FIELD OF THE INVENTION

The present invention relates to an in vitro method for identifying at least one gene product involved in fractional killing of drug sensitive cell populations following cancer drug treatment and a combined preparation comprising a cancer drug and a compound that modulates the expression or the activity of the gene product identified by said method for use in cancer treatment to increase said cancer drug potency and reduce the development of resistance to said cancer treatment.

BACKGROUND OF THE INVENTION

Isogenic cell populations exhibit multiple scales of non-genetic heterogeneity, from gene expression to phenotypic response. Clonal cells in the same phenomenological state, such as cell cycle or differentiation stages, can still respond differently to the same stimulus (Purvis et al, 2012. Science. 336(6087): 1440-44; Spencer et al., 2009. Nature. 459(7245):482-32). A common hypothesis for this cell-to-cell phenotypic heterogeneity is the co-existence, within isogenic cell populations, of different cellular states that arise from stochasticity in the biochemical reactions controlling molecules biosynthesis and degradation (Balazsi et al, 2011. Cell. 144(6):910-25; Elowitz et al. 2002. Science. 297(5584): 1183-6; Loewer et al, 2013. BMC Biol. 11:114; Raj and van Oudenaarden, 2008. Cell. 135(2):216-26).

These processes generate biological noise which can facilitate cell decisions such as growth factor and drug responses, cell proliferation and differentiation, in a non-genetic fashion (Albeck et al, 2013. Mol Cell. 49(2) :249-61; Mitchell et al, 2018. PNAS. 115(12) :E2888-E2897; Purvis et al, 2012. Science. 336(6087): 1440-44; Saint et al, 2019. Nat Microbiol. 4(3):480-491; Santos et al, 2007. Nat Cell Biol. 9(3):324-30; Spencer et al, 2009. Nature. 459(7245):482-32; Siiel et al, 2006. Nature. 440(7083):545-50).

In addition, non-genetic heterogeneity has also been linked to tumor formation -cell evading apoptosis, and incomplete eradication of tumor clones (Marusyk et al, 2012. Nat Rev Cancer. 12(5):323-34; Reyes and Lahav, 2018. Curr Opin Biotechnol.51:109-115). Recent studies have shown that the incomplete response to cancer therapeutics, or fractional killing, is due to differences in cell response dynamics (Paek et ah, 2016. Cell. 165(3):631-42; Roux et al, 2015. Mol Syst Biol. 2015. 11 (5): 803), and play a significant role in the efficacies of many cancer drugs where even for “sensitive cell lines”, only a fraction of responding cells undergoes apoptosis at saturating doses (Fallahi-Sichani et al, 2013. Nat. Chem. Biol. 9(11):708-14; Forcina et al, 2017. Cell Syst. 4(6):600-610.e6). Drug-tolerant persister cells have been shown to mediate this fractional killing, even after repeated rounds of drug challenges, suggesting that the same individual cell can switch between opposing phenotypic responses (Flusberg et al, 2013. Mol. Biol. Cell. 24(14):2186-200; Sharma et al, 2010. Cell. 14 l(l):69-80; Su et al., 2017. Proc Natl Acad Sci U S A. 114(52): 13679-13684). Importantly, this mode of drug-evasion has been shown to lead to genetic drug resistance, ultimately accounting for cancer therapeutic failure (Ramirez et al, 2016. Nat. Commun. 7:10690; Shaffer et al, 2017. Nature. 555(7695):274; Sharma et al., 2010. Cell. 141(l):69-80).

As these studies have illustrated that each clonal tumor cell can process the same signaling pathway differently, understanding the molecular basis of this drug response phenotype switching at the single-cell level, has important implications in the cancer therapeutic development (Salgia and Kulkami, 2018. Trends Cancer. 2018. 4(2): 110-118).

Death receptor targeted agent such as TRAIL, are excellent models for studying the origins of drug response switching, as they have been promising cancer drugs, proven disappointing in clinical trials due to a low maximum drug effect (Emax) evidenced by fractional killing (Gonzalvez and Ashkenazi, 2010. Oncogene. 29(34):4752-65; Holland, 2013. Cytokine Growth Factor Rev. 25(2): 185-93). Cancer therapy research now actively investigates co-treatment strategies that sensitize tumor cells to death receptor agonists (Kretz et al., 2019. Cancers (Basel). 11(4)), and faces the challenge of identifying molecular determinants of drug response heterogeneity, so far elusive because of technical impediments of single-cell approaches.

Recent technological and analytical advances in single-cell approaches, from dynamic imaging of live cells, single-cell RNA sequencing (scRNAseq) and now multi-omics, have allowed the continuing characterization of cell variability between tissues, cell types and across several regulatory layers (Chen et al, 2015. Science. 348(6233):aaa6090; Macaulay et al., 2017. Trends Genet. 33(2): 155-168; Ren et al., 2018. Genome Biol. 19(1):211). Although live-cell imaging has linked the encoding of cellular dynamics into specific cell response phenotypes (Cheong et al, 2011. Science. 334(6054):354-8; Selimkhanov et al, 2014. Science. 346(6215): 1370-3), the molecular origins of the observed variable behaviors remain unclear, essentially because measuring signaling dynamics and global gene expression profiles in the same cell has been technically challenging.

Indeed, cell sample preparation is destructive, which impairs the monitoring of the same cell over time and moreover, the drug treatment itself either deteriorate the integrity of sensitive cells in the case of cytotoxic cancer drugs, or induce a genomic response. As a consequence, the relationship between global - omic signatures and phenotypic measures has stayed mainly correlative.

Thus, it remains a need to develop a method for identifying target genes involved in fractional killing which could represent drug targets for co-treatment with cancer therapy to reduce the development of resistance to said cancer treatment.

SUMMARY OF THE INVENTION

The scope of the invention is defined by the claims. Any subject-matter falling outside the scope of the claims is provided for information purposes only.

To overcome the previously described limitations, the inventors compared the expression profiles of cancer drug predicted responding and non-responding cells of sensitive cell populations as early as detectable changes in fractional killing factor activity can predict cell fate and before cells exhibit the drug effect causing the responding cells to die or to induce further genomic response in the non-responding cells.

A workflow coupling three single-cell technologies is designed: a predictive measure of single-cell response by live-cell microscopy is used in order to isolate predicted responding and predicted non-responding cells at early time by laser-capture microdissection, and each captured cell by scRNAseq is then profiled. This allowed to determine the genome-wide transcriptomic profile at the origin of cellular drug response, and establish a single-cell signature of cancer therapeutic efficacy. Using this approach to compare the predicted responding versus predicted non-responding cells of the same treated population, the inventors identified and experimentally validated a set of gene products that discriminate the two opposing cell responses and represent new potential drug targets for co-treatments. Thus, the present invention relates to an in vitro method for identifying at least one gene product involved in fractional killing of sensitive cell populations following cancer drug treatment said method comprising: i) treating said sensitive cell populations with cancer drug, ii) measuring the activity of fractional killing factor in each single cell of the sensitive cell populations at early time after cancer drug treatment, iii) determining from the fractional killing factor activity whether said cell is predicted to be responding or non responding to said treatment, iv) isolating said single cell, v) determining the expression profile of said single cell, preferably by single-cell RNA sequencing vi) comparing the expression profiles of predicted responding cells to predicted non-responding cells, and vii) determining at least one gene product differentially expressed.

