Context-specific synthetic T cell promoters from assembled transcriptional elements

Abstract Genetic engineering of human lymphocytes for therapeutic applications is constrained by a lack of transgene transcriptional control, resulting in a compromised therapeutic index. Incomplete understanding of transcriptional logic limits the rational design of contextually responsive genetic modules 1 . Here, we juxtaposed rationally curated transcriptional response element (TRE) oligonucleotides by random concatemerization to generate a library from which we selected context-specific inducible synthetic promoters (iSynPros). Through functional selection, we screened an iSynPro library for “IF-THEN” logic-gated transcriptional responses in human CD8 + T cells expressing a 4-1BB second generation chimeric antigen receptor (CAR). iSynPros exhibiting stringent off-states in quiescent T cells and CAR activation-dependent transcriptional responsiveness were cloned and subjected to TRE composition and pattern analysis, as well as performance in regulating candidate antitumor potency enhancement modules. These data reveal synthetic TRE grammar can mediate logic-gated transgene transcription in human T cells that, when applied to CAR T cell engineering, enhance potency and improve therapeutic indices.

A CAR-responsive synthetic promoter could serve this purpose by harnessing the natural signal transduction of T cell antigen stimulation to drive supplemental transgene expression.The simplest synthetic promoters contain repeated endogenous transcription factor response elements (TREs) [19][20][21][22][23][24] to amplify signals from endogenous signal transduction pathways.For example, promoters composed of repeat Nuclear factor of activated T-cells (NFAT) TREs have been used to drive exogenous cytokine secretion upon TCR or CAR activation but suffer from leakiness and limited transcriptional activity in the ON-state 22,25,26 .Screens of repeated promoter elements have identi ed constitutive and inducible promoters in cell lines 27,28 and stem cells 24 .More complex synthetic promoters could be constructed by combining varied TREs upstream of a minimal (core) promoter 29 , better mimicking natural promoters that utilize combinations of TREs to regulate gene expression by integrating activity of multiple signaling pathways.While these studies interrogated modestly sized libraries created from pre-de ned TREs combinations, we reasoned an agnostic search of synthetic promoters formed by combinatorial ligation of selected TREs may e ciently reveal candidates more precisely tuned to de ned cellular contexts.
Here, we screen a library of synthetic promoters constructed by a random ligation of T cell activationspeci c TREs for candidates that exhibit stringent OFF-state ON-state switching.By encoding the library within a lentiviral vector (LVV) pool and conducting a screen in the context of CAR T cell tumor challenges, we identify transiently active promoters in therapeutically relevant contexts.A supporting computational framework allowed us to identify clusters of promoter behavior in response to antigen stimulation, revealing TRE combinations associated with speci c gene expression patterns.This approach selected inducible synthetic promoters (iSynPros) that exhibit stringent OFF states and strong induction following in vitro antigen stimulation.iSynPro induction was speci c to antigen stimulation, as exposure to co-stimulation alone, Toll like receptor agonists, or stimulatory cytokines failed to activate transcription.When integrated into CAR T cells delivered in vivo to tumor bearing mice, the selected synthetic promoter iSynPro1 showed robust inducibility that was spatially and temporally restricted to antigen presence.iSynPro1-driven expression of potency-enhancing synthetic transgenes enhanced CAR T cell function and prevented off target toxicity in vivo.This paradigm of rational formulation and random assembly leverages existing clinical CAR designs to enhance T cell potency through regulated expression of supplemental transgenes and has the potential to be applicable to other emerging cellular therapies.

RESULTS
Generation of an inducible synthetic promoter library.
We posited that random ligation of rationally selected TREs would generate a library of cell state responsive synthetic promoters.This library could then be screened for prescribed transcriptional outputs in a cellular context of choice -here we explore "IF-THEN" transcription in antigen-activated CAR T cells (Fig. 1a).We rst identi ed genes upregulated in CD8 + T cells within 12 to 48 hours of T cell activation (drawn from clusters I, II, III and VI in Best et al. 2013).Next, transcription factor (TF) binding sites within the promoters of representative genes were identi ed by cross-referencing the TRED 30 and TRANSFAC 31 databases, with eleven TF binding sites appearing in both.We interrogated the JASPAR database 32 to identify TRE sequences associated with each of the eleven TFs to serve as functional units of our promoter library.Promoters were assembled by random ligation of pooled double-stranded oligonucleotides encoding TREs.Ligation reactions included a 3x amount of the nuclear factor-κB (NF-κB) TRE and a 1x amount of all other TREs, as CARs bearing a 4-1BB domain are known to engage NF-κB 33,34 .Following gel electrophoresis (Extended Data Fig. 1a), DNA products between approximately 200 and 500 base pairs were recovered and cloned into a LVV transfer plasmid upstream of an IL-2 minimal promoter (IL2mp) and a green uorescent protein: re y luciferase fusion (GFP: uc).The resulting plasmid library was packaged into LVV, yielding a functional titer of 1.16 x 10 9 infectious units per mL, corresponding to 2.2 x 10 8 total infectious units.
We retained DNA samples from the TRE ligation product ("raw"), the cloned plasmid library ("plasmid") and cDNA from the LVV ("virus") for evaluation by next generation sequencing (NGS).The variable length templates and absence of an alignment reference present a challenge for NGS analysis, which is exacerbated by the block structure of sequences constructed from a limited set of TREs.Accordingly, we devised a computational approach for recovery and analysis of promoter sequences (Extended Data Fig. 1b,c).A rst module, alignMerge, reconstructs the underlying sequence templates by rst sliding and concurrently aligning each read against its pair.A second module, TREsCaller, annotates each reconstructed synthetic promoter with TREs from our starting set while accounting for sequencing noise.We used the resulting promoter sequences, rendered as strings of TREs (i.e.TRE contigs) for downstream analysis.
The raw library diversity, as assessed using Lorenz distributions, proved highly polyclonal with a vast majority of promoter sequences occurring only once (Extended Data Fig. 1d).This likely re ects the large potential diversity of synthetic promoters created by our method, of which we were only able to sample a small proportion.The plasmid library showed a skewing in clonality, possibly due to differential clonal expansion after transformation into bacteria.This skewing was retained in the lentiviral library, as demonstrated by strongly correlated TRE contig frequencies among the lentiviral and plasmid libraries (Extended Data Fig. 1e).The lentiviral library showed enrichment of shorter promoters (as measured in TRE units), compared to the plasmid library (Extended Data Fig. 1e,f), likely re ecting a lentiviral packaging bias toward shorter plasmid sequences.
To assess the gene regulatory potential of TRE-based synthetic promoters in bulk, we stimulated human Jurkat T cells with PMA and ionomycin following transduction with our lentiviral promoter library driving a GFP reporter (Fig. 1b).In comparison to Jurkat T cells transduced with a GFP reporter driven by repeated NFAT responsive elements (NFAT-RE), TRE-based synthetic promoters achieved similar levels of induced GFP expression at 24 hours (Fig. 1c).A time course showed GFP expression as early as four hours post-stimulation (Fig. 1d).We observed re-induced GFP expression upon a second stimulation demonstrating some members of the library achieved repeat inducibility (Extended Data Fig. 2a-c).This response demonstrates our library contains inducible synthetic promoters (iSynPros).

