Design and Characterization of the RAS Sensor.

(a) Design of the binding component of the RAS sensor. Inspired by natural RAF (left), the RAS-binding component of the sensor (right) comprises RAS binding domain (RBDCRD), a linker, and a NarX-derived transphosphorylation domain. (b) Schematic of the RAS sensor composition and mechanism of action. The sensor’s genetic payload is encoded on four plasmids. Two plasmids express the RAS-binding components: RBDCRD fused, respectively, to NarXN509A or NarXH399Q. The third plasmid expresses NarL-VP48, and the fourth plasmid encodes the output protein (mCerulean) under the control of a NarL response element (NarL-RE) in front of a minimal promoter (TATA). Upon RAS activation, the RBDCRD domain of the RAS-binding components bind to RAS-GTP. This binding leads to a forced dimerization of the NarX domains and a transphosphorylation of NarXN509A, in turn phorsphorylating NarL. Phosphorylated NarL binds its response element on the output plasmid, inducing the expression of the output protein. (c) Sensor activation by mutated RAS. The bar chart shows output expression in HEK293 cells co-transfected with the RAS Sensor and either KRASG12D or KRASWT. (d) Dose-response curve and dependence of the RAS sensor on functional RAS binding. Output expression of RAS sensors with either RBDCRD wild-type (blue) or RBDCRD with R89L (red), C168S (green), or both (purple) mutations. The dashed line represents conditions where the NarX-fusion plasmids were replaced with a non-coding plasmid (control). (e) Dependence of sensor output on RAS levels. Output expression of the RAS sensor measured with increasing amounts of KRASG12D (blue), KRASWT (orange), or negative control (black) plasmid. (f) Input-output curve. Correlation of the output expression with the RAS-GTP levels in HEK293 cells, measured by a luminescence RAS-pulldown ELISA assay. To alter RAS-GTP levels, the cells were transfected with different amounts of either KRASG12D (blue), KRASWT (orange), or KRASWT + Sos-1 (purple) plasmids. Pearson’s correlation is shown as R2. (g) Generalizability across RAS variants. Output expression of the RAS sensor when co-transfecting increasing amounts of different RAS isoforms and mutants. mCerulean output expression was measured by flow cytometry and normalized to a constitutively expressed mCherry transfection control. Mean values were calculated from biological triplicates. Error bars represent +/- SD. Significance was tested using an unpaired two-tailed Student’s t-test. ****p < 0.0001.

Mechanism of Action.

Effect of differential RAS activation on the steps considered necessary for RAS Sensor activation. RAS activation in HEK293 cells was manipulated by co-expressing KRASG12D, KRASWT, Sos-1 (a guanine nucleotide exchange factor that activates endogenous RAS), or NF1 (a GTPase-activating protein that deactivates endogenous RAS). In the control condition the cells are transfected with a non-coding plasmid, here it represents the endogenous RAS activation. Schematics on top of the graphs illustrate how and what part of the Mechanism was investigated. (a) Expression levels of the RBDCRD-NarX-SYFP2 fusion protein measured by flow cytometry in the presence of various KRAS modulators (x axis labels). (b) RAS binding of the RBDCRD-NarX-SYFP2 fusion protein approximated by calculating the ratio of membrane to total SYFP2 signal for each cell. Intracellular localization of SYFP2 was measured using confocal microscopy. The micrographs below show representative images for each condition. Scale bars = 50 µm. (c) Dimerization of the NarX fusion proteins assessed by transfecting two complementary RBDCRD-NarX-split.mVenus fusions and measuring the mVenus fluorescence by flow cytometry. (d) Output expression after transfection with the full RAS Sensor measured by flow cytometry. In a, c & d the fluorescent signals were normalized to a constitutively expressed transfection control. Each symbol represents one biological replicate (a: n=9, c-d: n=3). The error bars represent +/- SD. In b the fluorescence at the membrane was normalized to the total fluorescence for each cell. Violin plots in b represent 560 (KRASG12D), 322 (KRASWT), 482 (Control), 226 (NF1), or 1194 (Sos-1) cells from three biological replicates. Significance was tested using an ordinary one-way ANOVA with Dunnett’s multiple comparisons to compare each condition with the control condition (endogenous RAS activation). *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Tunability of the RAS sensor.

