A RAS-Independent Biomarker Panel to Reliably Predict Response to MEK Inhibition in Colorectal Cancer

Simple Summary Today, clinical management for the majority of cancer patients is still based on a “one-size-fits-all” approach. To improve the outcomes in the era of personalized medicine, it is essential to stratify patients based on established and novel biomarkers. In the present study, we investigated a SMAD4 loss-of-function mutation, which is associated with chemoresistance and decreased overall survival in colorectal cancer (CRC). To investigate the molecular mechanism behind the impact on drug response, we used CRISPR technology on patient-derived organoid models (PDOs) of CRC. We showed that PDOs with loss-of-function SMAD4 mutations are sensitive to MEK-inhibitors. Using a novel four-gene signature reliably predicts sensitivity towards MEK-inhibitors, regardless of the RAS and BRAF status. The present study is a significant step towards personalized cancer therapy by identifying a new biomarker. Abstract Background: In colorectal cancer (CRC), mutations of genes associated with the TGF-β/BMP signaling pathway, particularly affecting SMAD4, are known to correlate with decreased overall survival and it is assumed that this signaling axis plays a key role in chemoresistance. Methods: Using CRISPR technology on syngeneic patient-derived organoids (PDOs), we investigated the role of a loss-of-function of SMAD4 in sensitivity to MEK-inhibitors. CRISPR-engineered SMAD4R361H PDOs were subjected to drug screening, RNA-Sequencing, and multiplex protein profiling (DigiWest®). Initial observations were validated on an additional set of 62 PDOs with known mutational status. Results: We show that loss-of-function of SMAD4 renders PDOs sensitive to MEK-inhibitors. Multiomics analyses indicate that disruption of the BMP branch within the TGF-β/BMP pathway is the pivotal mechanism of increased drug sensitivity. Further investigation led to the identification of the SFAB-signature (SMAD4, FBXW7, ARID1A, or BMPR2), coherently predicting sensitivity towards MEK-inhibitors, independent of both RAS and BRAF status. Conclusion: We identified a novel mutational signature that reliably predicts sensitivity towards MEK-inhibitors, regardless of the RAS and BRAF status. This finding poses a significant step towards better-tailored cancer therapies guided by the use of molecular biomarkers.


Introduction
Colorectal cancer (CRC) is the third most diagnosed cancer in the world [1]. The main cause of death in CRC is metastases [2]. At first diagnosis, 20% of colorectal cancer patients already have developed metastases, and more than 30% of early-stage patients will eventually develop metastases [3]. Making a successful treatment even more difficult, the metastatic disease is often refractory to systemic agents such as 5-fluorouracil (5-FU)-based therapies [4,5], so progression-free survival after first-line treatment did not significantly improve in the last decade [6,7]. The main reason for this unsatisfactory situation is that large cohorts of patients may benefit from guideline treatments, yet the individual tumors are so heterogenous that for most metastatic CRC patients, this "one-size-fits-all" approach does not culminate in a successful treatment [8,9]. This fact is fully accepted by the approving bodies (EMA, FDA); therefore, the development of companion diagnostic biomarkers is highly encouraged for future targeted treatment regimens.
Loss-of-function mutations of the SMAD4 gene are present in up to 15% of CRC cases [10]. They have been correlated with decreased overall survival [11,12] and a strong line of evidence suggests that functional loss of SMAD4 promotes chemoresistance in multiple cancer types, including CRC [13].
In the present study, we link SMAD4 loss-of-function mutations and the subsequent loss of the TGF-β/BMP signaling to increased sensitivity towards MEK-inhibitors of affected cells. More specifically, the BMP axis of this signaling pathway seems responsible for resistance towards MEK inhibition. This led to the identification of a mutational signature present in more than 30% of CRC cases as sufficient for predicting sensitivity to MEK-inhibitors independent of the RAS and BRAF mutational status of the tumor.

3 D Cell Culture
PDO cultures were generated and propagated as previously described [14,15]. Models used are listed in Table S2.