Selection of antigen-responsive iSynPro clones in CD19(4-1BB)CAR T cells
To select the antigen responsive iSynPros within the iSynPro library, we co-transduced CD8 + T cells with the iSynPro LVV library and a separate LVV encoding a CD19-targeting CAR 9 (Fig. 2a).We rst sorted the resulting T cell population for CAR + GFP -T cells, thereby excluding promoter sequences with constitutive expression.To identify synthetic promoter sequences that exhibit increased transcription following antigen exposure, the sorted CAR + GFP -population was co-cultured with CD19 + tumor cells for CARspeci c stimulation.GFP + T cells were recovered by ow sorting 24, 48, and 72 hours after an initial tumor challenge and 24 hours after a second tumor challenge one week later.We isolated DNA from recovered T cells and performed targeted NGS to identify inducible promoter sequences as described previously (Fig. 2b).
We rst investigated whether the frequency of individual TREs within the promoter library changed during the time-course.We found that, compared to the proportion of TREs in the reference libraries (raw, plasmid, virus), individual TRE proportions among tumor activated GFP+ CD19CAR T cells exhibited low variation throughout our screen (mean coe cient of variation = 0.108, range 0.021 -0.305, Extended Data Fig. 3a).We next cataloged the abundance of synthetic promoters among inducible GFP + CAR T cells in response to tumor co-culture across timepoints.In contrast to TRE proportions, which did not change during the course of our screen, we found many promoters decreased in frequency at later time points, while few others increased (Extended Data Fig. 3b).After discarding minimally detected promoters (< 4 counts per million at each time point post tumor co-culture) we projectedpromoter frequencies in multi-dimensional space, linking promoter abundance across collection time points.We found a subset of promoter sequences were dominant throughout, while others were abundant only within one or several time points (Fig. 2c).We also found longer synthetic promoters, harboring more TRE units, were overrepresented in the cell populations compared to virus (Fig. 2d).Together these data indicate selective preference for additive or synergistic TRE combinations and suggest that promoter frequency is re ective of promoter expression dynamics.
To further examine the potential role of sequence complexity within individual promoters, we studied changes in frequencies of each TRE contig across all collection points in relation to the lentiviral library (taken as a baseline).Fold-change time course trajectories revealed evidence of clustering, as evaluated by the Hopkins statistic (H = 0.91, p < 10 -16 ) [35][36][37] .Accordingly, we clustered the dynamic behaviors of individual TRE contigs to reveal broad patterns (Fig. 2e).While some promoter sequences were present at similar frequencies across all timepoints, others exhibited increased or decreased abundance following antigen stimulations.
We next explored whether individual or combinations of TREs were enriched within behavioral clusters.For a handful of clusters, we found enrichment of individual TREs (Fig. 2f), while for others we did not.Some TREs enriched in solo were also enriched as parts of a combinatorial motif.Concurrently, there were other enriched motifs that did not have TRE sub-components enriched individually (Fig. 2g).These ndings suggest that TREs, both individually and in combination, form functional units underlying synthetic promoter behavior patterns.Enrichment of complex TRE motifs within clusters indicate synergism of individual TREs re ective of a "TRE grammar".The kinetic clusters and inherent TRE signatures could serve as a resource to inform design of promoters with behaviors appropriate to speci c cell therapy applications.