(a) 3D structure of the RAS Sensor dimerizing at the membrane. The structure of the RBDCRD-NarX fusion proteins (orange and yellow) was predicted using AlphaFold and aligned with existing NMR structures of RBDCRD (blue) bound to a KRAS-dimer (green) at the membrane. The ATP-binding site N509 (purple) and the phosphorylation site H399 (pink) are highlighted as spheres. On the top RBDCRD and NarX are fused with a 1x GGGGS linker and on the bottom with a 6x GGGGS linker. (b) Effect of different linkers in the RBDCRD-NarX fusion protein of the RAS Sensor. The bar charts show the output expression in HEK293 co-transfected with KRASG12D (blue) or KRASWT (orange) when using different numbers of repeats of a flexible (GGGGS) or a rigid (EAAAK) linker in the RBDCRD-NarX fusion proteins. The heatmap below shows the corresponding fold change between output expression in cells with KRASG12D and KRASWT. (c) Effect of different binding domains (BD) fused to NarX in the RAS Sensor. The heatmap shows the output expression in HEK293 co-transfected with 15 ng/well of KRASG12D (blue), KRASWT (orange), or NF1 (green), a GTPase-activating protein that deactivates endogenous RAS. The bars above show the corresponding fold changes between cells with KRASG12D and KRASWT (black) or KRASG12D and NF1 (green). (d) Effect of different transactivation domains (TAD) fused to NarL in the RAS Sensor. The heatmap shows the output expression in HEK293 co-transfected with KRASG12D (blue), KRASWT (yellow) or NF1 (green). The bars above show the corresponding fold changes between cells with KRASG12D and KRASWT (black) or KRASG12D and NF1 (green). mCerulean output expression was measured using flow cytometry and normalized to a constitutively expressed mCherry transfection control. Mean values were calculated from three (b) or two (c-d) biological replicates. Error bars were calculated using error propagation rules.

Design of Multi-Input RAS-targeting Circuits.

(a) Schematic of the RAS-targeting circuit with an AND-gate between mitogen-activated protein kinase (MAPK) sensors and the direct RAS Sensor. Dimerization of RAS activates the MAPK pathway and its downstream transcription factors. These transcription factors bind the synthetic response elements (RE), expressing the parts of the RAS Sensor. The RBDCRD-NarX proteins then bind activated RAS, dimerize, and propagate the signal to NarL, leading to output expression. The logic diagram of the resulting coherent feed-forward loop with AND-gate logic is shown in the dotted box. (b) Schematic of the transcription factor binding sites present in the response elements. Multiple repeats of the binding sites were placed upstream of a minimal promoter (TATA) driving expression of the RAS Sensor parts. (c) Expression levels with the MAPK response elements. The heatmap shows the direct expression of mScarlet of the different REs in HEK293 cells co-transfected with 15 ng/well of either KRASG12D (blue), KRASWT (orange), or NF1, a protein that deactivates endogenous RAS (green). The bars above show the corresponding fold changes between cells with KRASG12D and KRASWT (black) or KRASG12D and NF1 (green). (d) RAS-dependency of the SRE-, pFos-, and PY2-response elements. Direct mScarlet expression of the response elements in HEK293 cells co-transfected with different amounts of KRASG12D (blue), KRASWT (orange), or non-coding control plasmids (red). (e) RAS-dependency of the Elk-RE when additionally overexpressing Elk-1. RAS titration as described in d. (f) Dynamic range of the RAS-targeting circuits. Fold change in mCerulean output expression between HEK293 co-transfected with 1.67 ng/well of KRASG12D and KRASWT. In the RAS-targeting circuits, NarL-VP48 and/or the RBDCRD-6xfL-NarX fusion proteins were expressed using different MAPK-REs or a constitutive promoter (EF1a). Fluorescent protein expression was measured by flow cytometry and normalized to a constitutively expressed transfection control. Mean values were calculated from two (c-e) or three (f) biological replicates. PY2: polyoma virus enhancer domain; SRE: Serum response element; pFos: minimal promoter of c-fos; AP1: activator protein 1; Elk: Ets-like protein; Myc: myelocytomatosis protein. Detailed response element design is shown in Supplementary Fig. 5.

Modularity of the RAS-targeting Circuits.