Semi-Automated High-Throughput Drug Response Assay
PDO cultures were digested with TrypLE Express (ThermoFisher Scientific, Waltham, MA, USA) to a single cell suspension. Single cells were recovered in 5 mL complete PDO medium with ROCK-inhibitor (Miltenyi Biotec, Bergisch Gladbach, Germany) for 15 min at 37 • C. Cells were embedded in Matrigel ® (Corning, Tewksbury, MA, USA) at a density of 1500 cells/well in 384-well plates (Greiner Bio-One, Frickenhausen, Germany) using a Biomek FX automated workstation (Beckman Coulter, Brea, CA, USA). Plates were incubated for 15 min at 37 • C to allow Matrigel ® solidification, and a 50 µL/well complete PDO medium with ROCK-inhibitor was added. PDOs were cultured until organoid sizes reached 50 µm in diameter prior to treatment. Subsequently, drug treatments were performed by removing the medium daily and adding 50 µL/well complete medium without ROCK-inhibitor, containing drugs in 4 replicates. Concentrations of the tested compounds were distributed around the c max (maximal plasma concentration) values observed in clinical settings. Cells treated with DMSO (Sigma-Aldrich, Steinheim, Germany) (0.3%) were used as "untreated control," while Matrigel ® only was used for background readout. Staurosporine was used as an internal positive control.
After 4-day treatment, the cell viability was evaluated by Cell Titer Glo ® (CTG) assay (Promega, Madison, WI, USA) following the manufacturer's protocol. The intensity of luminescence was measured using the SpectraMax i3x plate reader (Molecular Devices, San Jose, CA, USA) 30 min after the addition of the reagent. Luminescence intensity data were used to evaluate the percentage of viability after background subtraction. To interpret the data, organoids are defined as sensitive if the absolute IC 50 value is equal to or lower than the c max value and resistant if the absolute IC 50 value is higher than the c max value.

Genomic DNA and RNA Isolation
For DNA preparation, PDOs from 4-8 wells of a 24-well plate were collected, centrifuged (5 min, 300× g), washed in 1× PBS (ThermoFisher Scientific), and stored at −80 • C for further use. Genomic DNA was isolated using the Wizard Kit (Promega) according to the manufacturer's protocol. DNA was eluted in 30 µL RNAase-free water (Sigma-Aldrich) and stored at −20 • C for further use.
For total RNA isolation, PDOs from 12 wells of a 24-well plate were collected. PDOs were centrifuged (300× g for 5 min at 4 • C) and kept on ice for the following steps. Pellets were washed with 5 mL sterile ice-cold PBS and centrifuged. Cell pellets were resuspended in 600 µL ice-cold RLT buffer (Qiagen, Venlo, The Netherlands) supplemented with 1% β-mercaptoethanol. Samples were frozen at −80 • C for further use. Total RNA was isolated by ATLAS Biolabs GmbH (Berlin, Germany). Nucleic acids were quantified using the Nanodrop 2000 (ThermoFisher Scientific).

PCR
Primers were designed using the web-interface Primer3Plus. The PCR reaction mixture (25 µL) contained 100 ng isolated DNA, 1× Phusion GC buffer, 3% DMSO, 200 µM dNTPs, 0.5 µM of each primer (SMAD4_fw 5 -AAATTCTCAGTTGACCTGGTCC-3 and SMAD4_rev 5 -ACCGACAATTAAGATGGAGTGC-3 ), and 0.02 U/µL of Phusion DNA polymerase (all PCR reagents from ThermoFisher Scientific). Initial denaturation at 95 • C for 5 min was followed by 34 cycles of denaturation at 95 • C for 30 s, primer annealing at 58 • C for 30 s, and extension at 72 • C for 30 s in an automated DNA thermal cycler (Biometra TRIO Thermal Cycler Series, Analytik Jena AG, Jena, Germany). The final extension was performed at 72 • C for 5 min. The PCR products, expected to be 688 bp, were resolved on 2% agarose gel stained with ethidium bromide (Roth, Karlsruhe, Germany) and visualized under the UV trans illuminator (VWR ® International, Radnor, PA, USA). Molecular size markers (GeneRulerTM100bp DNA Ladder, ThermoFisher Scientific) were run concurrently. PCR products were purified using the QIAquick PCR purification Kit (Qiagen), according to the manufacturer's protocol.