In a particular embodiment, said fractional killing factor is caspase 8 and the activity of said caspase 8 is measured prior one hour after cancer drug treatment. In this case, said cancer drug is preferably death receptor targeted agent, more preferably selected from the group consisting of TRAIL, DR5 agonistic antibody and DR4 agonistic antibody and a small molecule that transcriptionally induces death receptor ligand. In a preferred embodiment, caspase 8 activity is measured by determining the proteolysis rate of at least one caspase-8/10 substrate, preferably at least one IETD peptide sequence. In a more preferred embodiment, said sensitive cell populations are genetically engineered to express caspase 8 activity reporter and proteolysis is determined using luminescence or fluorescence, more preferably using fluorescence electron transfer (FRET)-assay, again more preferably by using live-cell fluorescent microscopy.

In another particular embodiment said fractional killing factor is p53 and the activity of p53 is measured prior ten hours after cancer drug treatment. In this case, cancer drug is preferably chemotherapy drug. In a preferred embodiment, p53 activity is p53 nuclear translocation.

In another aspect, the invention relates to a combined preparation comprising a death receptor targeted agent and a compound that increases the expression of at least one gene selected from the group consisting of: PRSS56, SIVA1, SLC25A1, UBE2D4 and UQCC3, preferably a nucleic acid construct comprising at least one of said genes, for use in cancer treatment to reduce the development of resistance to said cancer treatment in a subject in need thereof. The present invention also relates to a combined preparation comprising a death receptor targeted agent and an inhibitor of Dynamin-l-like protein, preferably Dynasore or Mdivi-1 for use in cancer treatment to reduce the development of resistance to said cancer treatment in a subject in need thereof.

FIGURE LEGENDS

Figure 1. Evidence for drug response phenotype switching in isogenic population, undetected in classic single-cell transcriptomic analyses. (A) Clonal expansion and drug selection reveal the plasticity of drug response phenotypes in isogenic cells. Over 10 clones were derived from the parental cell cultures to ensure the capture of single-cell in different cell states representative of the population. IC50 measures of drug potency, and Emax measures of drug efficacy. Bar graphs represent data from at least three biological repeats of clonal expansion and drug selection. Cell viability was assessed by Celltiter-glo and the absorbance results are in percentage of control (cell population treated with rhTRAIL- vehicle). (B) Natural variations of gene expression was shown by UMAP representation of the joint analysis of the single-cell RNA sequencing data of control resting cells and 1-hr treated samples. Three cell sub-populations were found by clustering analysis. (C) Differential analysis of the 1-hr treated cells vs. control resting cells, for each of the 3 cell sub-population clusters observed in panel B. The x-axis is the average gene expression in a given cluster, and the y-axis is the log2 fold-change of gene expression. No gene reached statistical significance after adjustment for multiple testing.

Figure 2: Phenotype-coupled same-cell profiling (LCM2Seq). (A) Example live-cell microscopy experiment for caspase-8 early activation measurement in each identified cell. Fifty-minute live-cell microscopy experiment was performed to collect caspase-8 dynamics for each cell cultured on a membrane ready for micro-dissection. About 20 fields of cells are monitored, and all cells are tracked for their FRET signal in time. (B) Single cell trajectories are reported for each cell labelled with its field number, so to identify each cell position for the subsequent laser-capture microdissection step of the chosen cells. (C) Each selected cell has been tracked so to position the laser-capture micro-dissection: this allows the same-cell to be phenotyped in live-cell microscopy and genotyped by single cell -RNA-seq.

Figure 3: Caspase-8 early activation rates can be used as a new metric to predict drug response to death receptor ligands. (A) Twenty-four hours live cell microscopy experiments were performed to collect caspase-8 dynamics for each cell and their corresponding cell fate after TRAIL stimulation (Hela Caspase-8 reporter cells, 25 ng/ml TRAIL). (B) Using caspase-8 activation rate at 50 minutes after stimulation the inventors show that they could accurately predict cell response for about 25% of the cell population. Using constant thresholds (lines, measured in several experiments) that allow satisfying accuracy and sufficient cell yield (>95% accuracy for 10% of the cells per 50 mins runs), the inventors isolate chosen single cells (laser capture microdissection) to carry on with single-cell RNA seq (transcriptomics). The inventors normalized each cell FRET curves to its minimal values within the first 5 time points (FRET tends to go down first, known as “photoactivation”).

Figure 4: Single-cell RNA profiling of Hela cells, 50 min post TRAIL treatment. (A)

Principal component analysis of single cell RNA seq dataset. Each dot represents a cell, colored by its predicted phenotype. (B) Heatmap of gene expression correlation coefficients between cells assayed by scRNAseq. The color bars at the top represent potential cofounding factors, namely batches of single cell collections, cell cycle phase prediction, percentage of dropouts and predicted phenotype. (Pearson correlation). (C) Scatterplot representing for each gene measured by scRNA seq the relationship between the mean log-expression and the variance. ERCC spike-ins are represented in grey. (D) Heatmap of gene expression for the genes with the most significant biological variation when compared to ERCC spike-ins. The predicted phenotype of each cell is represented at the top. (E) Cell cycle assessment for each cell in scRNAseq data. Potential cofounding factors such as batch of single cell collection and predicted phenotype are represented on the left side.

Figure 5: Phenotype-coupled single-cell RNA profiling of HeLa cells, 50 minutes’ post TRAIL treatment. TRAIL sensitivity signature of HeLa cells. Phenotyped cells of interest were assayed by RNA seq. Hierarchical clustering of selected cells highlights the heterogeneity of anticancer drug response information contained in identical cells treated together.

Figure 6: Single-cell differential gene expression analysis between cells predicted responding and predicted non-responding to TRAIL treatment. Boxplots highlighting the gene expression differences between cell predicted to be responding and cells predicted to be non-responding for the 135 genes with the most statistically significant differences: FDR<0.05 or non-adjusted p-value < 0.05 and | logFC | > 2 and max dropouts < 50% of cells for both responding and non-responding. Figure 7: Single-cell differential gene expression analysis between predicted responding and predicted non-responding cells to TRAIL treatment. (A) MA-plot. X- axis represents the average gene expression for each gene (log2 scale). Y-axis represents the difference in average gene expression between cells predicted to be responding vs. cells predicted to be non-responding (log2 scale). (B) Volcano-plot. X-axis represents the difference in average gene expression between cells predicted to be responding vs. cells predicted to be non-responding (log2 scale). Y-axis represents the false discovery rate (- loglO FDR). Black dots represent the experimentally validated genes.