(a) Screening of RAS circuit variants with different parts. mCerulean output expression in HEK293 co-transfected with 1.67 ng/well of KRASG12D versus KRASWT representing the ON-versus OFF-state of the screened circuits. The gray shading of the symbols represents the dynamic range (fold change between ON and OFF). Circuits with a high dynamic range (> 100+/-SD) are highlighted. The pie chart shows the prevalence of the response elements (RE) among the hits. (b) Prevalence of the circuit parts among the hits. On the left for all circuits where the RE expressed NarL (RE_NarL) and on the right where the RE expressed the NarX proteins (RE_NarX). TAD: transactivation domain, BD: binding domain. RA: Ras association domain of Rassf5 (c) Effect of different circuit parts on the output expression in HEK293 with KRASG12D (blue) and KRASWT (orange) fitted using a generalized linear regression. The EF1a expressed RAS Sensor with RBDCRD, a 6x flexible linker in the NarX fusion proteins, and VP48 as TAD fused to NarL was set as the reference sensor. The graph shows the model coefficients, which can be interpreted as the effect on output expression when a part in the reference sensor is replaced by the part indicated on the x-axis. (d) Expression of all circuit parts via response elements. Fluorescence histograms of mCerulean-positive cells comparing ON-(KRASG12D) and OFF-state (KRASWT) of RAS circuits when either NarL (blue), the NarX proteins (violet), or all parts (red) are expressed via REs. The tested circuits contain RBDCRD as BD, a 6x flexible linker and F.L.T. as TAD, the parts that led to the hits with the highest dynamic range. The table below shows the parts used in each of the tested circuit. (e) Microscopy images showing the mCerulean expression of the conditions from d. mCerulean output expression was measured by flow cytometry and normalized to a constitutively expressed mCherry transfection control. Each circuit was measured in three biological replicates. Error bars represent +/- SD. Significance in c was tested using the Wald test. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Translation into cancer cells – Detection of endogenous RAS levels in HCT-116.

(a) RAS Sensor activation in colon cancer cells. Microscopy images of the mCerulean output expression in HCT-116 wild-type cells harboring a homozygous KRASG13D mutation (HCT-116WT; top row) and HCT-116 KRAS knock-out cells (HCT-116k.o.; bottom row) transfected with the initial RAS Sensor (right) or a RAS Sensor with an R89L mutation in the Ras binding domain (left). (b-d) Effect of different circuit parts in colon cancer cells. Boxplots of dynamic range of different RAS-targeting circuits grouped by the circuit parts of interest they contain. The circuit parts investigated were: the response elements in b, the binding domain fused to the NarX proteins in c, and the transactivation domain fused to NarL in d. Each black circle represents a different RAS circuit. (e) Best performing RAS-targeting circuits in colon cancer cells. The parts used in each RAS circuit are listed in the table below. The bar graph shows the dynamic range, while the fluorescence histograms show mCerulean-positive cells obtained in the On-(HCT-116WT) and Off-state (HCT-116k.o.) of the circuits. mCerulean output expression was measured by flow cytometry and normalized to a constitutively expressed mCherry transfection control. Dynamic range was calculated as fold change between normalized output expression in HCT-116WT and HCT-116k.o.. Each circuit was measured in three biological replicates. Error bars in e were calculated using error propagation rules.

Translation into cancer cells – Selectivity for RAS-driven cancer cells.

(a) RAS-targeting circuits are classifiers for cells with mutated RAS. Output expression of RAS-targeting circuits in different cancer cell lines with (RASMUT = blue) or without (RASWT = orange) mutation leading to increased RAS activation. (b) Output expression of the best performing RAS-targeting circuits for each RASMUT cancer cell line. Each RASMUT cell lines is only shown in the circuit that performed best in the respective cell line. A boxplot of all RASWT cell lines is shown in each circuit and fold changes were calculated between the mean of the individual RASMUT and the mean of all RASWT cell lines. The colored symbols represent biological replicates of the different cell lines. The parts used in each RAS circuit are indicated in the tables below. mCerulean output expression was measured by flow cytometry and normalized to a constitutively expressed mCherry transfection control. Each circuit was measured in three biological replicates. Significance was tested using an unpaired two-tailed Student’s t-test. **p < 0.01, ***p < 0.001.

Killing of RAS-driven cancer cells.