CRC Panel Sequencing (DNAseq)
Targeted high-depth sequencing was performed for genomic areas comprising 100 genes using a custom-designed CRC panel (Table S1) [16]. From each sample, 10 ng of DNA were used for multiplexed PCR amplification with the Ion Ampliseq Library kit (ThermoFisher Scientific), using two amplicon pools per DNA sample. Samples were ligated to Ion Xpress Barcode Adapters (ThermoFisher Scientific) and purified using Agencourt AMPure beads (Beckman Coulter). All six samples were combined on a 316v2 chip and sequenced on an Ion Torrent PGM device (ThermoFisher) with an average read depth of 580-700.

Sanger Sequencing
Mutations found via panel sequencing were validated via direct Sanger sequencing of PCR amplicons using both forward and reverse SMAD4 PCR primers mentioned before. Aliquots of 14 µL 20 ng/µL purified PCR product were sent to LGC Genomics GmbH (Berlin, Germany) for sequencing. Electropherograms and FASTA files from Sanger sequencing were analyzed internally using CLC Genome Viewer (Qiagen).

Whole Transcriptome Sequencing (RNAseq)
PDOs were treated with trametinib (0.03 µM) or vehicle control DMSO (0.3%) for 24 h. Drug concentrations were chosen based on clinically achievable plasma concentrations [17]. RNA isolation and whole-transcriptome sequencing were performed at ATLAS Biolabs GmbH. An average of 1 µg total RNA per sample was used to generate barcode-labeled libraries using the Illumina TruSeq Standard mRNA Library preparation kit. Paired-end libraries were sequenced on an Illumina NovaSeq sequencer with the 125 bp reads protocol. The quality of raw RNA-seq reads was estimated with the following trimming of adapters and low-quality reads with the fastP [18] and MultiQC tools [19]. To generate gene expression data, RNA-seq reads were mapped to the latest version of Homo sapiens GRCh37 reference genome from the Ensembl collection [20] using STAR aligner [21]. Rsubread package [22] and featureCounts tool [23] were used to evaluate read counts and generate the expression data matrix. Further differential gene expression analysis was performed using the NOISeq package [24] with the pipeline customized for the replicate simulation analysis. Read counts were normalized using RPKM (Reads per kilobase per million mapped reads) [25] and TMM (Trimmed Mean of M-values) methods [26], low count values were filtered away, and the ARSyNseq function of the NOISeq package was used to filter out noise-associated results and unidentified batch effects. To increase the significance of differential expression probability, we set parameter q = 0.95 for the NOISeq-sim function. Functional enrichment analysis was done using parameter q = 0.9 to have a more representative and broader picture of affected pathways. Bioconductor packages org.Hs.eg.db (https://doi.org/10.18129/B9.bioc.org.Hs.eg.db, accessed on 13 October 2021) and Genomic Features [27] were used to annotate the results. GOannotation and functional enrichment analysis was implemented utilizing GOexpress [28] and EnrichR [29,30] packages.  Figure S2A-D). The plasmid was digested with BbsI (NEB #R3539) for 1.5 h at 37 • C. The mixture of linearized plasmid and inserts were incubated with T4 DNA ligase (NEB #M0202L) overnight at 16 • C, followed by transformation into E. coli. DH5α. Plasmid minipreps were performed according to the standard protocol (GeneJET Plasmid Miniprep Kit, Thermo Fisher Scientific). The resulting plasmid (pU6-SMAD4-I-CAG-Cas9-venus-bpA) was sequenced using U6 sequencing primer 5 -GAGGGCCTA-TTTCCCATG-3 and was further used for transfection. All primers and single-stranded oligonucleotides were synthesized by BioTEz (Berlin, Germany).

Cell Sorting
72 h after transfection, cells were collected, centrifuged at 300× g for 5 min, and digested in 500 µL TrypLE Express for 15 min. Single cells were washed in a complete PDO medium with ROCK-inhibitor and incubated for 15 min at 37 • C, washed in PBS, and filtered through a 35 µm cell strainer (Corning). Single cells were gated for BFP/Venus and sorted directly in 500 µL of complete PDO medium supplied with 10 µg/mL gentamycin (Lonza, Basel, Switzerland) using a BD Aria cell sorter (BD Biosciences, Franklin Lakes, NJ, USA). BFP/Venus-positive cells were embedded in Matrigel ® . After 3 weeks in culture, one-half of the cells was used for isolation of genomic DNA and performing genotyping PCR, while the other was used for expansion and clonal selection.