Figure 8: Target genes validation using engineered cell lines: single-cell assessment of caspase-8 activation and cell survival upon treatment with TRAIL. Each box represents at least 3 repeats of live-cell microscopy experiments comparing the response of two engineered cell lines to TRAIL treatment (25 ng/ml): control cell expressing ICRP and a mRFPl fluorescent protein only, and a cell line expressing ICRP and a target gene in transcriptional fusion with mRFPl. Boxplots show the maximal caspase-8 activity of each cell for each condition and the survival curves show the cell death times of each condition in one representative experiment. Bar graphs show the average cell variability at the end of the experiments, for at least 3 experimental repeats (p-value < 0.05). The boxes are arranged in two groups (a top group of 5, and the bottom group of two) which indicate two large experiments of cell engineering with their associated control cells. The diagrams with the target gene names indicate the network distance and some intermediary nodes between the target genes and caspase 8.

Figure 9: Target gene DNM1L validation using pharmacological inhibitor: single-cell assessment of caspase-8 activation and cell survival upon treatment with TRAIL.

Each box represents at least 3 repeats of live-cell microscopy experiments comparing the cell response to TRAIL treatment (25ng/ml) alone, to TRAIL treatment in combination with an inhibitor drug in parental cells expressing ICRP only. Boxplots show the maximal caspase-8 activity of each cell for each condition and the survival curves show the cell death times in each condition in one representative experiment. Bar graphs show the average cell viability at the end of the experiments, for at least 2 experimental repeats (p- value < 0.05 for Mdivi-1). The diagrams with the target gene names indicate the network distance and some intermediary nodes between the target genes and caspase-8. DETAILED DESCRIPTION OF THE INVENTION

The inventors surprisingly showed that at early time after drug treatment prior mitochondrial outer membrane permeabilization and drug gene induction, the therapeutic response of a drug sensitive cell population can be predicted by the determination of fractional killing factor activity.

They develop a method for identifying gene products involved in fractional killing by measuring fractional killing factor activity at early time after drug treatment in a single cell of drug sensitive cell populations and comparing expression profiles of predicted responding cells and predicted non-responding cells. The response phenotype of a cell is thus analyzed before the induction of drug transcriptional response that would change expression profile and before the eventual death of the cell.

As used herein gene product refers to either RNA or protein, resulting from expression of a gene.

As used herein, fractional killing refers to the incomplete response of sensitive cell populations following cancer drug treatment. In another terms, fractional killing is the phenomenon that will kill a constant fraction of the cells with the same or similar genotypes at a given drug dose, independent of the absolute numbers of cells following cancer drug treatment applied for a defined period of time. Fractional killing emerges through non-genetic mechanisms. Each cell of a genetically homogeneous tumor cell population retains its ability to respond differently to a therapy, due to the unique state of its transcriptome/proteome at the time of treatment.

Fractional killing contributes to drug resistance in cancer. Indeed, the ability of a cell from a “drug sensitive cell population” to evade drug treatment (“non-responding cells”) is a first step in the emergence of genetically-driven drug resistance (mutational), giving rise to “drug resistant cell populations” (Ramirez et al, 2016. Nat. Commun. 7:10690; Shaffer et al, 2017. Nature. 555(7695):274; Sharma et al., 2010. Cell. 141(l):69-80).

As used herein, the therapeutic response to a drug refers to a positive medical response characterized by objective parameters or criteria such as objective clinical signs of the disease, patient self-reported parameters and/or the increase of survival. The objective criteria for evaluating the response to drug-treatment will vary from one disease to another and can be determined easily by one skilled in the art by using clinical scores. A positive medical response to a drug can be readily verified in appropriate animal models of the disease which are well-known in the art.

Thus, the sensitive cell populations according to the invention can be cells sample from patient presenting a positive medical response to a drug.

As used herein, the term "patient cell sample" refers to cells isolated from a patient.

The term "patient" refers to a mammal including a non-primate (e.g. a cow, pig, horse, cat, dog, rat and mouse) and a primate (e.g. a monkey and a human), and more preferably a human. The patient to treat, who is preferably a human patient, is affected with a cancer.

Said patient-derived tumor cells can be from haematologic cancer, in particular acute myelogenous leukaemia (AML), chronic lymphocytic leukaemia (CLL), multiple myeloma, Hodgkin's disease, non-Hodgkin's lymphoma, B cell, cutaneous T cell lymphoma, or a non-haematologic cancer, for instance brain, epidermoid (in particular lung, breast, ovarian), head and neck (squamous cell), bladder, gastric, pancreatic, head, neck, renal, colon, prostate, cervical, testicular, liver, colorectal, oesophageal or thyroid cancer, and melanoma.

The resistance of a cell to a therapeutic agent may also be determined by the survival of the cells after some time of administration of the treatment. The sensitivity of a cell to a therapeutic agent may be determined by the death of the cells after some time of administration of the treatment.

The drug sensitive cell populations according to the invention can also be a human cancer cell line which refers to an ex vivo cancerous cell culture which a fraction is killed after cancer therapy. Sensitive cell populations comprise responding cells and non-responding cells to the cancer therapy, resulting in fractional killing at a given dose.

Cell lines that may be used as test cells include human cancer cell lines. Examples of human cancer cell lines which may be used according to the invention include: Lung Cancer Human Cell Lines (Non-small cell lung cancer adenocarcinoma cell line, A549; adenosquamous cell carcinoma, NCI-H125; squamous cell carcinoma, SK-MES-1, bronchial-alveolar carcinoma, NCI-M322; large cell Carcinoma, A 427, mucoepidermoid carcinoma, NCI-M292, small cell lung cancer (SCLC) "Classic", NCI-M69; SCLC "Variant", NCI-M82; SCLC "Adherent", SHP77; colon cancer human cell lines (COLO 205, DLD-1, HCT- 15, HT29, LoVo); breast cancer human cell lines, (MCF7 WT, MCF7 ADR, MDA-MB-231, HS 578T); prostate cancer human cell lines (D4 145, LNCaP, PC-3, UMSCP-1); melanoma human cell lines (RPMI-7951, LOX, SK-MEL 2, SK-MEL-5, A 375); renal cancer human cell lines (A 498, A 704, Caki-1, SNI2 C, UO-31); ovarian cancer human cell lines (IGROV-1, OVCAR-3, SK-OV-3, A2780, OVCAR-4, OVCAR-5, OVCAR-8); leukemia human cell lines (Molt-4, RPMI 8336, P388, P388/ ADR-Resist CCRF-CEM, CCRF-SB); central nervous system cancer human cell lines (SF 126, SF 295, SNB19, SNB 44, SNB 56, TE 671, 4251); sarcoma human cell lines (A-204, A 673, MS 913T, Ht 1080, Te 85); head and neck squamous cancer human cell lines (UM-SCC- MB,C, UM-SCC- 21A, UM-SCC-22B); normal fibroblasts (MRC-5-lung, human, CCD- 194Lu- lung, human, IMR-90-lung, human, NIH 3 T3 -mouse). In a preferred embodiment, cells are HeLa cells derived from cervical cancer.

In the first step of the method disclosed herein, the sensitive cell populations as described above are treated with a cancer drug agent. By “treating sensitive cell populations with a cancer drug agent” as used herein is meant contacting sensitive cell populations with said cancer drug. This method is carried out in vitro , i.e. outside of the body of the patient. In particular embodiment, sensitive cells disclosed herein are cultured in vitro and cancer drug is added to the cell culture.