Recursive and persistent antigen activates iSynPro1
We identi ed 24 TRE contigs that were (i) present at a baseline read frequency in the viral library of at least 0.001 (i.e. 1 in 1000 reads), (ii) exhibited a fold change of >1.25 at multiple time points during cell selection and (iii) were among the most abundant TRE contigs (comprising 90% of the cell library) (Supplementary Table 2) .We the generated lentiviral vectors encoding GFP: uc reporters driven by either individual iSynPros, control promoter sequences composed of only NFAT or NF-κB responsive elements, or an "empty" IL2mp-only construct.Finally, manufactured CD8+ CAR T cell products via co-transduction of a second generation CD19CAR lentivirus and iSynPro or control lentiviral vectors and then assessed iSynPro induction potential following antigen exposure.Eight of eight iSynPro sequences tested (numbered in Supplementary Table 2) induced GFP expression upon each recurrent exposure to CD19 + LCL and returned to baseline levels after two weeks (Fig. 3a, Extended Data Fig. 4a).Induction was also observed for the NF-κB control promoter, but not for the IL2mp, nor the NFAT control promoters.Because individual iSynPros showed a range of dynamic behaviors, we ranked promoters in order of decreasing baseline expression and increasing peak expression of GFP (Fig. 3b,c).Several iSynPros showed lower background expression and higher peak expression than the NF-κB promoter.Among sequences with the desired features of low background and high induction potential, one sequence, identi ed as iSynPro1 (Extended Data Fig. 4c), exhibited the most stringent OFF state and greatest degree of transcription induction.
We further investigated the iSynPro1 transcriptional response to both pulsed and persistent in vivo antigen exposure.In a rst model we adoptively transferred iSynPro1-GFP: uc/CD19CAR CD8 + T cells into NOD scid gamma (NSG) mice, then injected irradiated Raji cells (a CD19 + human B lymphoma line) or PBS as a control 16, 26, and 40 days later (Fig. 3d).Following the initial injection of Raji tumor cells, mice harboring iSynPro1-GFP: uc CD19CAR T cells showed signi cant increases in luciferase activity at 24 h (P = 0.000599), 48 h (P = 0.00011), and 72 h (P = 0. 0.003269), compared to mice injected with PBS alone (Fig. 3e).Five days following tumor injection, increases in bioluminescence imaging were no longer signi cant (P = 0.08).Imaging at similar timepoints following subsequent tumor injections revealed continued OFF-ON-OFF luciferase activity patterns and minimal uctuation in response to PBS (Fig. 3e).
We further assessed the response of iSynPro1 to prolonged antigen exposure in vivo, using a model that creates a slowly progressive intraperitoneal lymphoma despite CD19CAR T cell in ltration.In this model, iSynPro1-GFP: uc/CD19CAR CD8 + T cells are adoptively transferred to NSG mice, followed injection of CD19 + Raji tumor cells in numbers su cient to result in tumor engraftment (Fig. 3f).We tracked iSynPro1 bioluminescence before and after adoptive transfer of Raji tumors and observed increasing iSynPro1 signal over 41 days (Fig. 3g).At the end of the experiment, ex vivo ow analysis of tumor biopsies from mice showed sustained CD19 antigen presence (Extended Data Fig. 4d).Interestingly, one Raji-injected mouse showed attenuated iSynPro1 activity.Necropsy at end of study revealed that this mouse was tumor-free (data not shown), consistent with a model in which the CD19CAR T cells eliminated the tumor and subsequently iSynPro signal returned to its OFF state.In summary, persistent antigen exposure led to iSynPro1 activation over a prolonged time course.Our ndings with both transient and persistent tumor exposure a rms that iSynPro1 responds to antigen-mediated CAR and TCR signaling.iSynPro1 induction is restricted to signal 1 T cell activation pathways Antigen encounters promote immunoreceptor tyrosine activation motif (ITAM) phosphorylation transmitting a T cell activating stimulus known as "signal 1" 38 .We investigated whether signal 1 was responsible for iSynPro1 transcriptional induction.We therefore correlated mRNA and protein abundance in CAR T cells harboring iSynPro1-driven GFP: uc following antigen stimulation.To ensure equal gene dosage of CD19CAR and iSynPro-GFP: uc modules, we utilized transposase-based integration of a single dual-promoter construct housing iSynPro1-regulated GFP: uc and a constitutively expressed CD19CAR (iSynPro1-GFP: uc/CD19CAR).Live cell uorescent imaging revealed iSynPro1 GFP induction occurred within 6 hours of stimulation, and transcriptional output positively correlated with antigen abundance (Fig. 4a).We also observed iSynPro1 induced expression of the GFP: uc transgene, following overnight CD3/CD28 stimulation by ow cytometry (Extended Data Fig. 5a).Because accumulation of stable GFP proteins can mask rapid transcriptional off rates, we assessed transcript levels by reverse transcriptase digital droplet PCR.We found induction of iSynPro1 within one day post stimulation, with a decrease in expression over the next 8 days while transcript levels of the constitutively expressed CD19CAR did not uctuate (Extended Data Fig. 5b).Therefore, iSynPro driven transcription induction is rapid and limited to periods of several days following stimulation.
Antigen stimulation of T cells triggers a set of signal transduction/gene expression pathways leading to T cell activation and effector function, some of which overlap with signaling pathways activated by receptors of the innate immune system.To explore the speci c signaling inputs necessary or su cient for iSynPro1 activation, we exposed iSynPro1-GFP: uc/CD19CAR T cells to a set of stimulatory ligands and chemicals.We found T cells receiving a stimulus through CD3ζ (via TCR-CD3 complex or CAR stimulation) resulted in iSynPro1 induction, and that this activation could be abrogated by addition of dasatinib, an inhibitor that blocks activity of SRC family kinases involved in proximal TCR/CAR signaling (including LCK) 39,40 (Fig. 4b, Extended Data Fig. 5c).T cells activated with PMA and ionomycin (PMA/Iono), which bypasses the membrane-proximal signaling cascade of TCR engagement, elicited iSynPro1 induction resistant to dasatinib inhibition.Co-culture of CD19CAR T cells with CD19 -tumor cells bearing CD80 and CD86 alone had no effect on iSynPro1-driven transgene expression, demonstrating that co-stimulation via CD28 is not su cient for iSynPro1 induction (Extended Data Fig. 5d).Likewise, compared to 7X NF-κB and 7X gamma-interferon activation site (GAS) promoters, iSynPro1 does not respond to stimulation with TLR5 agonist agellin, IFN, nor TNF .Because iSynPro1 activation is less responsive to these stimulatory molecules, it offers the bene t of restricting gene expression to areas and times of antigen exposure (Fig. 4c).These results show LCK-dependent pathways activated TCR/CAR signaling via CD3ζ are su cient for iSynPro1 induction, while costimulatory signals and in ammatory cytokines are not.
While our studies thus far have featured the CD19CAR, we hypothesized that other CARs featuring the CD3ζ domain would likewise lead to iSynPro1 induction.Coculture of antigen expressing tumor cells with two additional CAR speci cities currently under clinical investigation (B7H3- 41 and EGFR-targeted CAR T cells 42 ) also resulted in iSynPro1 activation (Extended Data Fig. 5e).Thus, we nd that activation of iSynPro1 is both CAR speci city-agnostic and likewise tumor-agnostic, as iSynPro1 induction in CD19CAR T cells occurs upon exposure to a variety of human tumor cell lines transfected to express CD19 (Extended Data Fig. 5f).These data are consistent with a model in which iSynPro1 responds via signal 1.