Limited Dilution
Cells were dissociated using TrypLE Express as previously described. A cell solution at a concentration of 5 cells/mL was prepared in a 10 mL complete PDO medium. 100 µL per well were transferred into a round 96-well plate (Greiner Bio-One) to achieve an average density of 0.5 cells/well. Individual colonies were allowed to grow until they could be genotyped individually to confirm gene editing using PCR and the pJet cloning method (Thermo Fisher Scientific #EF694056.1).

High-Content Western Blotting-DigiWest ®
DigiWest ® on PDO cultures was performed essentially as previously described [32]. For organoid generation, 3.0 × 10 6 cells per condition were plated. After 72 h, PDOs were treated with trametinib (0.03 µM) or vehicle control DMSO (0.3%) for 24 h. PDOs were isolated with cell recovery solution (Corning) according to the manufacturer's protocol and stored at −80 • C. For analysis, organoid pellets were lysed by the addition of 2× LDS lysis buffer (Life Technologies) containing Protease-Inhibitor Mix M (Serva, Heidelberg, Germany), PhosSTOP (Roche Applied Science, Mannheim, Germany), and PMSF (Ther-moFisher Scientific) and heating to 95 • C for 10 min; protein was quantified using a Pierce BCA Protein Assay Kit (ThermoFisher Scientific). The NuPAGE SDS-PAGE gel system (Thermo Fisher) was used for protein separation and blotting; 12 µg protein per sample was separated using 4-12% Bis-Tris gels and transferred onto PVDF membranes (Millipore, Billerica, MA, USA).
For the DigiWest ® procedure, relative protein expression was determined using the primary antibodies listed in Table S4 and signal intensity was determined from the integral of the analyte-specific peak. The values were normalized to beta-actin and data were presented relative to the DMSO controls.

Establishment of 5-FU Resistant PDO Models
PDO models were cultivated as described before. To establish 5-FU resistant PDO models, cells were grown in the presence of a sublethal concentration (9.2 µM) of 5-FU for 4 weeks. Surviving cells were allowed to recover and challenged further with 46.1 µM of 5-FU over 4 months. Subsequently, 5-FU concentration was increased to 230 µM (c max ) [33] for another 3 months. Before drug screening, 5-FU resistant PDOs, as well as the nonresistant counterparts, were cultivated and passaged in a complete PDO medium. 5-FU resistant PDOs were treated with 5-FU, regorafenib, cobimetinib, and trametinib and compared with their non-resistant counterparts.

SMAD4 Loss-of-Function Is Associated with Differential Drug Response
Intra-tumor heterogeneity poses a significant obstacle in cancer therapy. However, comparing different cell populations from the same tumor can be used to understand mechanistic complexity and subsequently identify prospective biomarkers. We have previously reported that PDO "sibling"-culture models established from separate regions of a primary chemo-naïve CRC tumor respond differently to drugs [15]. Here, we focused on two of these sibling PDO models (R1 R361H and R4 wt ), displaying multiple identical mutations (KRAS G12D , PIK3CA H1047R , and TP53 C242F ), but differ regarding a SMAD4 R361H mutation ( Figure 1A). As reported by others [34], PDO model R1 R361H carries the well-known loss-offunction mutation SMAD4 R361H [35] and tends to proliferate faster in comparison to R4 wt ( Figure 1B). The cell density was measured on day 2, 3, 5, and 7 by measuring cellular ATP content using Cell Titer Glo ® assay. Results from four independent experiments were combined (one-way ANOVA * p < 0.05). (C) Drug response curves of R4 wt and R1 R361H treated with 5-Fluorouracil (5-FU). (D) Scheme of the MAPK signaling pathway with interruption of signal transduction due to small molecules (mRTK-inhibitor: regorafenib, BRAF-inhibitor: dabrafenib, MEK-inhibitors: cobimetinib, trametinib, selumetinib, and ERK-inhibitor: ulixertinib). (E) Drug response curves of R4 wt and R1 R361H treated daily for 4 days with small molecules targeting the MAPK signaling pathway. Cell sensitivity in three biological and four technical replicates was defined according to IC 50 values in relation to the maximal plasma concentrations (c max ), which are indicated for each drug with dotted lines. In all panels: Nonlinear curve fit, comparison of best-fit values, extra sum-of-squares F Test p < 0.05. (See also Figure S1). There were no significant differences between the two PDOs in response to 5-FU, a pivotal drug in first-line therapy for CRC ( Figure 1C). Notably, the absolute IC 50 values for both models were very close to the c max values of 5-FU [33], reflecting the donor patient's poor response to FOLFOX/FOLFIRINOX in the clinic. Morphological changes observed in R1 R361H upon 5-FU treatment ( Figure S1) may signify epithelial to mesenchymal transition (EMT). In contrast, responses to targeted therapeutics, however, revealed drastic differences in sensitivity to MEK-inhibitors cobimetinib, trametinib, and selumetinib. The mutant PDO R1 R361H was 10-to 100-fold more sensitive towards the MEK-inhibitors trametinib and cobimetinib, respectively, compared to the SMAD4 wt counterpart R4 wt ( Figures 1E and S1). Whereas targeting the mitogen-activated protein kinase (MAPK) pathway upstream or downstream ( Figure 1D) of MEK resulted in similar values in both models ( Figures 1E and S1). Thus, the SMAD4 mutational status is very likely responsible for a differential sensitivity specifically to MEK inhibition.