In a preferred embodiment, the sensitive cell populations are cultured on an adherent culture system. The culture system may be in any form suited to the method according to the invention, in particular in the form of a flask, a multi-well plate or a dish. According to a preferred embodiment, the adherent culture system is an adherent monolayer culture system. This system comprises a solid support, for example glass or plastic, usually coated with a matrix or a substrate promoting cell adherence. The substrate may be a protein substrate consisting of attachment factors and promoting the adhesion of cells to the support. These attachment factors may in particular be selected from poly-L-lysine, collagen, fibronectin, laminin or gelatin, preferably collagen.

The cells are cultured in a basal medium. Numerous basal media are available commercially and are well-known to the person skilled in the art. This medium may be a minimum medium particularly comprising mineral salts, amino acids, vitamins and a carbon source essential to cells and a buffer system for regulating pH. The basal medium able to be used in the method according to the invention includes, for example, but are not limited to, DMEM/F 12 medium, DMEM medium, RPMI medium, Ham's F12 medium, IMDM medium and KnockOut™ DMEM medium (Life Technologies). Depending on the medium used, it may be necessary or desirable to add serum, glutamine, vitamin C, one or more antibiotics such as streptomycin, penicillin and/or anti-mycotic such as Fungizone (amphotericin B).

In a particular embodiment, the cancer drug agent is added to the medium of cell culture.

As used herein, “drug”, “therapeutic agent” or “cancer drug” refers to a compound or agent that provides a desired biological or pharmacological effect when administered to a human or animal, particularly results in an intended therapeutic effect or response on the body to treat or prevent conditions or diseases. Therapeutic agents include any suitable biologically-active chemical compounds, biologically derived components such as for example small molecules, cells, proteins, peptides, antibodies, enzymes, polynucleotides, and radiochemical therapeutic agents, such as radioisotopes.

In a particular embodiment, said cancer drug agent is a death receptor targeted agent such as death receptor agonist and small molecule that transcriptionally induces death receptor ligand. In another particular embodiment, said cancer drug target is chemotherapy.

By “death receptor” is meant a receptor that induces cellular apoptosis once bound by a ligand. Death receptors include, for example, tumor necrosis factor (TNF) receptor superfamily members having death domains (e.g., TNFRI, Fas, DR3, 4, 5, 6) and TNF receptor superfamily members without death domains LTbetaR, CD40, CD27, HVEM.

By “death receptor-targeted agent” is meant a substance that is able to increase the activation of death-receptor that initiates caspase activation leading to apoptosis.

In particular, death receptor-targeted agent is a death receptor agonist or a small molecule that transcriptionally induces death receptor ligand, such as TRAIL.

By “agonist” is meant a substance (molecule, drug, protein, etc.) that is capable of combining with a receptor (e.g. death receptor) on a cell and initiating the same reaction or activity typically produced by the binding of the endogenous ligand (e.g., apoptosis).

The death receptor agonist as described herein can be TNF, Fas Ligand, or TRAIL. The agonist can further be a fragment of these ligands comprising the death receptor binding domain such that the fragment is capable of binding and activating the death receptor. The agonist can further be a fusion protein comprising the death receptor binding domain such that the fusion protein is capable of binding and activating the death receptor. In one embodiment, said TRAIL is a recombinant human protein, such as dulanermin also known as AMG951.

The agonist can further be an apoptosis-inducing antibody that binds the death receptor. The “antibody” can be monoclonal, polyclonal, chimeric, single chain, humanized, fully human antibody, or any Fab or F(ab')2 fragments thereof, crosslinked or not using crosslinking agents such as FcyR or antihuman Fc-specific reagents. Thus, the agonist of the present method can be an antibody specific for a Fas, TNFR1 or TRAIL death receptors, such that the antibody activates the death receptors. The agonist can be an antibody specific for DR4 or DR5. In one embodiment, the DR5 agonist is an antibody selected from the group consisting of Lexatumumab (also known as ETR2-ST01), Tigatuzumab (also known as CS-1008), Conatumumab (also known as AMG 655), Drozitumab, zaptuzumab (also known as AD5.10), HGSTR2J/KMTRS, DS-8273a, LBY- 135 and TAS266. In another embodiment, the DR4 agonist is mapatumumab (also known as HGS-ETR1 or TRM1).

By a “small molecule that transcriptionally induces death receptor ligand” is meant a substance that induces an up-regulation of the endogenous death receptor ligand such as TRAIL in tumor and normal cells leading to apoptosis activation. For example, said small molecule that transcriptionally induces TRAIL is ONC-201 (TIC- 10).

Cancer drug agent can also be a chemotherapy drug which includes cytotoxic anti neoplastic agents, such as alkylating agents, platinum-based agents, nucleotide analogs and precursor analogs, kinase inhibitors, anti-metabolites, anti-microtubule agents, Topoisomerase inhibitors, cytotoxic antibiotics and others. As non-limiting examples, said drug for chemotherapy is selected from the group consisting of: cisplatin (also with bevacizumab or pemetrexed), carboplatin, 5-FU Fluorouracil, doxorubicin, actinomycin, bortezomib, docetaxel, Crizotinib, Gefitinib, Erlotinib, Afatinib.

In a preferred embodiment, said cancer drug agent is added to the medium of the cell culture at a concentration comprised between 1 to 1000 ng/mL, preferably between 10 and 100 ng/mL, more preferably 25 ng/mL. The therapeutic response of each single cell of said sensitive cell populations is then predicted by the determination of a fractional killing factor activity at early time after drug treatment. By “after the drug treatment” is meant after contacting said sensitive cell populations with cancer drug. In a preferred embodiment, said cancer drug is contacting to sensitive cell populations by adding said cancer drug to sensitive cell populations culture.

The term “predicting therapeutic response to a treatment with a drug”, as used herein, refers to an ability to assess whether the treatment with a drug will induce or not death of the cell after some time of administration of the treatment. In another terms, it refers to an ability to determine whether a cell is likely to be responding or non-responding to a drug after some time of administration of the treatment.

The term “responding cell”, as used herein, refers to a cell, within a drug- sensitive cell population as described above, that is going to commit to cell death following cancer drug treatment. The term “non-responding cell”, as used herein, refers to a cell, within the same drug-sensitive cell population, that is going to survive cancer drug treatment.

According to the invention, such an ability to assess whether the treatment will induce or not death of the cell is typically exercised at early time after the drug treatment.

The measure of the activity at early time allows to predict accurately the response phenotype of a cell after a time when the contact of the sensitive cell populations with said cancer drug induced only little transcriptional response that would change its profile and allows to recover both non-responding and responding cells that eventually die when its response phenotype manifests itself.

By “early time” used herein, it is intended time before which the cells exhibit the drug effect causing the responding cells to die and as early as detectable changes in fractional killing factor activity can predict cell fate with accuracy for at least a fraction of the cell population, preferably at least 1% of the cell population, more preferably at least 5, 10, 12, 15, 17, 20, 25% of the cell population.