The graphs show the overall confluence, or confluence of mCherry transfection control positive cancer cells, transfected with RAS-targeting circuits that express herpes simplex virus thymidine kinase (HSV-TK) as output protein or controls over time. The used RAS circuits (purple and green) and controls are described in the dotted box, with a RAS circuit expressing GFP as output as negative control without HSV-TK (turquoise) and a EF1a-expressed constitutive HSV-TK as positive control (red). Gray arrow at t=46h indicate the addition of the prodrug ganciclovir (GCV) that is activated by HSV-TK. Statistical significance and fold chances are calculated between each condition and the corresponding GFP-circuit control at t=180h. Representative microscopy images of both brightfield and mCherry confluence at the time of ganciclovir addition (t=46) and end of the measurement (t=180) are shown for each condition and cell line. (a) Overall confluence in the well of KRASG13D-mutated HCT-116 cells. (b) Confluence of transfected KRASG13D-mutated HCT-116 cells. (c) Microscopy images corresponding to a & b. (d) Overall confluence in the well of RAS wild-type Igrov-1 cells. (e) Confluence of transfected RAS wild-type Igrov-1. (f) Microscopy images corresponding to d & e. (g) Overall confluence in the well of KRASG12V-mutated SW620 cells. (h) Confluence of transfected KRASG12V-mutated SW620 cells. (i) Microscopy images corresponding to g & h. (j) Overall confluence in the well of KRASG13D-mutated HCT-116 cells transfected with lower (0.5x) amounts of the circuits. (k) Confluence of transfection control positive KRASG13D-mutated HCT-116 cells transfected with lower (0.5x) amounts of the circuits. (l) Microscopy images corresponding to j & k. Confluence was quantified from microscopy images. Mean confluence was calculated from biological triplicates and background adjusted to the EF1a_HSV-TK condition of the same cell line at t=46h. Standard deviation is shown as ribbons. Significance was tested using an ordinary one-way ANOVA with Dunnett’s multiple comparison test. ns = non-significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

RAS Sensor activation with RBD or RBDCRD as binding domain.

mCerulean output expression of the RAS Sensor with either RBD or RBDCRD of CRAF as binding domain in the NarX fusion proteins in HEK293 cells co-transfected with 15 ng of KRASG12D. mCerulean expression was measured using flow cytometry and normalized to a constitutively expressed mCherry transfection control. Mean values were calculated from three biological replicates.

RAS Isoforms and Mutants.

(a) Comparison of oncogenic KRAS mutants. Output expression of the RAS sensor when co-transfecting increasing amounts of different KRAS mutants. The dashed line represents wildtype KRAS (b) Bar chart of output expression at 1.67 ng KRAS plasmid/well from a. (c) Comparison of RAS isoforms. Output expression of the RAS sensor when co-transfecting increasing amounts of KRAS (blue), NRAS (green) or HRAS (red). The solid line represents G12D mutant, dashed line the corresponding wildtype isoform (d) Bar chart of output expression at 0.56 ng plasmid/well of different RASG12D Isoform from c. (e) Bar chart of output expression at 1.67 ng plasmid/well of different RASG12D Isoform from c. (f) Bar chart of output expression at 15 ng plasmid/well of different wildtype RAS Isoform from c. mCerulean output expression was measured by flow cytometry and normalized to a constitutively expressed mCherry transfection control. Mean values were calculated from biological triplicates. Error bars represent +/- SD. Significance was tested using an ordinary one-way ANOVA with Dunnett’s multiple comparisons to compare each condition with KRASG12D (b, d & e) or KRASWT (f). ns = non-significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Prediction of MAPK transcription factor binding sites in EF1a.

(a) Elongation factor 1a promoter (EF1a) with localization of predicted binding sites of MAPK transcription factors downstream of RAS. The width of the symbols indicates the sequence length of the potential binding site. (b) List of all predicted binding sites of Myc, AP1 (Fos/Jun), SRF, CREB, and Ets/Elk transcription factors with LASAGNA 2.0 using the TRASFAC TFBS search. For each binding site the sequence, position in EF1a, strand, binding strength (score), statistical significance (p-value), and expected number of random hits with similar scores in the data set (e-value) are provided.

Contribution of RAS-dependent Increase in Sensor Expression Levels on total Output.

Comparison of the functional RAS sensor (red) with inactive non-functional (different shades of gray) and constitutively active sensor controls (shades of blue). The details of the controls are illustrated in the table below. Black: constitutive EF1a driven mCerulean expression; light gray: only output plasmid present; gray: EF1a-driven NarL-VP48 expression and output plasmid present; dark gray: EF1a-driven expression of RAS-binding impaired mutated RBDR89LCRDC168S-NarX fusion proteins and NarL-VP48 and output plasmid present; light blue: same plasmids as RAS sensor but RBDCRD in the NarX fusion proteins are replaced by FKB or FRBP, which constitutively dimerized in the cytosol (100 nM of A/C heterodimerized were added to induce full dimerization); dark blue: same plasmids as RAS sensor but the NarX fusion proteins were replaced by a full length NarX protein that constitutively dimerizes at the membrane. (a) Increase in absolute mCerulean signal (non-normalized) compared to 0 ng/well of KRASG12D. (b) Increase in relative mCerulean signal (= mCerulean output expression normalized to a constitutively expressed mCherry) compared to 0 ng/well of KRASG12D plasmid. Mean values were calculated from biological triplicates. Error bars represent +/- SD.