The restriction of iSynPro1 transcriptional activation to antigen-stimulated T cells may also allow spatial regulation of gene expression in vivo.To test the localization of iSynPro1 activation, we assessed iSynPro1 activity in in NSG mice bearing subcutaneous SK-N-Be(2) tumors (human neuroblastoma, hereafter referred to as Be2) modi ed to express CD19 (Fig. 4d).We administered CD8 + iSynPro1-GFP: uc/CD19CAR T cells intravenously fteen days after injection of control CD19 -Be2 tumors into the left ank and CD19 + Be2 tumors into the right ank of NSG mice.To evaluate the spatial activation of iSynPro, bioluminescent imaging was conducted every four to seven days.We observed luminescence (Fig. 4e) and GFP expression (Extended Data Fig. 5g) within the right ank CD19 + tumor, but not the left CD19 -ank tumor, demonstrating the iSynPro1-driven gene expression is restricted to anatomical areas with antigen presence.An iSynPro1-driven transgene improves CAR T cell function.
Within the eld of CAR T cell therapy, there is an unmet need to develop T cell products with potency enhancements.Potency enhancement of CAR T cell products could be mediated by promoters featuring tumor-responsive activity so that effects of transgenes are swiftly deployed upon tumor encounter and extinguished after tumor clearance.We designed a PD1:MyD88 transgene with dual functionality to abrogate CAR T cell deactivation by PD1 ligands and supplement CAR T cell activation via MyD88 signaling 43,44 (Fig. 5a).We then assessed whether this novel immune switch receptor would be an effective potency enhancement 45,46 when regulated by logic-gated iSynPro1 transcription.
We rst evaluated transcriptional pro les of CD19 CART cell products harboring either iSynPro1-PD1:MyD88 or iSynPro1-GFP by performing RNAseq on sorted CD4+ and CD8+ T cell products prior to and following an in vitro tumor challenge (Extended Data Fig. 6 a-d).Among differentially expressed genes we detected high induction of the PD1:MyD88 or GFP transgenes (as expected).Within iSynPro1-PD1:MyD88 containing CAR T cells, we also detected upregulated expression of IFNG, IL1B, IL2 and CSF2, suggesting enhanced effector function and supporting the immunomodulatory role of iSynPro1-PD1:MyD88.We therefore used PD1:MyD88 as a model payload to evaluate potency enhancement by a deployed iSynPro1 regulated transcription module.
We designed a set of dual-promoter piggyBac transposon-based vectors incorporating CD19CAR together with iSynPro1-driven PD1:MyD88 or TurboGFP (Fig. 5b).Our vector design included the 2A separated marker tags appended to CD19 CAR (EGFRt) and PD1:MyD88 (HER2tG) to track constitutive and iSynProdriven expression via ow cytometry.Following in vitro manufacture, T cell products demonstrated constitutive expression of the CAR cassette and activation-dependent induction of the iSynPro1 module, as measured by HER2tG (Fig. 5c, Extended Data Fig. 7 a,b).We reasoned that because MyD88 signaling activates NF-κB motifs, which are present in iSynPro1, a potential self-sustaining positive feedback loop may emerge.We therefore reassessed the stringency of iSynPro1 regulation following introduction of the PD1:MyD88 transgene.Following a CD3/CD28 stimulation, we found that iSynPro1 ON-OFF state transitions were similar when driving expression of PD1:MyD88 or a control, TurboGFP, indicating that iSynPro1 maintains stringent expression control in the context of expressing PD1:MyD88 (Fig. 5d).
To understand the potency enhancing effects of iSynPro1 regulated PD1:MyD88, we subjected T cells to a series of in vitro tumor challenges every three days.We utilized live cell uorescent imaging to monitor CAR T cell cytotoxicity and proliferation in response to challenges with CD19t + /mCherry + Be2 tumor cells (Extended Data Fig. 7c).Tumor tracking revealed that the addition of iSynPro1 regulated PD1:MyD88 expression resulted in marked reduction in tumor signal 3 days after the third tumor challenge whereas control CD19CAR T cells lost capacity to inhibit tumor outgrowth (Extended Data Fig. 7d).
Simultaneously, iSynPro1-regulated PD1:MyD88 CAR T cells exhibited increased survival and proliferation in response to repeated tumor exposures (Extended Data Fig. 7e).
We further evaluated the impact of iSynPro1-regulated expression of PD1:MyD88 in an in vivo tumor rechallenge model in which CD19CAR T cells were challenged two successive exposures to subcutaneous CD19 + Be2 tumors (Fig. 5e).Following engraftment of the initial tumor, mice subsequently treated with iSynPro1-PD1:MyD88 CD19CAR T cells controlled both initial and secondary tumor challenges (Fig. 5f), resulting in markedly improved survival (Fig. 5g).These ndings provide con rmation that iSynPro1regulated PD1:MyD88 is a CAR T cell potency enhancing genetic module.These results also provide a basis for the design of potency-enhanced CAR T cell products wherein the synthetic module is deployed selectively by CAR T cells that have tra cked to tumor and engaged target antigen.iSynPro1 regulation of a mitogenic potency enhancement prevents T cell overgrowth.
CAR T cell potency enhancement via transgenic pro-growth and pro-survival signals increases the risk of T cell mediated toxicity.Our group has developed a ligand-autonomous STAT inducer protein that provides constitutive IL-7 and IL-21 signals (LASI-7/21) , including activation of STAT5 and STAT3 (Fig. 6a).LASI-7/21 incorporates a constitutively active IL7R mutant originating in T cell leukemia 47 and an appended IL21R peptide.Transposon mediated delivery of the LASI-7/21 transgene into primary human T cells under control of constitutive and iSynPro promoters (Extended Data Fig. 8a) achieved constitutive (Fig. 6b) and antigen-dependent expression of LASI-7/21 (Fig. 6c).
Continuous STAT5 and STAT3 signaling delivered by LASI-7/21 may contribute to CAR T cell growth 48 and survival 49 , respectively.To assess this possibility, we examined the effects of CD19CAR T cells bearing constitutively expressed LASI-7/21 against the human leukemia cell line Nalm-6 in NSG mice (Fig. 6d).Constitutive expression of LASI-7/21 led to improved tumor clearance (Fig. 6e), however 60% of mice became moribund after tumor clearance (Fig. 6f), suggesting LASI-7/21 caused T cell overgrowth and subsequent toxicity.A CAR T cell lymphoproliferative disorder driven by LASI-7/21 is further suggested by necropsy ndings that included enlarged and discolored livers and spleens (Extended Data Fig. 8b).
We hypothesized that signal 1-dependent expression of LASI-7/21 by iSynPro1 could ensure that T cell growth enhancement is self-limited upon tumor clearance.To test this possibility, we assessed the safety and e cacy of CAR T cells bearing iSynPro1-regulated LASI-7/21 in vivo (Extended Data Fig. 8c).Indeed, iSynPro1-regulated LASI-7/21 enhanced CAR T cell tumor clearance, achieving undetectable Nalm-6 levels in four of ve mice (Fig. 6g, Extended Data Fig. 8d).Remarkably, despite e cacy improvements by iSynPro1-regulated LASI-7/21, no adverse health events or deaths were observed in mice following tumor clearance (Fig. 6h).This resulted in signi cantly improved survival compared to mice treated with CAR T cells bearing iSynPro1-regulated marker-only control (Extended Data Fig. 8e).Examination of iSynPro1regulated LASI-7/21 CAR T cells in the peripheral blood of mice revealed T cell numbers decreased after tumor clearance (Extended Data Fig. 8f).Among circulating human T cells, iSynPro1 limited LASI-7/21 expression to 1.8% of cells at day 35, compared to 74.6% of cells in constitutive LASI-7/21 controls (Extended Data Fig. 8g).Taken together, we conclude that iSynPro1 regulation enables simultaneous CAR T cell functional enhancements by a mitogenic transgene while preventing toxicity due to unrestricted T cell growth.