Early time can be determined for each fractional killing factor activity by determining the time after contacting said sensitive cell populations with the cancer drug at which the fractional killing factor activity allows to separate cells that will respond from the ones that will not, for at least a fraction of the cell population, preferably at least 1% of the cell population, more preferably at least 5, 10, 12, 15, 17, 20, 25% of the cell population. Thus, “early time” is determined by measuring after drug treatment, the fractional killing factor activity over time in single responding cell. The fractional killing factor activity measured over time in each single cell is associated to the actual response phenotype at the end of the experiment, that is to determine the time at which the therapeutic response can be predicted accurately for at least a fraction of the cell population, more preferably at least 1, 5, 10, 12, 15, 17, 20, 25% of the cell population.

Fractional killing factors are compounds involved in the incomplete response of sensitive cells to a drug. In particular fractional killing factors are caspase 8 or p53.

In the particular embodiment wherein said fractional killing factor is caspase 8, caspase 8 activity is measured, preferably over time, prior one hour after drug cancer treatment, preferably prior 55, 45, 40, 35 or 30 minutes after drug cancer treatment. Said drug cancer treatment is preferably a death receptor targeted agent treatment as described previously.

As non-limiting examples, Caspase 8 activity can be determined by measuring the proteolysis of its substrates. Caspase 8 activity can be determined using genetically encoded substrate or chemical compounds added to cells prior to drug treatments. In particular said caspase 8 activity can be determined by measuring the cleavage of Caspase- 8/10 substrate, preferably IETD sequence that is efficiently cleaved by caspase-8. Caspase- 8/10 substrate cleavage can be measured by fluorescence or bioluminescence. Preferably the cleavage of IETD sequence is measured with fluorescence energy transfer (FRET) technology, more preferably by using caspase-8 activity reporter (ICRP), a FRET donor- acceptor pair connected via a linker that contains IETD sequence as described in Albeck et al. 2008. Mol. Cell. 30(1): 11-25.

Said donor acceptor pair are donor-acceptor pair of fluorescent proteins (FP), preferably selected from the group consisting of Cyan fluorescent protein (CFP)- yellow fluorescent protein (EYFP), EGFP-Red fluorescent protein (RFP) and Far-red fluorescent protein (FFP)- Infrared fluorescent protein (IFR), preferably Cyan fluorescent protein (CFP)- yellow fluorescent protein (EYFP). As non-limiting examples said donor-acceptor pair can be selected from the group consisting of ECFP-EYFP, mTurquoise2-sEYFP, mTurquoise2-mVenus, EGFP-mCherry, Clover-mRuby2, mClover3-mRuby3, mNeonGreen-mRuby3, eqFP650-iRFP, mAmetrine-tdTomato, LSSmOrange-mKate2, EGFP-sREACh, EGFP-ShadowG, EGFP-activated PA-GFP, EGFP-Phanta, mTagBFP- sfGFP, mVenus-mKoK, CyOFPl-mcardinal, preferably ECFP and EYFP. In a particular embodiment, said cells are genetically engineered to express the caspase 8- activity reporter (ICRP). Methods by which said cells can be genetically modified to express ICRP are well known in the art. A nucleic acid molecule encoding the ICRP may be introduced into the cell in the form of e.g. a vector, or any other suitable nucleic acid construct. Vectors, and their required components, are well known in the art. Nucleic acid molecules encoding ICRP can be generated using any method known in the art, e.g. molecular cloning using PCR. In particular, said cells are engineered by introducing a viral vector, preferably a retroviral vector encoding said ICRP. In a particular embodiment, nucleic acid construct or vector introduced into a cell may be expressed episomally, or may be integrated into the genome of the cell.

At early time after cancer drug treatment of genetically engineered cells as described above, fluorescence intensity is measured over time. The measure of the fluorescence intensity can be realized by any fluorescent methods well known in the art, preferably by fluorescent microscopy. In a preferred embodiment, the fluorescence intensity of the cell is measured in a live-cell fluorescent microscopy. After image acquisition, the fluorescence intensity is measured in each cell by using software well known in the art, for example ImageJ or image analysis algorithms developed in Python or MATLAB. In particular, Caspase 8-mediated proteolysis of ICRP was monitored by acquiring at each time point, for instance every 5 minutes or less, images for each fluorescent channel (donor and acceptor) and calculating the FRET ratio, that is the fluorescence intensities ratio or image ratio of donor: acceptor or donor :FRET, after suitable correction for background. The average FRET ratio trajectory (over time) of untreated cells was subtracted from the FRET ratio of each treated cell trajectory.

In a preferred embodiment, FRET ratio is measured over time, preferably prior one hour after cancer drug treatment and the time derivative of the FRET ratio is computed, in particular as described in Roux et al, 2015. Mol Syst Biol. 2015. 11(5):803. The time derivative of FRET ratio (dFR/dt) is computed to distinguish between cells predicted responding and cells predicted non-responding to cancer drug treatment, preferably death receptor agonist.

In a preferred embodiment, by recording the value of dFR/r/z for each cell prior one hour after drug cancer treatment, preferably before 55, 50, 45, 40, 35, 30 minutes, more preferably 50 minutes after drug cancer treatment, the cells are predicted responding to cancer drug treatment for dFR/dt > 75 ^th percentile of the population, preferably > 80 ^th , 85 ^th , 90 ^th , 95 ^th percentiles of the population, and are predicted non-responding to cancer drug treatment for dFR/dt < 25 ^th percentile, preferably < 20 ^th , 15 ^th , 10 ^th , 5 ^th percentiles of the population.

In the another particular embodiment wherein said fractional killing factor is p53, p53 activity is measured, preferably over time, prior ten hours after drug cancer treatment, preferably prior 10, 9, 8, 7, 6, 5, 4, 3, 2 or 1 hour, more preferably prior 6, 5, 4, 3, 2 and 1 hour after drug treatment, again more preferably between 1 and 6 hours or between 6 to lOh after drug cancer treatment. Said drug cancer treatment is preferably chemotherapy as described previously.

As non-limiting example, p53 activity can be determined by measuring the nuclear translocation of p53. The nuclear translocation of p53 can be determined by any method well-known in the art, for example by detecting the subcellular localization of luminescent or fluorescently-tagged p53 using fluorescent microscopy or image-based flow cytometry. In particular, the p53 nuclear translocation can be determined by measuring the ratio of the p53 signal in the nuclei and the cytoplasm of tagged p53 or by measuring the normalized average nuclear quantity of p53.

In a particular embodiment, said responding cells are genetically engineered to express the p53 fused to a fluorescent or luminescent protein, also named herein p53 fusion protein. Methods by which said cells can be genetically modified to express p53 fusion protein are well known in the art. A nucleic acid molecule encoding the p53 fusion protein may be introduced into the cell in the form of e.g. a vector, or any other suitable nucleic acid construct. Vectors, and their required components, are well known in the art. Nucleic acid molecules encoding p53 fusion protein can be generated using any method known in the art, e.g. molecular cloning using PCR. In particular, said cells are engineered by introducing a viral vector, preferably a retroviral vector encoding said p53 fusion protein. In a particular embodiment, nucleic acid construct or vector introduced into a cell may be expressed episomally, or may be integrated into the genome of the cell.