Introduction of SMAD4 Point Mutation Reverses MEK-Inhibitor Resistance
To substantiate our hypothesis that loss-of-function SMAD4 mutations create sensitivity to MEK-inhibitors, we generated syngeneic PDOs harboring SMAD4 R361H . The R4 wt PDO model was CRISPR-engineered to introduce a single SMAD4 R361H mutation (Figures 2A and S2). Three such clones (clones A, B, and C) were established and the successful introduction of the SMAD4 R361H mutation was confirmed ( Figure 2B,C, Table S1). SMAD4 loss-of-function was associated with a significant increase in proliferation ( Figure 2D), confirming our previous observations ( Figure 1E). CRISPR-SMAD4 R361H PDOs became sensitive towards MEK inhibition but not other compartments within the MAPK pathway ( Figures 2E and S3). This further strengthens a SMAD4 loss-of-function driven increased sensitivity towards MEK-inhibitors that may be of use as a predictive biomarker for clinically successful MEK-inhibitor therapy in CRC.

BMP-Pathway Is Activated in Response to MEK Inhibition
To pin down the mechanism underlying the differential MEK-inhibitor sensitivity, we treated the sibling-PDOs and the CRISPR-SMAD4 R361H PDOs with trametinib (0.03 µM) or vehicle control DMSO (0.3%) for 24 h and performed RNA sequencing. In a Principal Component Analysis, all CRISPR-SMAD4 R361H PDOs and R1 R361H clustered together and were clearly separated from R4 wt ( Figure 3A). We found 618 genes affected in all SMAD4 R361H PDOs and 222 unique genes in R4 wt ( Figure 3B). Gene enrichment analysis of the SMAD4 R361H models revealed that among the significantly altered biological processes were DNA replication and cell cycle progression consistent with a MEK inactivationassociated growth arrest (Table S3). Moreover, affected in GO terms were signal transduction, specifically of the TGF-β/BMP signaling, especially in SMAD4 wt PDOs. Yet, DNA replication or cell cycle progression were not enriched (Table S3).
As a complementary approach, we subjected trametinib and DMSO-control treated sibling cultures to DigiWest ® analysis [32] (Figure 3C,D, Table S4). All SMAD4 mutated PDOs clustered together according to their protein profile ( Figure 3C). We observed a pronounced decrease in proliferation markers cyclin B1 and Aurora kinase A in SMAD4 R361H PDOs after 24 h trametinib treatment ( Figure 3D). While no differences in post-translational phosphorylation of MAPK pathway proteins, such as MEK and ERK, or p-ERK were observed ( Figure 3D) in R4 wt PDOs, phosphorylation of mTOR, SMAD5, and SMAD1/5 was higher after treatment with trametinib ( Figure 3D). These findings confirm the activation of the TGF-β/BMP pathway, specifically of the BMP branch, in response to exposure to a MEK inhibitor.  (clones A, B, and C). The cell density was measured on day 2, 3, 5, and 7 by measuring cellular ATP content using Cell Titer Glo ® assay. Results from four independent experiments were combined (one-way ANOVA * p < 0.05). (E) IC 50 values of R4 wt and R1 R361H and CRISPR-SMAD4 R361H PDOs (clones A, B, and C each in biological triplicates) treated daily for 4 days with small molecules targeting the MAPK signaling pathway. Achievable maximal plasma concentrations (c max ) are indicated for each drug with dotted lines. In all panels: ns = not significant; * p < 0.05, values as mean ± SEM using one-way ANOVA and Tukey's test for multiple comparisons. (Scale bars = 100 µm (A and B)). (See also Figure S3).  also Table S4).

