests in primary MEFs, taking advantage of the p53 germline knockout mouse model. A clear de-repression of EGFP output was observed in p53^−/− cells, in comparison to the WT MEFs (Fig. 4f). Interestingly, further experiments showed that a P_M2-ACRmax inhibitory module was engaged even in p53^+/− MEFs, leading to marked suppression of the EGFP output driven by a constitutive SV-40-dCas9 (Supplementary Fig. 4i). These results from the isogenic control and p53-null cells demonstrate that the P_M2-ACRmax module was operated by basal p53 signal (even with a single allele), and consequently enabled stringent gating of CRISPRa output only by severe p53 deficiency. It is also encouraging that in these experiments, the basally engaged P_M2-ACRmax module was sufficiently potent in relation to a strong CRISPRa output (driven by constitutive dCas9/U6-sgTGTa), which in turn would be favorable for driving therapeutic activities.

An improved P_Suv/P_M2 AND–NOT logic circuit (v2) rewires p53-deficient tumor cells to produce immunostimulatory ligands

Given the establishment of a stringent p53-responsive NOT gate for CRISPRa activity, we proceeded to establish an improved AND–NOT tumor-targeting circuit (v2) and to additionally wire it toward an immunotherapeutic output (endogenous IFNγ). To this end, a tumor-preferential CRISPRa module (P_Suv-dCas9/U6-sgIFNG) was assembled with a P_M2-ACRmax inhibitory module (Fig. 5a, left illustration). Following transfection to the p53-tet cells, such a circuit (v2) enabled a marked activation of the IFNγ-STAT1 signaling in p53-null, but not p53-rescued (+DOX) cells (Fig. 5a, right). Additionally, the conditioned medium from p53-null, but not p53-rescued cells showed apparent paracrine activities to upregulate HLA proteins in tumor cells (Class I HLA, in H1299 cells), and in the freshly prepared peripheral blood mononuclear cells (Class II HLA) (Supplementary Fig. 5a and Fig. 5b). Furthermore, consistent with the reported direct actions of IFNγ on tumor cell growth⁵¹, noticeable circuit-engaged growth-inhibitory effects (triggered by two IFNG-activating sgRNAs in combination) occurred selectively in p53-null (−DOX) cells (Fig. 5c). The circuit could also be conveniently adapted to target p53-deficient cells for multiplexed induction of immune stimulators, i.e., IFNγ and CCL21, via co-delivery of their corresponding activating sgRNAs (Supplementary Fig. 5b).

The mouse LLC cell line is p53-deficient^52,53 and represents another suitable model to examine the p53 loss-gated tumor rewiring by our circuit. For comparison purposes, we established an LLC-derived knock-in line (p53⁺), where the WT p53 coding region was re-introduced into the endogenous locus to rescue the bi-allelic point mutations (a nonsense mutation at E32 and an R334P mutation, respectively) in the parental LLC cells (Supplementary Fig. 5c, d). The resultant p53⁺ line showed moderately increased levels of basal and Nutlin-enhanced p53 activities measured via a transient P_M2-EGFP reporter (Supplementary Fig. 5e).

The parental LLC and their p53⁺ derivatives were next used to further test the cell-selectivity for the synthetic output by a P_Suv/P_M2 AND-NOT circuit (v2). Here, the choice of P_Suv for driving CRISPRa was based on our earlier comparative experiments in the LLC tumor cells and MEFs (Supplementary Fig. 1c). In an application context, it is desirable that an evenly distributed gene circuit can restrictively trigger outputs in the targeted subpopulation. In a proof-of-concept experiment, we established a co-culture of LLC cells and their p53⁺ derivatives, respectively representing the targeted and non-targeted subpopulations. For easy distinguishment of the two subpopulations, the parental LLC cells had been labeled in advance with mCherry. Subsequently, the co-culture was transfected with the P_Suv/P_M2 AND–NOT circuits (v2) and an output reporter [EGFP] (Fig. 5d, left illustration). Indeed, with the mock NOT gate, the P_Suv-CRISPRa-dependent EGFP output showed a similar order of magnitude in the two subpopulations (Fig. 5d, right panel). In contrast, the inclusion of the P_M2-ACRmax NOT gate largely prevented CRISPRa activation in the p53⁺ LLC derivatives (mCherry⁻), which led to the specification of the p53-deficient LLC subpopulation (mCherry⁺) for an EGFP output (Fig. 5d). These results indicate the capability of a transfected P_Suv-CRISPRa/P_M2-ACRmax circuit for stringent identification of the p53-deficient subset within a cell population.

Analogous to earlier experiments in the human-derived p53-tet cells (Fig. 5a), the selectivity by a P_Suv/P_M2 AND–NOT IFNγ-inducing circuit (v2) for p53-deficient cells were additionally tested in the mouse LLC cells and their p53⁺ derivatives (Fig. 5e). To this end, activating sgRNAs for mouse Ifng (two pre-screened sgRNAs cloned in tandem) was adopted. As expected, such a circuit selectively enabled the p53-deficient, parental LLC cells to robustly induce their expression of IFNγ and its target gene Irf1 (Fig. 5e).