DISCUSSION
Endogenous promoters achieve ne-tuned temporal and stimulation-responsive regulation of gene expression via the combinatorial effects of cis-and trans-acting regulatory elements 50,51 .Stoichiometric and spatial relationships of DNA-bound transcription regulators integrate with protein activity states and other factors to specify gene expression levels 52,53 .Currently, incomplete understanding of TRE combinatorial logic precludes rational design of compact promoters responsive to T cell receptor antigen stimuli.Our approach circumvents incomplete knowledge of the complex architecture governing eukaryotic gene expression to permit de novo identi cation of promoter element combinations with desirable regulatory patterns.
Accordingly, we hypothesized libraries of synthetic promoters composed of randomly arrayed regulatory elements could yield synthetic promoters that impart transcriptional regulation speci c to contextually de ned cellular inputs.We constructed a promoter library through random assembly of T cell associated regulatory elements and tuned a high throughput screen to identify promoters with features attractive for application in CAR T cells including activation state dependence, stringent OFF states, a tuned range of magnitude and kinetics, and recursive state switching.Therefore careful selection of screening parameters ensured the applicability of regulated synthetic promoters to sophisticated CAR T designs.
The ndings from our screen can guide future endeavors to build synthetic promoters.We found that concatemerization of curated TREs that bind transcription factors mediating speci c cell status modules resulted in a diverse library of sequences that could be selected in a cellular context speci c for intended applications.This strategy of rational formulation and random assembly, followed by functional selection in context has potentially broad applications for cell type and context-speci c derivation of synthetic promoters that regulate therapeutic transgene expression.
We investigated the mechanism and context dependence of iSynPro1 activation when deployed in CAR T cells.We showed that iSynPro1 induction depends on signal 1 of T cell activation and is unresponsive to other T cell activation inputs.The quiescence of iSynPro1 following addition of ligands that activate NF-κB signaling (despite the presence of NF-κB TREs) demonstrates the speci city of transcriptional regulation achieved by TRE combinations.Further, the requirement for signal 1 to activate iSynPro1 restricts the potential of payload expression due to other environmental stimuli.Indeed, iSynPro1 activity in vivo only occurred within antigen-bearing tumors, highlighting the spatial control of transgene expression afforded by this system.
The synthetic promoter described here may also serve as a tool in other contexts requiring the linkage of gene expression to T cell activation.First, iSynPro1 could be used as a reporter for T cell activation, thereby streamlining screening processes for identi cation of antigen-speci c T cell clones or elimination of tonically signaling CAR designs.Second, utilization of iSynPro1 as a gate within a cellular logic circuit could trigger expression of a secondary and more promiscuous CAR, which would otherwise pose the risk of off-tumor toxicity.Third, regulated expression could limit safety risks posed by constitutive production of immunomodulatory cytokines such as IL12, IL18, and IL15, which exhibit promising preclinical T cell anti-tumor anti-tumor potency enhancements [3][4][5] .Our approach therefore is a general method to equip CAR T cells with regulated expression of elements that augment anti-tumor function.
Subsequent studies may address some of the shortcomings of this approach.Our synthetic promoter library was constructed from 11 T cell activation-speci c TREs ligated as contiguous blocks.Increasing library complexity could better mirror naturally occurring promoters, for example by inclusion of variable spacing between TREs, inclusion of a wider set of TREs, or by including TREs from orthogonal gene sets.To overcome the signi cant clonal skewing we observed during incorporation of our ligated 'raw' library into plasmids, repeated ligation/cloning reactions could permit evaluation of additional promoter candidates.Nevertheless, despite the sparse exploration of the large numbers of potential promoters that could result from random ligation, isolation of many suitable iSynPros suggests our library design approach e ciently produced promoter sequences capable of repeated transcriptional induction.
The work presented here demonstrates the utility of synthetic promoters assembled from candidate TREs identi ed using analysis of gene expression and transcription factor networks.Such a strategy yields a rich library and eliminates the need for exhaustive design-build-test iterations.Moreover, the library when formulated as a lentiviral pool allowed for at-scale selection in primary human T cells.Feature optimization of parameters such as ON state magnitude, OFF state stringency and kinetic induction of thousands of unique clones can be explored for particular use cases.While our work has focused on engineering therapeutic T cells, this strategy should allow discovery of novel regulatory logic operable in a wide variety of potentially therapeutic primary cells, organoids, or iPS-derived cell lines for subsequent clinical deployment.