SFAB Signature Predicts Sensitivity to MEK Inhibition in CRC PDOs
Next, we evaluated the eligibility of using the SMAD4 mutational status as a predictive biomarker for MEK-inhibitors in 62 PDOs with known mutational status (Table S2). In total, 9 PDOs harbor pathogenic SMAD4 mutations (stated in "The Human Gene Mutation Database"; HGMD ® [36,37]). All of these were sensitive to MEK-inhibitor cobimetinib, 6 were sensitive to trametinib, and 5 to selumetinib ( Figure S4). From the data generated on the sibling models, we would have expected that only PDOs with loss-of-function SMAD4 mutations were sensitive to MEK inhibition. However, a significant amount of SMAD4 wt PDO models were also MEK-inhibitor sensitive ( Figure S4).
To solve this conundrum, we looked at additional genes involved in TGF-β/BMP signaling that are frequently mutated in CRC. In addition to SMAD4, we identified FBXW7 (Fbox/WD repeat-containing protein 7) [38], ARID1A (AT-rich interactive domain-containing protein 1A) [39], and BMPR2 (bone morphogenetic protein receptor type II) [40] for the marker signature [41,42] (Figure 4A) as our SFAB-(SMAD4, FBXW7, ARID1A or BMPR2) signature. We found that at least one loss-of-function mutation is present in one of these genes determines the sensitivity towards MEK inhibition. The frequency of SFAB in the CRC patient cohort (TCGA n = 594; https://www.cbioportal.org, accessed on 18 August 2021) was comparable to the frequency of SFAB in our PDOs ( Figure 4B,C). For PDOs with SFAB-signature, we found up to 95% and 70% significant positive prediction for cobimetinib and selumetinib, respectively, and also up to 70% positive prediction for trametinib ( Figure 5). Thus, the SFAB signature can be used to reliably predict response to MEK inhibition.
We further investigated if SFAB is superior to the RAS status of CRC PDOs in predicting sensitivity to MEK inhibition. About 53% (33/62) of the CRC PDOs carry a pathogenic KRAS mutation (Table S2), which is concordant with published data [43]. Compared to the predictive power of SFAB, the RAS mutational status failed to better predict a model's sensitivity to the MEK-inhibitors cobimetinib, trametinib, and selumetinib ( Figure S5). Furthermore, we investigated if the BRAF status of CRC PDOs could predict sensitivity to MEK inhibition. About 10% (6/62) of the CRC PDO carry a pathogenic BRAF mutation, which is concordant with published data (https://www.cbioportal.org/, accessed on 20 November 2021). BRAF status alone failed to better predict sensitivity to the MEKinhibitors cobimetinib, trametinib, and selumetinib ( Figure S5). Apart from one PDO model, all BRAF mutant PDOs have an SFAB signature (Table S2). Therefore, BRAF mutations alone are not better than SFAB ( Figure 5).
Overall, these findings establish that the SFAB signature predicts sensitivity to MEK inhibition in CRC in a RAS/BRAF status-independent manner.   also Table S2). SMAD4 = mothers against decapentaplegic homolog 4, FBXW7 = F-box/WD repeat-containing protein 7, ARID1A = AT-rich interactive domain-containing protein 1A, BMPR2 = Bone morphogenetic protein receptor type II.