In a preferred embodiment when fluorescently tagged p53 is used, at early time after cancer drug treatment of genetically engineered cells as described above, fluorescence intensity is measured over time. The measure of the fluorescence intensity can be realized by any fluorescent methods well known in the art such as fluorescent microscopy, flow cytometry, fluorescent spectroscopy, preferably by fluorescent microscopy. In a preferred embodiment, the fluorescence intensity of the cell is measured in a live-cell fluorescent microscopy. After image acquisition, the fluorescence intensity is measured by using software well known in the art, for example ImageJ or image analysis algorithms developed in Python or MATLAB.

In a preferred embodiment, fluorescence intensity of p53 tagged protein is measured over time, preferably prior ten, nine, eight, seven, six, five, four, three, two hours, or one hour after cancer drug treatment, more preferably prior 6, 5, 4, 3, 2 and 1 hour after drug treatment, again more preferably between 1 and 6 hours or between 6 to 10 hours after cancer drug treatment and the normalized average nuclear intensity is determined to distinguish between cells predicted responding and non-responding to chemotherapy drug, preferably cisplatin. In a preferred embodiment, by recording the normalized average nuclear intensity for each cell prior ten hours after drug cancer treatment, preferably prior 10, 9, 8, 7, 6, 5, 4, 3, 2 or 1 hour, more preferably prior 6, 5, 4, 3, 2 and 1 hour after drug treatment, again more preferably between 1 and 6 hours or between 6 to 10 hours, the cells are predicted responding for normalized average nuclear intensity > 75 ^th percentile of the population, preferably > 80 ^th , 85 ^th , 90 ^th , 95 ^th percentiles of the population and are predicted non-responding for normalized average nuclear intensity < 25 ^th percentile of the population, preferably < 20 ^th , 15 ^th , 10 ^th , 5 ^th percentiles of the population.

In another preferred embodiment, fluorescence intensity of p53 tagged protein is measured over time, preferably prior ten, nine, eight, seven, six, five, four, three, two hours, or one hour after cancer drug treatment, more preferably prior 6, 5, 4, 3, 2 and 1 hour after drug treatment, again more preferably between 1 and 6 hours or between 6 to 10 hours after cancer drug treatment and cytoplasmic-nuclear ratio intensity of p53 is determined to distinguish between cells predicted responding and non-responding to chemotherapy drug, preferably cisplatin. The subcellular localisation of p53 may be determined by using subcellular markers well-known in the art for the nucleus, such as Hoechst, SiR-DNA or fluorescently tagged nuclear protein such as H2B. In a preferred embodiment, by recording cytoplasmic-nuclear ratio intensity of p53 for each cell prior ten hours after drug cancer treatment, preferably prior 10, 9, 8, 7, 6, 5, 4, 3, 2 or 1 hour, more preferably prior 6, 5, 4, 3, 2 and 1 hour after drug treatment, again more preferably between 1 and 6 hours or between 6 to 10 hours, the cells are predicted responding for cytoplasmic-nuclear ratio intensity < 25 ^th percentile of the population, preferably < 20 ^th , 15 ^th , 10 ^th , 5 ^th percentiles of the population and are predicted non-responding for cytoplasmic-nuclear ratio intensity > 75 ^th percentile of the population, preferably > 80 ^th , 85 ^th , 90 ^th , 95 ^th percentiles of the population.

After predicting whether cell will respond or not to drug cancer treatment, single cells are isolated to determine the expression profile of each single cell. Each single cell can be isolated by any methods well known in the art, such as sorted by flow cytometry, preferably, cells are isolated by microdissection, preferably laser-capture microdissection.

The terms “expression cell profile”, “expression profile”, “gene expression cell profile” or “gene expression profile” as used herein can be used interchangeably and refer to transcriptome as well as proteome or metabolome. The expression profile is then determined in each single cell. In a particular embodiment, the transcriptome profile of each single cell is determined by any methods well-known in the art. In particular, the mRNA content of each single cell is first extracted according to standard methods, for example using lytic enzymes or chemical solutions or extracted by nucleic-acid-binding resins following the manufacturer's instructions. In a preferred embodiment, the mRNA expression level is then measured by single-cell RNA sequencing (RNAseq) method to analyse the cellular transcriptome. Single-cell RNAseq can be performed for example in plate, micro or nano-wells, droplet-based microfluidics, microfluidics, tubes. In RNAseq method, the extracted mRNA is retro-transcribed and cDNA is amplified with barcode primer to obtain a barcoded nucleic acid.

As used herein, the term “barcode nucleic acid” refers to a unique nucleotide sequence that is distinguishable from any other barcode as well as from any other nucleotide sequences within the nucleic acid sequence, wherein it is comprised. Production of these sets of barcodes to identify RNA molecules (Unique Molecular Identifier, UMI) or barcodes to identify each sequenced single cell (cell barcode) in case of multiplexing, are known by persons skilled in the art. Because each cell is exposed to a barcode sequence, the nucleic acids arising from these cells may be subsequently distinguished. Said barcoded nucleic acid is then sequenced. The approach aims at identifying genes product differentially expressed between predicted responding and non-responding cells.

In another embodiment, the proteome or metabolome profile can be determined by any methods well-known in the art. Based on the comparison of expression profiles of predicted responding and non responding cells, genes products differentially expressed in both populations are identified.

As used herein, a gene product differentially expressed in cells predicted to be responding and non-responding refers to an expression level, such mRNA or protein expression level in predicted responding cells which is, after normalization, at least 1.5-fold higher, or 2, 3, 4, 5-fold higher or lower, than the expression level in predicted non-responding cells. Expression levels may be normalized by using any commonly used suitable methods such as for example the “Trimmed Mean of M-values” normalization (TMM) method from edgeR package (Robinson et al. 2010. Genome Biol. 11, R25). After normalization, gene product expression profiles can be compared using standard statistical tests for mean comparison such as the Wilcoxon rank sum test, or by using dedicated statistical frameworks such as MAST (Finak et al. 2015. Genome Biol. 16, 278), DESeq2 (Love et al. 2014. Genome Biol. 15, 550-21) and edgeR (McCarthy et al. 2012. Nucleic Acids Res. 40, 4288-4297). Using this method, a set of gene products that discriminates the two opposing cell responses was identified and represents new potential drug targets for co-treatments to increase said cancer drug potency and reduce the development of said cancer drug resistance.

Thus, in another aspect, the present disclosure also relates to the therapeutic use of a combined preparation comprising a cancer drug agent and a compound that modulates the expression of gene product identified by the method as described herein.

In a preferred embodiment, said combined preparation comprised a cancer drug agent, preferably death receptor targeted agent as described previously, and a compound that increases the expression of at least one gene selected from the group consisting of PRSS56, SIVA1, SLC25A1, UBE2D4 and UQCC3.

Gene expression can be increased by compounds that include, but are not limited to, chemicals, antibiotic, compounds known to modify gene expression, modified or unmodified polynucleotides (including oligonucleotides), polypeptides, peptides, small RNA molecules. In a particular embodiment, gene expression is increased by administering to subject a nucleic acid construct or vector comprising a transgene selected from the group consisting of: PRSS56, SIVA1, SLC25A1, UBE2D4 and UQCC3.