The P_Suv/P_M2 AND–NOT circuit (v2) empowers immune rewiring of p53-deficient tumors, driving inhibition of tumor progression in vivo

To determine the specificity and activity of the above logic-gated Ifng-inducing gene circuit (v2) at a global level, we performed RNAseq analyses on circuit-transfected LLC cells and their p53⁺ derivatives. The two cell lines showed differences in basal levels of hundreds of genes, most of which possessed a grouped pattern of higher expression in the p53⁺ cells (Supplementary Fig. 6a, top red box). A number of classical p53 targets were identified in this group (Supplementary Fig. 6b)⁵⁴, consistent with the reconstitution of p53 activity in the knocked-in cells. Notably, the logic circuit strongly induced a different gene set in the parental, but not the p53⁺ cells (Supplementary Fig. 6a, green box). The principal component analyses also illustrated such a parental cell-selective regulatory program by the logic circuit (Fig. 6a). A closer examination confirmed the induction of Ifng. Importantly, the vast majority of the circuit-induced genes are known downstream targets of IFNγ (Fig. 6b, annotations in orange)⁵⁵, confirming the specificity of a CRISPRa-driven transcriptional rewiring. Overall, this gene set shows strong enrichment for biological functions such as cytokine actions, immune responses, and antigen presentation/processing (Supplementary Fig. 6c). These unbiased analyses demonstrate that our AND–NOT logic circuit (v2) can empower a highly specific and effective tumor recognition/immune rewiring axis with therapeutic implications.

Effective therapeutic tumor targeting by a gene circuit is likely to require a viral vector-mediated delivery method. As a prototype for viral circuit delivery, we constructed a mix of two lentiviral vectors (non-fluorescent, LV_NF) incorporating the P_Suv/P_M2-directed IFNγ-inducing circuit [v2] (Supplementary Fig. 6d, left illustration). As expected, in a co-culture system with p53-deficient and -sufficient LLC cells (see Fig. 5d), the virally delivered circuit also exhibited an evident cell-state selectivity and drove a much greater induction of IFNγ in the p53-deficient subpopulation [mCherry⁺] than in its p53-sufficient counterpart (Supplementary Fig. 6d, right panel).

Tumor cells introduced with the gene circuit through lentiviral vectors would enable investigations regarding the therapeutic potential of the circuit in an in vivo context. For this line of investigation, additional development was made to adapt the IFNγ-inducing circuit v2 into another mix of two lentiviral vectors with fluorescent labels (LV_FL), which would allow the assessment of transduction efficiency. As the circuit was divided into two units of LV_FL, full circuit delivery to the cells would be indicated by co-labeling of markers [mCherry and EGFP] (Supplementary Fig. 6e, left illustration). A circuit with a non-targeting sgRNA was used as the negative control. With a single round of transduction in culture, about 6% of LLC cells, and to the same extent, the p53⁺ cells were introduced with the complete functional circuit (Supplementary Fig. 6e, the mCherry⁺EGFP⁺ cells in the sgIfn groups). As expected, the circuit-transduced parental LLC cells displayed a robust IFNγ signal (Supplementary Fig. 6f). On the other hand, no changes in the in vitro growth characteristics were observed for these particular cells (Supplementary Fig. 6g). Subsequently, the circuit-transduced LLC and p53⁺ cells were transplanted subcutaneously into syngeneic mice (C57/BL6) for assessment of tumor progression in vivo. Importantly, the circuit-introduced parental LLC tumors, but not their p53⁺ counterparts, showed clear inhibition in progression (Fig. 6c). A few (two out of ten) tumors in this group even failed to become palpable. Since the circuit-induced IFNγ did not affect the growth of these cells in vitro (see above), such effects in vivo is conceivably attributed to stimulation of anti-tumor immunity. Consistent with this notion, in tumor samples harvested at the endpoint (day-17), the circuit-introduced parental LLC group showed substantial increases in not only the direct immunostimulatory targets of the IFNγ axis (e.g., STAT1 and MHC molecules), but also markers for T cell activation and their cytotoxic function, including the mRNAs for CD25, CD69, Granzyme B and Perforin (Fig. 6d). Interestingly, the levels of Cd8a mRNA, indicative of the abundance of CD8⁺ T cells did not show similar changes in these samples. Flow cytometry analyses of tumor tissues at day-7 and day-17 also failed to reveal significant changes in the numbers of CD8⁺ T cells in response to the circuit activities (Supplementary Fig. 6h). Therefore, the associated enhancement of cytotoxic T cell activities, rather than their abundance, is most likely to underlie the therapeutic effect by our p53 loss-gated, IFNγ-inducing circuit. Collectively, these results from different levels suggest the potential of our AND–NOT logic-controlled CRISPRa circuit as a useful tool for tumor-specific delivery of immunological interventions.