Design of the MAPK Response Elements.

(a) Detailed representation of the MAPK response elements design. Three repeats of the transcription factor binding site pattern with restriction enzyme cutting sites were placed in front of a minimal promoter (TATA). Exact sequences of the restriction sites, TATA, 3’ UTR, and Kosak are shown in italics. (b) Sequences of the transcription factor binding site of each repeat of the different response elements. The schematics show which transcription factors bind to which response element.

MAPK Response Element with overexpressed transcription factors.

mScarlet expression from plasmids with the different MAPK response elements when additionally overexpressing the MAPK transcription factors Elk-1, c-Jun, c-Fos, and c-Myc. Heatmap showing the normalized mScarlet expression in HEK293 cells co-transfected with 15 ng/well of either KRASG12D (blue), KRASWT (orange), or NF1, a GTPase-activating protein that deactivates endogenous RAS (green). Bars above show the corresponding fold changes between cells with KRASG12D and KRASWT in black or KRASG12D and NF1 in green. mCerulean expression was measured by flow cytometry and normalized to a constitutively expressed mCherry transfection control. Mean values were calculated from two biological replicates.

Comparison of the RAS-binding dependent AND-gate RAS circuits and MAPK response element expressed constitutively dimerized NarX-NarL TCS.

(a) Dynamic range of the RAS-targeting circuits. Fold change in mCerulean output expression between HEK293 co-transfected with 1.67 ng/well of KRASG12D and KRASWT. In the RAS targeting circuits, NarL-VP48 and/or the RBDCRD-6xfL-NarX fusion proteins were expressed using different MAPK-REs or a constitutive promoter (EF1a). (b) Output expression of RAS-targeting circuits with different response elements in HEK293 co-transfected with 1.67 ng/well of KRASG12D. (c) Output expression of RAS-targeting circuits with different response elements in HEK293 co-transfected with 1.67 ng/well of KRASWT. (d-f) Same set-up as in a-c, but the RBDCRD-NarX fusion protein sequence on the NarX plasmids were replaced by full length NarX that constitutively dimerizes at the membrane. Fluorescent protein expression was measured by flow cytometry and normalized to a constitutively expressed transfection control. Mean values were calculated three biological replicates. PY2: polyoma virus enhancer domain; SRE: Serum response element; pFos: minimal promoter of c-fos; AP1: activator protein 1; Elk: Ets-like protein; Myc: myelocytomatosis protein.

Differential circuit activation of the binding domains (BD) depends on the transactivation domain fused to NarL.

Boxplots of mCerulean output expression in HEK293 co-transfected with 1.67 ng of KRASG12D. The boxplots include all RAS-targeting circuits from the screening in Fig.5, where the NarX fusion proteins were expressed through the response elements split by the binding domain used in the circuit. (a) Effect of binding domain in circuits with F.L.T. as transactivation domain fused to NarL. (b) Effect of the binding domain in circuits with VP48 as transactivation domain. mCerulean expression was measured by flow cytometry and normalized to a constitutively expressed mCherry transfection control. Each of the boxplot shows all circuits from the screening that used the indicated binding and transactivation domain, resulting in a total of 18 circuits in a or 34 in b. For each circuit three biological replicates were measured. Significance was tested by one-way ANOVA with a Tukey’s multiple comparison. *p < 0.5, ***p < 0.001, ****p < 0.0001.

Effect of the RAS circuit parts: response elements (RE).