De-Novo Acquired Drug Resistance Can Affect Sensitivity to MEK-Inhibitors
In CRC, 5-FU-based therapies, such as FOLFOX and FOLFIRINOX, are the first-line treatment of choice. MEK-inhibitors are currently not approved for first-line treatment of CRC [44], and thus, they could only be considered for subsequent therapy lines. To recapitulate this clinical situation, we wanted to establish 5-FU resistant sibling-and CRISPR-SMAD4 R361H PDOs. PDOs were treated with increasing concentrations of 5-FU up to 230 µM (c max ) [33]. Interestingly, 5-FU resistant lines could only be established from SMAD4 R361H PDOs. SMAD4 wt PDO (R4 wt ) did not recover from long-term 5-FU treatment in two independent approaches (data not shown). This is consistent with the previous observations of SMAD4 functional loss being associated with first-line therapy resistance [13]. 5-FU resistant PDO models and their passage-matched counterparts did not differ in proliferation ( Figure S6A). As expected, drug screening revealed a significantly increased 5-FU resistance ( Figure S6B), as well as an increased resistance to all tested targeted agents, including MEK-inhibitors ( Figure S6C). These observations suggest that de-novo acquired drug resistance to 5-FU might share similar mechanisms with resistance to MEK-inhibitors.

Discussion
The ultimate goal of precision oncology is to identify effective treatments for the individual patient. One route to reaching that goal is linking specific tumor mutations to clinical response to treatment. Combined with chemotherapy, targeted agents are already an important pillar in the standard-of-care treatment of multiple cancers [45]. Among studied examples are the anti-epidermal growth factor receptor (EGFR) antibody cetuximab and the vascular endothelial growth factor (VEGF) antibody bevacizumab [7,46].
The MAPK pathway is dysregulated in more than 30% of human cancers [47]. The aberrant activation of this pathway is usually triggered by gain-of-function mutations in RAS and BRAF proteins [48]. Mutations in KRAS occur in 36-50% of CRC [47] and they interfere with anti-EGFR therapy [49,50]. Only 10-20% of CRC patients are responsive to anti-EGFR treatment, and the KRAS/NRAS mutational status is the standard-of-care biomarker to exclude patients from therapies based on anti-EGFR anti-bodies [51]. Cetuximab is therefore only approved for RAS wild-type tumors. However, 35-45% of these cases still do not respond to this treatment [52,53]. Besides KRAS/NRAS and BRAF mutations, alterations in PI3KCA or PTEN are likely involved in the resistance to anti-EGFR therapy [50,52,54]. Also, mutations in FBXW7 and SMAD4 are prevalent in non-responsive cases [53]. To plan successful therapies for those patients, additional biomarkers are needed. Using a set of biomarkers will help stratify cancer patients into cohorts based on the individual tumor biology [55] rather than only the location of the tumor. We observed that SMAD4 loss-of-function is associated with differential sensitivity to MEK inhibition (Figure 1).
We were able to introduce the R361H point mutation into R4 wt PDOs and generated three CRISPR-PDOs carrying the homozygous SMAD4 R361H mutation (Figure 2A-C, Table S1). However, the three CRISPR-PDOs lost the heterozygous PI3KCA H1047R mutation after clonal selection ( Figure 2C, Table S1), which was present in both R1 R361H and R4 wt . PIK3CA is frequently mutated in CRC with up to 32% [56,57] and is associated with poor prognosis [58,59]. Although the clones lost PI3KCA H1047R , they showed similar behavior to R1 R361H in terms of proliferation time and drug response ( Figure 2D-E). Thus, we conclude that this PI3KCA H1047R mutation is not involved in the MEK-inhibitor response.
Gene editing confirmed that mutations of SMAD4 are responsible for the observed differential drug response. We showed that the introduction of the R361H mutation is sufficient to (re-)induce sensitivity towards the MEK-inhibitors cobimetinib, trametinib, and selumetinib ( Figure 2).
Further, activation of the TGF-β/BMP signaling pathway, specifically the BMP pathway branch, was observed in SMAD4 wt PDOs, indicating that BMP signaling is likely a driver of resistance towards MEK inhibition (Figure 3).
We observed an upregulation of mTOR phosphorylation in trametinib-resistant SMAD4 wt PDO (R4 wt ) and a decrease in phosphorylation for SMAD4 R361H PDOs (Figure 3). Several mechanisms of MEK-inhibitor resistance are associated with the activation of cellular signaling pathways, such as the PI3K/AKT/mTOR pathway [60]. This finding is consistent with a previously published study showing that activation of mTOR promotes tumor growth and metastases [61].