Declarations
CD4 + or CD8 + T cells were resuspended to 2 to 4 x 10 6 cells / mL in RPMI medium supplemented with cytokines as above and stimulated with Dynabeads™ Human T Activator CD3/CD28 (Invitrogen) at 1:1 ratio overnight.The activated T cells were added with protamine sulfate (40 ug/ml, APP Pharmaceuticals)) and then transferred into a 12-well plate.Viruses were thawed, vortexed and added to each corresponding well.For co-transduction, both viruses were added to the cells at certain ratios.For example, in iSynPro library screening, CD19CAR-T2A-EGFRt virus and the iSynPro-IL2mp-GFP: uc library virus were co-transduced at MOI of 2 and 0.1 respectively.The plate was then transferred to 37 o C after spinoculation for 30 minutes at 800 x g.Equal volume of warm complete RPMI media supplemented with cytokines was added to each well at 4 hours or overnight post transduction.Six days post transduction, CD3/CD28 beads were removed from the cells and the transduction e ciency was checked by ow cytometry analysis of cell-surface marker or GFP expression.
For T cells transduced with CD19CAR-T2A-EGFRt-T2A-DHFRdm, methotrexate (MTX) was used to select transgene-containing cells.MTX was added to the T cell culture medium at 50 µM 6 days after transduction and supplemented when medium was changed.Transduced T-cells were expanded by stimulation with irradiated (8000 rad) TM-LCL at a 1:7 E:T ratio in the presence of 50U/mL IL-2 (CD8 + ), 5ng/mL IL-7 (CD4 + ) and 1ng/ml IL-15.The in vitro and in vivo assays were performed 10 days after T cell expansion.