Boxplots of activation of all RAS-targeting circuits with either EF1a as promoter or pFos, Py2, or SRE as response element expressing different circuit parts. (a-c) Effect of response elements when expressing NarL-F.L.T. via REs, while the NarX fusion proteins are expressed with EF1a. (d-f) Effect of response elements when expressing NarL-VP48 via REs, while the NarX fusion proteins are expressed with EF1a. (g-i) Effect of response elements when expressing NarX fusion proteins via REs, while NarL is expressed with EF1a. a,d&g show the mCerulean output expression in HEK293 co-transfected with 1.67 ng of KRASG12D plasmid. b,e&h show the mCerulean output expression in HEK293 co-transfected with 1.67 ng of KRASWT plasmid. c,f&i show the fold change between cells with KRASG12D and KRASWT. mCerulean expression was measured using flow cytometry and normalized to a constitutively expressed mCherry transfection control. Mean values were calculated from three biological replicates. Each of the Boxplot shows all circuits from the screening that used the indicated response element to express the indicated part, resulting in a total of 4 circuits in a-c, 14 in d-f, or 48 in g-i. For each circuit three biological replicates were measured. Significance was tested using a one-way ANOVA with a Dunnet’s multiple comparison against the EF1a condition. *p < 0.5, **p < 0.01, ***p < 0.001.

Effect of the RAS circuit parts: binding domain (BD) and transactivation domain (TAD) in RE_NarX circuits.

(a-c) Effect of the binding domain. Boxplots show the mCerulean output expression in HEK293 of all tested RAS-targeting circuits where the NarX fusion proteins are expressed via MAPK response elements (RE_NarX circuits) split by the used binding domain. a shows the ON-state in cells co-transfected with KRASG12D, b the OFF-state in cells co-transfected with KRASWT, and c the dynamic range calculated as fold change between ON- and OFF-state. (d-f) Effect of the transactivation domain. Boxplots show mCerulean output expression of the same circuits as in a-c but split by the transactivation domain fused to NarL. d shows the ON-state, b the OFF-state, and f the dynamic range. (g-i) Effect of the transactivation domain only in circuits with RBDCRD as binding domain. Boxplots show mCerulean output expression in the circuits from d-f that use RBDCRD as binding domain. g shows the ON-state, h the OFF-state, and i the dynamic range. Each circle represents one circuit condition using one of the following binding domains: RA(Rassf5) (purple), K55 (cyan), or RBDCRD (blue). For each circuit three biological replicates were measured. mCerulean output expression was measured using flow cytometry and normalized to a constitutively expressed mCherry transfection control. Significance was tested using a one-way ANOVA with a Tukey’s multiple comparisons in a-c and unpaired two-tailed Student’s t-tests in d-i. *p < 0.5, **p < 0.01, ***p < 0.001.

RAS Sensor activation in HCT-116.

mCerulean output expression of the RAS Sensor in HCT-116 wildtype (blue) or KRAS knock out (orange) cells. mCerulean expression was measured using flow cytometry and normalized to a constitutively expressed mCherry transfection control. Mean values were calculated from three biological replicates.

Correlation of activation of the MAPK sensors and the RAS-targeting circuits in individual cell lines.

Output expression of the MAPK sensors or the RAS circuits when using pFos (red) PY2 (green) or SRE (blue) as response elements. For the MAPK sensors, the response elements directly expressed mScarlet. For the RAS Circuits, the response elements expressed NarL-F.L.T. leading to mCerulean output expression. a A549, b AsPC-1, c HCT-116, d HeLa, e HT-29, f Igrov-1, g K562, h LoVo, i MCF-7, j SKOV-3, k SW480, l SW620. Output expression was measured using flow cytometry and normalized to a constitutively expressed transfection control. Each symbol represents one biological replicate (n=3).

Comparison of RAS-targeting circuits in individual cancer cell lines with mutations overactivating RAS.

Output expression after transfecting different RAS circuits into various cancer cell lines with mutations overactivating RAS. The dashed lines show the median activation of the RASWT cell lines with the highest background for the corresponding circuit, namely Igrov-1 for RAS Sensor_F.L.T. (green), SRE_NarL-F.L.T. (purple) and PY2_NarL-F.L.T. (red), SKOV-3 for pFos_NarL-F.L.T. (blue) and HT-29 for PY2_all_F.L.T. (olive). As HT-29 has a BRAF mutation overactivating the RAS pathway, for PY2_all_F.L.T. HeLa, the highest RASWT cell line without MAPK mutation, is also shown as a dotted line (olive). (a) A549, (b) AsPC-1, (c) HCT-116, (d) K562, (e) LoVo, (f) SW480, (g) SW620 cells. The composition of the parts used in the different RAS circuits is shown in h. mCerulean output expression was measured using flow cytometry and normalized to a constitutively expressed mCherry transfection control. Mean values were calculated from three biological replicates.

Differences between cell lines and transfection amount in killing assay.