The term “nucleic acid construct” as used herein refers to a man-made nucleic acid molecule resulting from the use of recombinant DNA technology. A nucleic acid construct is a nucleic acid molecule, either single- or double-stranded, which has been modified to contain segments of nucleic acids sequences, which are combined and juxtaposed in a manner, which would not otherwise exist in nature. A nucleic acid construct usually is a “vector”, i.e. a nucleic acid molecule which is used to deliver exogenously created DNA into a host cell.

Examples of appropriate vectors include, but are not limited to, recombinant integrating or non- integrating viral vectors and vectors derived from recombinant bacteriophage DNA, plasmid DNA or cosmid DNA. Preferably, the vector is a recombinant integrating or non integrating viral vector. Examples of recombinant viral vectors include, but not limited to, retrovirus, adenovirus, parvovirus (e. g. adenoassociated viruses), coronavirus, negative strand RNA viruses such as orthomyxovirus (e. g., influenza virus), rhabdovirus (e. g., rabies and vesicular stomatitis virus), paramyxovirus (e. g. measles and Sendai), positive strand RNA viruses such as picornavirus and alphavirus, and double-stranded DNA viruses including adenovirus, herpesvirus (e. g., Herpes Simplex virus types 1 and 2, Epstein-Barr virus, cytomegalovirus), and poxvirus (e. g., vaccinia, fowlpox and canarypox). Other viruses include Norwalk virus, togavirus, flavivirus, reoviruses, papovavirus, hepadnavirus, and hepatitis virus, for example.

As used herein, the term “increase”, “overexpress” or “overexpression” refers to an expression level which is, after normalization, at least 1.5-fold higher, or 2, 3, 4, 5-fold higher, than the expression level in untreated cell. Expression levels may be normalized by using expression levels of mRNA which are known to have stable expression such as ribosomal 18S, GAPDH (glyceraldehyde 3-phosphate dehydrogenase) or b-actin.

The increase of said genes may be determined by measuring the expression level of genes. The gene expression level may be determined by any suitable methods known by skilled persons. Usually, these methods comprise measuring the quantity of mRNA. Methods for determining the quantity of mRNA are well known in the art. For example the nucleic acid contained in the sample is first extracted according to standard methods, for example using lytic enzymes or chemical solutions or extracted by nucleic-acid-binding resins following the manufacturer's instructions. The extracted mRNA is then detected by hybridization (e.g., Northern blot analysis) and/or amplification (e.g., RT-PCR). Quantitative or semi- quantitative RT-PCR is preferred.

The level of the target genes protein may also be determined by any suitable methods known by skilled persons. Usually, these methods comprise contacting a cell sample, preferably a cell lysate, with a binding partner capable of selectively interacting with the target gene protein present in the sample. The binding partner is generally a polyclonal or monoclonal antibodies, preferably monoclonal. The quantity of the protein may be measured, for example, by semi-quantitative Western blots, enzyme-labelled and mediated immunoassays, such as ELISAs, biotin/avidin type assays, radioimmunoassay, Immunoelectrophoresis or immunoprecipitation or by protein or antibody arrays.

In another preferred embodiment, said combined preparation comprised a cancer drug agent, preferably death receptor targeted agent as described previously, and an inhibitor of Dynamin-l-like protein, preferably dynasore (CAS No. 1202867-00-2) or Mdivi-1 (CAS No. 338967-87-6).

The present invention also relates to a method of treating a cancer in a subject in need thereof comprising administering to said subject a therapeutically effective amount of a combined preparation as described above.

In a particular embodiment, the disclosure relates to a combined preparation as disclosed above for the use in cancer treatment to reduce the development of resistance to said cancer treatment in a subject in need thereof.

In the context of the invention, the term "treating" or "treatment", as used herein, means reversing, alleviating, inhibiting the progress of, or preventing the disorder or condition to which such term applies, or reversing, alleviating, inhibiting the progress of, or preventing one or more symptoms of the disorder or condition to which such term applies.

"Treating cancer" includes, without limitation, reducing the number of cancer cells or the size of a tumor in the patient, reducing progression of a cancer to a more aggressive form (i.e. maintaining the cancer in a form that is susceptible to a therapeutic agent), reducing proliferation of cancer cells or reducing the speed of tumor growth, killing of cancer cells, reducing metastasis of cancer cells or reducing the likelihood of recurrence of a cancer in a subject. Treating a subject as used herein refers to any type of treatment that imparts a benefit to a subject afflicted with cancer or at risk of developing cancer or facing a cancer recurrence. Treatment includes improvement in the condition of the subject (e.g., in one or more symptoms), delay in the progression of the disease, delay in the onset of symptoms or slowing the progression of symptoms, etc. Preferably, the combined preparation as disclosed herein is used to reduce the development of resistance to said cancer.

The cancer is any cancer type as disclosed above. In some embodiments said cancer is selected from the group consisting of: haematologic cancer, in particular acute myelogenous leukaemia (AML), chronic lymphocytic leukaemia (CLL), multiple myeloma, Hodgkin's disease, non-Hodgkin's lymphoma, B cell, cutaneous T cell lymphoma, or a non-haematologic cancer, for instance brain, epidermoid (in particular lung, breast, ovarian), head and neck (squamous cell), bladder, gastric, pancreatic, head, neck, renal, colon, prostate, cervical, testicular, liver, colorectal, oesophageal or thyroid cancer..

The terms "subject" and "patient" are used interchangeably herein and refer to a mammal including a non-primate (e.g. a cow, pig, horse, cat, dog, rat and mouse) and a primate (e.g. a monkey and a human), and more preferably a human. The patient to treat, who is preferably a human patient, is affected with a cancer.

In some preferred embodiments, said combined preparation is administered to a cancer patient previously classified as resistant to treatment with said cancer drug.

As used herein, a "therapeutically effective amount" or an "effective amount" means the amount of a composition that, when administered to a subject for treating a state, disorder or condition is sufficient to achieve the desired therapeutic result. The therapeutically effective amount will vary depending on the compound, formulation or composition, the disease and its severity and the age, weight, physical condition and responsiveness of the subject to be treated.

The compound that modulates the identified gene product as described previously and the cancer drug may be used simultaneously, separately or sequentially. The compound that modulates the identified genes as described previously may be administered before, after, or concurrently with the therapeutic drug. The compounds or cancer drugs described herein may be administered by any means known to those skilled in the art, including, without limitation, intravenously, orally, intra- tumoral, intra-lesional, intradermal, topical, intraperitoneal, intramuscular, parenteral, subcutaneous and topical administration. Thus the compositions may be formulated as an injectable, topical, ingestible, or suppository formulation. Administration of the compounds or therapeutic agents to a subject in accordance with the present invention may exhibit beneficial effects in a dose-dependent manner. Thus, within broad limits, administration of larger quantities of the compositions is expected to achieve increased beneficial biological effects than administration of a smaller amount. Moreover, efficacy is also contemplated at dosages below the level at which toxicity is seen.