As previously published, BMP signaling, particularly through SMAD5, can promote cancer cell proliferation and tumor growth [62]. In breast cancer, resistance to paclitaxel can be reversed with the depletion of SMAD5 [63]. The role of the BMP signaling in CRC is historically controversial, potentially due to the differential status of this signaling in different CRC molecular subtypes [64,65]. There is evidence of BMP pathway activation in the mesenchymal molecular subtype of the CRC, while in other subtypes, it is suppressed [66]. R1 R361H and R4 wt were derived from a tumor that was described as consensus molecular subtype 2 (CMS2), which is characterized by WNT and MYC activation [15,64]. In addition to CMS classification, the CRC intrinsic subtype B (CRIS-B) is associated with TGF-β-pathway activity, EMT, and poor prognosis [67], which is more applicable to this tumor and correlates with our observations. From these observations, it is plausible to assume that functional loss of SMAD4 and thus loss of BMP signaling renders SMAD4 mutated tumors more sensitive to MEKinhibitors. However, a loss-of-function mutation or deletion of the SMAD4 gene alone did not significantly positively predict response to MEK inhibition ( Figure S4). By broadening the scope to additional frequently mutated genes involved in TGF-β/BMP signaling [11,[38][39][40][41], we identified the novel gene mutational SFAB-signature (SMAD4, FBXW7, ARID1A, or BMPR2) (Figure 4). SMAD4 is considered a signaling mediator of the TGF-β/BMP-pathway, but its role as an important suppressor of CRC progression remains elusive [68]. SMAD4 loss has been shown to promote chemoresistance in multiple cancer types, including CRC [11,12].
ARID1A participates in the TGF-β/BMP-pathway as a tumor suppressor that interacts with p53 regulating cyclin-dependent kinase inhibitor 1A (CDKN1A) and SMAD3 transcription and subsequently tumor growth [71]. ARID1A has emerged as a candidate "driver gene" tumor suppressor based on its frequent mutations in cancer cells, such as ovarian clear cell and endometroid cancers, as well as CRC [72].
BMPR2 is directly involved in BMP branch signal transduction in the TGF-β/BMPpathway [73] and has been reported to play a dual role in regulating tumor growth [74]. In CRC lacking SMAD4, BMPR2 can bind to LIM domain kinase 1 to activate the Rho/Rhoassociated protein kinase (ROCK) pathway to promote tumor invasion and metastases [75].
The SFAB signature predicts the sensitivity towards MEK inhibition in CRC PDOs with a high probability ( Figure 5). PDO models have been consistently shown to faithfully recapitulate features of the tumor of origin in terms of cell differentiation, genomic background, intra-tumor heterogeneity, and drug response [14,15,[76][77][78]. We show that the frequency of SFAB-signature in our PDOs mirrors the expression data of CRC patients ( Figure 4). Therefore, it is plausible to assume that the SFAB signature could reliably predict sensitivity to MEK inhibition in CRC patients. Alterations in the TGF-β/BMP pathway due to loss-of-function mutations or deletions affecting SMAD4, FBXW7, ARID1A, or BMPR2 have previously been shown to correlate with decreased overall survival and to promote chemoresistance in multiple cancer types, such as CRC [11,13], ovarian cancer [79][80][81] and squamous cell carcinoma [82]. This exciting observation suggests that the same alterations that cause resistance to first-line therapies and provide proliferative benefits could at the same time render cancer cells sensitive to certain targeted therapeutics.
In addition to the previously published gene expression signature scores representing MEK-inhibitor sensitivity and MEK-inhibitor adaptive resistance [83,84], SFAB could serve as a predictive biomarker for successful cancer therapy by MEK inhibition.
The present study significantly adds to the concept that multimodal strategies must be considered to deliver effective and specific therapies.

Conclusions
We identified a novel RAS/RAF-independent biomarker signature that could correctly predict the outcome of MEK inhibition in vitro. In conclusion, we propose using the SFAB signature as a RAS/BRAF-independent biomarker for the clinical stratification of patients. Furthermore, we strongly suggest including MEK-inhibitors in the portfolio of therapeutic regimens against CRC.
As the next step to renew the guidelines for CRC treatment, retrospective clinical correlation studies in tumors with known mutation status are ongoing, proofing the clinical benefit of the SFAB signature.

Data Availability Statement:
The data presented in this study are available in Supplementary Tables S1-S4.