The graphs show the confluence of mCherry transfection control positive cancer cells (a,b,e,f) or overall confluence (c,d) transfected with a RAS circuit expressing GFP as output as negative control without HSV-TK (a-d) or a EF1a-expressed constitutive HSV-TK as positive control (e-f). Control circuits are illustrated in the dotted boxes below the graphs. Gray arrows indicate the timepoint of transfection and the addition of the prodrug ganciclovir (GCV) that is activated by HSV-TK. a Difference in transfection efficiency between cell lines measured as confluence of the constitutively expressed mCherry transfection control. b Difference in transfection efficiency in HCT-116 when transfecting different amounts of circuit DNA. c Different growth curves between cell lines. Growth was measured as overall confluence in the well; red = HCT-116 (0.5x DNA amount), blue = SW630, green = Igrov-1. d Growth curves in HCT-116 when transfecting different amounts of circuit DNA. Red = 0.5x DNA amount, Turquoise = 1x DNA amount. e Difference in killing curves between cancer cell lines when constitutively expressing herpes simplex virus thymidine kinase (HSV-TK). Killing of transfected cells is approximated using the confluence of mCherry positive cells. f Difference in killing curves in HCT-116 when transfecting different amounts of DNA. Confluence was quantified from microscopy images. Mean confluence was calculated from biological triplicates with standard deviation shown as ribbons. Significance was tested using an ordinary one-way ANOVA with Dunnett’s multiple comparison test. ns = non-significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Assessment of the non-transformed regression model.

(a-c) Histograms showing the distribution of the measured mCerulean output expression of all circuits tested in the screening in HEK293 overexpressing KRASG12D in a, KRASWT in b, or the fold change between KRASG12D and KRASWT in c. (d-f) Residuals plots showing the difference between measured and fitted output (residuals) vs the predicted output of HEK293 overexpressing KRASG12D in d, KRASWT in e, or the fold change between KRASG12D and KRASWT in f used to assess linearity and homoscedasticity assumptions. (g-i) Goodness of fit: Measurements of output vs predictions of the regression models based on the parts used in the circuit. Predicted values were calculated using the following general linear regression models. a,d,g show the ON-state data in HEK293 co-transfected with KRASG12D. Model formula: RASmut_RU ∼ RASmut_mCherry_AU + BD + RE_NarL_TAD + Linker + RE_NarX + Conc_NarX + Conc_NarL. b,e,h show the OFF-state in HEK293 co-transfected with KRASWT. Model formula: RASwt_RU ∼ RASwt_mCherry_AU + BD + RE_NarL_TAD + Linker + RE_NarX + Conc_NarX + Conc_NarL. c,f,i show the dynamic range calculated as fold change between ON- and OFF-state. Model formula: Ratio ∼ RASmut_mCherry_AU + RASwt_mCherry_AU + BD + Linker + RE_NarX + RE_NarL + Conc_NarX + Conc_NarL Gray lines in g-I display the linear best fit used for calculating the Pearson correlation. RASmut_RU: mCerulean signal with KRASG12D normalized to transfection control, RASwt_RU: mCerulean signal with KRASWT normalized to transfection control, Ratio: fold change between RASmut_RU & RAS_wt_RU, RASmut_mCherry_AU: absolute transfection signal with KRASG12D, RASmut_mCherry_AU: absolute transfection signal with KRASWT, BD: binding domain, Linker: linker used in NarX fusion protein, RE_NarL_TAD: response element used to express NarL and transactivation domain fused to NarL, RE_NarX: response element used to express NarX proteins. Conc_NarX: transfected amount of plasmid expressing the NarX proteins. Conc_NarL: transfected amount of plasmid expressing the of NarL protein.