Sorting of CD8/iSynPro-GFP/CAR T cells using FACS
Eight days after and iSynPro-GFP: uc co-transduction, CD8 + T cells were sorted for EGFRt + GFP -population by FACS to exclude cells containing constitutively expressing synthetic promoters.
Four days later, these T cells were washed with RPMI medium to remove the cytokines and co-cultured with irradiated Tm-LCL cells at 1:2 ratio.The cell suspension was harvested separately at 24, 48 and 72h after co-culturing, resuspended in 1X PBS+10%FBS and ltered (30 µm lter).The GFP positive population was sorted out by FACS.The sorted cells were centrifuged down and stored at -80 o C until DNA extraction.

NGS library preparation
Libraries were generated from DNA samples prepared from the plasmid, virus and GFP sorted cell populations (Fig. 1b and 2a) for sequencing on the Illumina Miseq platform as follows.DNA was extracted using QIAamp DNA Micro Kit (Qiagen #56304) and concentration measured by Qubit DNA HS assay (Thermo Fisher Scienti c #Q32851).A rst round PCR was performed for targeted ampli cation of the synthetic promoter library (refer sequence schema below).First round PCR primers comprised of (1) a locus-speci c region (letters in 'bold red') that matched invariant bases immediately adjacent to the 5' and 3' ends of the variable synthetic promoter region (letters in italics), and (2) a linker oligo (green overhang) to permit subsequent conjugation of Illumina adapter sequences.For e cient library recovery, a touchdown PCR protocol was employed with KAPA HiFi HotStart Ready Mix (KAPA Biosystems #KK2601).Each reaction used 100 ng of DNA template based on our PCR optimization result.Samples with material in excess of 100 ng were split into several PCR reactions.Resulting products were combined and puri ed by Agencourt AMPure XP (Beckman Coulter #A63880).
From 15ng of each rst round PCR reaction, dual index amplicon libraries were generated by adding Illumina Index-Adapter sequences to the DNA templates.This was accomplished by a second round PCR with KAPA HiFi HotStart Ready Mix (KAPA Biosystems #KK2601) using Illumina Nextera-XT Index primers that incorporate linker oligos at their 3' end (Illumina #FC-131-1001).The shared linker oligo between primers of both rounds of PCR facilitate conjugation of the Index-Adapters to the DNA templates.
Completed libraries were puri ed with Agencourt AMPure XP beads (Beckman Coulter #A63880) and quanti ed using Qubit DNA HS assay (Thermo Fisher Scienti c #Q32851).Separately, a PCR-Free amplicon library was prepared from 100ng of puri ed DNA sourced from the raw ligation product with the Accel-NGS 2S DNA Library Kit (IDT #10009877).This PCR-Free library was quanti ed by qPCR using the KAPA Library Quant Kit (Roche Diagnostics #KK4824).Following each PCR round, amplicon libraries were quality assessed by Agilent's 2200 TapeStation D5000 assay (Agilent #5067-5588).

Sequencing
All sequencing was performed on an Illumina MiSeq with MiSeq Reporter v2.5.1 (Illumina), using Illumina's v3 reagents (#MS-102-3003) and following manufacturer's instructions.Our libraries featured amplicons with stretches of monotemplate sequence (~30 bases) at both ends juxtaposing variable synthetic promoter sequences in between -this is problematic for Illumina sequencers and can result in poor quality reads.As a diversity countermeasure, 10% PhiX Control v3 Library (Illumina FC-110-3001) was supplemented for sequencing.From this attempt it was evident signi cantly higher PhiX levels would be needed to improve read quality, in turn sacri cing usable read throughput.This prompted a search for custom Read 1 and 2 primers that would bypass the monotemplate sequence in our libraries.
We identi ed a suitable candidate for use as Read 1 primer (refer previous sequence schema) but were unable to design a custom Read 2 primer counterpart.For Read 2, we relied on the Illumina supplied sequencing primer mix, that primes off the 3' linker.Illumina sequencer color matrix calibrations that are based on the initial bases of Read 1 are of primary importance -thus we proceeded with the custom Read 1 primer (GCGTTTTGCGCTGCTTCGCGATCGA) and to work in conjunction with it, a custom PhiX Read 1 Primer (ACACTCTTTCCCTACACGACGCTCTTCCGATCT).We spiked 10% PhiX Control to support improved quality for Read 2. Such an approach greatly improved the overall quality of both Read 1 and Read 2. The raw library templates which lacked terminal monotemplate bases representative of plasmid backbone sequences and were directly ligated to Illumina adapters, were accordingly sequenced with the default Illumina sequencing primers.PhiX was added as a quality control for cluster generation, sequencing, alignment and matrix calibration.Sequencing con guration was 300 bp paired end.Read recovery is listed in Supplementary Table 3.

Transient and repeated stimulation of CD19CAR + iSynPros-GFP: T cells
For repeated in vitro stimulation, T cells were stimulated with CD3/CD28 Dynabeads overnight and cotransduced with CD19CAR and iSynPros-GFP: uc.On day 7, beads were removed, and a sample of the cells was analyzed for EGFRt and GFP expression by ow cytometry.The remainder of the cells were continuously cultured for 7 days, stimulated with irradiated (8000 rad) TM-LCL at 1:2 ratio for 16 to 18 hours and analyzed by ow cytometry.This procedure was repeated two more times on 14-day intervals.
In vitro Incucyte assay Incucyte (Sartorius) live imaging was used to track the cytotoxicity, proliferation, and iSynPro1 driven GFP expression CAR T cells.CD19t + /mCherry + Be2 tumor cells and CAR T cells were co-cultured in triplicates in 384-well imaging plates.The number of tumor cells and CAR T cells that were seeded varied depending on the E:T ratios of interest.For the serial re-challenge experiment, subsequent doses of Be2 tumor cells were added every 3 days for a total 9 days.Images were captured every 4-hours at 10x or 20x magni cation.IncuCyte 2020C software was used for analysis.Cytotoxicity of CAR T cells, represented by the target cell presence, was quanti ed by RCU.GFP expression upon iSynPro1 induction was quanti ed by GCU.Cytotoxicity and iSynPro1 induction were assessed using the Basic Analyzer analysis.Proliferation of CAR T cells was quanti ed using Non-Adherent Cell-by-Cell analysis.
Dasatinib assay CD4 + and CD8 + CAR T cells expressing iSynPro1-GFP: uc were co-cultured with unmodi ed, OKT3 + , or CD19t + K562 target cells at a 1:1 E:T ratio, or stimulated with PMA/Ionomycin, with or without Dasatinib at 60 nM.After 16 hours, cells were harvested and analyzed by ow cytometry.