Assessment of the regression model (log-transformed model used in

Fig.5). (a-c) Histograms showing the distribution of the measured mCerulean output expression of all circuits tested in the screening in HEK293 overexpressing KRASG12D in a, KRASWT in b, or the fold change between KRASG12D and KRASWT in c. (d-f) Residuals plots showing the difference between measured and predicted output (residuals) vs the predicted output of HEK293 overexpressing KRASG12D in d, KRASWT in e, or the fold change between KRASG12D and KRASWT in f used to assess linearity and homoscedasticity assumptions. (g-i) Goodness of fit: Measurements of output vs predictions of the regression models based on the parts used in the circuit. Predicted values were calculated using the following general linear regression models. a,d,g show the ON-state data in HEK293 co-transfected with KRASG12D. Model formula: log(RASmut_RU) ∼ RASmut_mCherry_AU + BD + RE_NarL_TAD + Linker + RE_NarX + Conc_NarX + Conc_NarL. b,e,h show the OFF-state in HEK293 co-transfected with KRASWT. Model formula: log(RASwt_RU) ∼ RASwt_mCherry_AU + BD + RE_NarL_TAD + Linker + RE_NarX + Conc_NarX + Conc_NarL. c,f,i show the dynamic range calculated as fold change between ON- and OFF-state. Model formula: log(Ratio) ∼ RASmut_mCherry_AU + RASwt_mCherry_AU + BD + Linker + RE_NarX + RE_NarL + Conc_NarX + Conc_NarL Gray lines in g-I display the linear best fit used for calculating the Pearson correlation. RASmut_RU: mCerulean signal with KRASG12D normalized to transfection control, RASwt_RU: mCerulean signal with KRASWT normalized to transfection control, Ratio: fold change between RASmut_RU & RAS_wt_RU, RASmut_mCherry_AU: absolute transfection signal with KRASG12D, RASmut_mCherry_AU: absolute transfection signal with KRASWT, BD: binding domain, Linker: linker used in NarX fusion protein, RE_NarL_TAD: response element used to express NarL and transactivation domain fused to NarL, RE_NarX: response element used to express NarX proteins. Conc_NarX: transfected amount of plasmid expressing the NarX proteins. Conc_NarL: transfected amount of plasmid expressing the of NarL protein

Illustration of gating in flow cytometry analyses.

Gating example in HEK293 cells transfected with a mCerulean expressing RAS circuit and mCherry transfection control (a) Live cells are gated by plotting all events on a side scatter area (SSC-A) versus forward scatter area (FSC-A) density plot. (b) Single cells are gated by plotting FSC-A versus FSC-height (FSC-H). (c) Real mCerulean signals are separated from false positive signals by plotting the mCerulean signal measured with two independent detectors against each other. Both measure the signal from the 445 nm excitation laser, only exiting mCerulean and not mCherry. Both have an emission filter that can strongly detect mCerulean. Detector one measures mCerulean with a 473/11 emission filter, while detector two measures mCerulean with a 515/20 emission filter. Only real mCerulean signals where the signal correlates between the detectors are inside the gate. (d) Plot showing mCerulean output signal (excitation at 445 nm, emission filter 473/10) versus mCherry transfection control signal (excitation at 561 nm, emission filter 610/20). mCerulean output expression is calculated by multiplying frequency of parent * mean of Q2, which represents mCerulean positive cells with high transfection (mCherry higher than 103). The mCherry signal is calculated by multiplying frequency of parent * mean of all cells with mCherry higher than 103 (Q2+Q3). 103 was chosen as threshold for transfection efficiency as cells below this rarely show circuit activation. (e) Binning makes the normalized Output signal less sensitive to different transfection amounts. Change in mCerulean output expression in HCT-116 cells transfected with 1x or 0.5x the plasmid amount of various RAS circuits to simulate different transfection efficiencies. All mCherry-positive cells were analyzed on the left (blue), while on the right the cells were binned for high transfection efficiency, and only cells with mCherry signal >103 were analyzed (red).

Correlation of Output Expression and Transfection Efficiency.

(a) Comparison of transfection efficiency in cancer cell lines with (blue) or without (orange) mutation overactivating RAS. (b) Correlation between normalized output expression and transfection efficiency in the different cancer cell lines. mCerulean output expression and mCherry transfection efficiency were measured using flow cytometry. Absolute fluorescence units (AU) were calculated as mean*frequency of parent. mCerulean output expression AU was normalized to the mCherry transfection control AU to get the relative fluorescence units [RU]. Both a & b contain all samples tested during the cell line screening. Significance was tested using an unpaired two-tailed Student’s t-test.

Cell line and culturing details.

Seeding and transfection conditions in different cancer cell lines (Fig.7 & 8d-l)

Seeding and transfection conditions for all experiments (Fig.1-6e) except the cancer cell line screening (Fig.7)

Excitation filter, dichroic mirror, and emission filter wavelengths used during fluorescence microscopy.

Excitation laser, long pass filter, and emission filter wavelengths used during flow cytometry.

Tested regression models to select the independent variables and assess the interaction between response element and expressed parts.

Numerical values from flow cytometry histograms in Fig.5d

Numerical values from flow cytometry histograms in Fig.6e

Segmentation parameters for image analysis using Agilent eSight in Fig.8