Jurkat reporter assay
Jurkat reporter cell lines were generated via lentiviral transduction to contain either 7x NF-κB, 6x NFAT, 7x GAS or iSynPro1 driving GFP expression.Cells were cultured in a base media of RPMI supplemented with 10% FBS and 1% glutamine and cell stimulants, TNFα 50 ng/mL (Bio-Techne), IFNγ 50 ng/mL (Bio-Techne) and PMA/Ionomycin 1x (Thermo Fisher), and TLR5 agonist, agellin 1.25 ug/mL (InvivoGen) for 16 hours at 37°C.7xNF-κB served as a positive control for TNFα, and 7xGAS for IFNγ.After the stimulation period, cells were harvested and analyzed by ow cytometry for GFP expression.As the PMA/Ionomycin stimulated condition presented the maximal GFP expression for all Jurkat lines, relative percent induction was calculated by dividing the GFP positive percentage for each stimulation condition by the GFP positive percentage induced by PMA/Ionomycin stimulation in each cell line.

RNA kinetics study
Cell pellets were collected and spun at 300 X g for 7 minutes, media aspirated and frozen at -80 o C.
Pellets were processed for RNA using a RNeasy Mini kit (Qiagen), cDNA was synthesized using a SuperScript™ IV Reverse Transcriptase (ThermoFisher), and cDNA was quanti ed using the QX-200 droplet digital PCR system (Bio-Rad) according to the manufacturer's recommendations.Primer and probes sets are detailed in Supplemental Table 5.
Bioluminescent imaging was performed weekly by (i.p.) injection of 4.29 mg/mouse D-luciferin (Xenogen), anesthesia by iso urane and imaging 10 minutes post D-luciferin injection using the IVIS Spectrum Imaging System (Perkin Elmer).Luciferase activity was analyzed using Living Image Software Version 4.3 (Perkin Elmer) and photon ux was analyzed within regions of interest.Systemic Nalm-6 Model: 11-13 week old NSG mice were injected with one million human Nalm-6 leukemia cells via the tail vein, modi ed to express a fusion protein of mCherry and re y luciferase.Six days after tumor injection, tumor engraftment in each mouse was quanti ed via bioluminescent imaging as described above.Mice were then assigned to treatment groups to equalize average tumor engraftment across groups.Seven days after tumor injection, mice were systemically injected via the tail vain with two or four million T cells.Thereafter, mice were monitored for tumor progression by bioluminescent imaging.Mice were euthanized when they showed moderate to severe hind-limb paralysis, an effect of leukemia progression, or otherwise as recommended by veterinary staff.Retro-orbital blood samples were taken on a weekly basis to track T cell engraftment in the peripheral blood.
T Cell Tracking by Retro-Orbital Bleeds.Beginning at day 17 after tumor injection, peripheral blood retroorbital bleeds of mice were taken on a weekly basis, and ow cytometry was performed on blood samples to quantify T cell engraftment.First, samples were subjected to red blood cell lysis using Pharm Lyse Buffer (BD, Cat.# 555899) and treated with Fc blocking reagent (Miltenyi, Cat.# 130-059-901) to prevent indiscriminate antibody binding.Next, cells were stained with the following panel of reagents: anti-CD3, anti-human CD45, anti-mouse CD45, and xable viability stain 520 to discriminate live human T cells.Finally, samples were xed in 0.5% paraformaldehyde (Electron Microscopy Sciences, Cat.# 15713) in PBS before analysis.CountBright absolute counting beads (Invitrogen, Cat.# 2207530) were then added to each sample to allow for calculation of T cell concentrations in analyzed blood.

Statistical analyses
Data wrangling and visualization were conducted using Prism (GraphPad software), R (www.rproject.com)and RStudio (www.rstudio.com).The number of replicates or T cell donors are indicated in gure legends.Comparisons of means between two groups was conducted using two-tailed t-tests.
Enrichment of TREs within promoter clusters was evaluated using one-sided (alternative="greater") Fisher's test (see Supplemental Methods for additional details).Survival comparisons following adoptive transfer of tumor xenografts used log-rank (Mantel-Cox) tests.The number of mice are indicated in the gure legends.For all tests, p < 0.05 was considered signi cant and was corrected for multiple comparisons.Exact values of test statistics, degrees of freedom, variance assumptions, and power are listed in Supplementary Table 5.   q values are denoted: *, q < 0.01, **, q < 0.001; exact q values available in Supplemental Table 5. Illustrations observed in d and g were generated using BioRender.p values are denoted: ****, P < 0.0001.Exact P values available in Supplemental Table 5 Illustrations observed in a and e were generated using BioRender.

Figure 1 Random
Figure 1