It will be appreciated that the specific dosage of compounds or cancer drugs administered in any given case will be adjusted in accordance with the composition or compositions being administered, the volume of the composition that can be effectively delivered to the site of administration, the disease to be treated or inhibited, the condition of the subject, and other relevant medical factors that may modify the activity of the compositions or the response of the subject, as is well known by those skilled in the art.

For example, the specific dose of compounds or cancer drugs for a particular subject depends on age, body weight, general state of health, diet, the timing and mode of administration, the rate of excretion, medicaments used in combination and the severity of the particular disorder to which the therapy is applied. Dosages for a given patient can be determined using conventional considerations, e.g., by customary comparison of the differential activities of the compositions described herein and of a known agent, such as by means of an appropriate conventional pharmacological protocol. The compositions can be given in a single dose schedule, or in a multiple dose schedule.

Suitable dosage ranges for a compound that modulates the identified gene by the method described previously may be of the order of several hundred micrograms of the agent with a range from about 0.001 to 10 mg/kg/day, preferably in the range from about 0.01 to 1 mg/kg/day.

EXAMPLES

1. Materials and methods

Cell culture and materials Caspase-8 activity reporters (ICRP) were constructed as previously described (Albeck et al. 2008. Mol. Cell. 30(1): 11-25). mRFPl version ofBLOCISl, DNM1L, MITF, PRSS56, SIVA1, SLC25A1, UBE2D4, UQCC3 (Cl lorf83) were constructed by PCR, from ORF Clones in pcDNA3.1+/C-(K)-DYK vectors (GenScript Biotech B.V., Netherlands) or from pEGFP-Nl-MITF-A (a gift from Shawn Ferguson, Addgene plasmid # 38132) and mCh- Drpl (a gift from Gia Voeltz (Addgene plasmid # 49152), all with mRFPl in position 2 after the IRES (transcriptional fusion) in pMSCV or pQCXIX (Clontech Takara Bio, France).

HeLa cells were obtained from the ATCC and cultured in DMEM Glutamax supplemented with 10% fetal bovine serum and penicillin / streptomycin (Thermo Fisher Scientific, France). HeLa cells stably expressing ICRP were derived by infection with retrovirus (Retro-X system, Clontech Takara Bio, France) and positive selection by FACS, single cell sort, and grown in 96-well plate until 1 colony appeared. Freshly cloned cells were used in all experiments (below passage 10). BLOC1S1, DNM1L, LLPH, MITF, PRSS56, SIVA1, SLC25A1, UBE2D4, UQCC3 (Cl lorf83) with mRFPl cells (transcriptional fusion) were obtained by infection of the respective mRFPl fusion gene into the parental clone of HeLa ICRP cells. After infection, positive cells were enriched by FACS sorting but not sub cloned to avoid the selection of a particular clone and to allow the emergence of a population of cells expressing various amount of each mRFPl tagged protein. Recombinant human (rh) TRAIL was obtained from R&D Systems Europe (Lille, France). Dynasore (ALX-270-502-M005) and Mdivi-1 (BML-CM 127-0010) were obtained from ENZO Life Science (Villeurbanne, France).

Live-cell microscopy and image analysis

Clonal HeLa cells stably expressing the FRET-based initiator caspase reporter (ICRP) were seeded into FrameSlides PET (Carl Zeiss, France) or 96-well plates, coated with rat- tail collagen I (Thermo Fisher Scientific, France). Cells in phenol red-free DMEM Glutamax supplemented with 10% fetal bovine serum and penicillin / streptomycin (Thermo Fisher Scientific, France), were imaged every 3 min for up to 24hrs in the temperature/C02-controlled environmental chamber of a Delta Vision Elite microscope (GE Healthcare Life Sciences, Velizy-Villacoublay, France), with a 20x objective (NA = 0.75) in transmitted light and using the following solid state illumination excitation wavelengths and single band pass emission filters : for CFP (Ex. 438/24 nm / Em. 475-24 nm), YFP (Ex. 513/17 nm / Em. 548/22 nm). In addition to the time-lapse runs, cells stably expressing target proteins tagged with mRFPl or mCherry, were imaged at the beginning of the experiment for mCherry (Ex. 575/25 nm / Em. 625-45 nm). For the analysis of the FRET signal, the background- subtracted CFP and YFP images were divided, to get a ratiometric image (CFP/YFP) using ImageJ and custom plug-ins to track cells (Albeck et al. 2008. Mol. Cell. 30(1): 11-25). Signals were normalized by subtracting the minimum value across all time points from each single-cell time course and the average trajectory of the corresponding drug vehicle-treated cells. Cell death times were determined by visual inspection, for each corresponding cell trajectory with the transmitted light image stacks, and used to determine population cell viability. Caspase-8 activity and maximum Caspase- 8 activity were determined from CFP/YFP ratio trajectories, as described before (Roux et al, 2015).

Laser capture microdissection

Right after a 50 min-live cell microscopy, the reaction was stopped by replacing the FrameSlide cell media with PBS Ca2+/Mg2+ at 4dC, and non-contact laser capture microdissection was performed with a Zeiss PALM MicroBeam (Carl Zeiss France). Cells were ranked based on their caspase-8 activity at 45-50min determined by live cell microscopy image and signal analyses, to pick “high” and “low” cells, and their xy coordinates were converted to position the Zeiss RoboStage accordingly (using MATLAB scripts developed in house). Chosen single-cells were then catapulted into 0.2ml domed caps tube (strips) containing 4.3 pi of lysis buffer (0,2% w/v Tween 20, 1.5U/pl Promega RNAseIN Plus). Tubes where kept on ice and frozen in dry ice directly after collection to minimize RNA degradation during sample collection and transport.

Isolated single-cell sequencing (after laser-capture microdissection) cDNA preparation

Samples were thawed on ice and 2.43 mΐ of a mix containing 10,000 molecules Life Technologies ERCC spike-In Mix 1, 2.5mM dNTPs, and 3.7 mM single barcoded reverse transcription primer was added to each tube, incubated for 10 min at room temperature, 3 min at 70°C and kept at 10°C. Immediately after lysis, 8.27 mΐ 1.8X reverse transcription buffer (Invitrogen), 4.53mM DTT, 1.8M Betaine, 10.8 mM MgC12, 3.62 mM template switching oligonucleotide, 0.6U/pl Promega RNAse IN Plus, 7.5 U/mI Invitrogen Superscript II reverse transcriptase were added and samples were reverse transcribed 10 min at 25°C, 90 min at 42°C, 15 min at 70°C and kept at 10°C until PCR amplification. Samples were PCR amplified (KAPA HiFi Hotstart) 20 cycles in 31 pi with 0.8mM forward PCR primer, 0.8mM biotinylated reverse PCR primer. Samples were incubated 3 min at 98°C followed by 20 cycles at 98°C for 20 sec, 64°C for 15 sec, 72°C for 6 min followed by a final extension for 5 min at 72°C. After PCR amplification, samples were pooled and purified with 0.8X Beckman Coulter SPRIselect beads, washed once with 85% EtOH and eluted in 30 mΐ.

Library preparation

Batches of barcoded cDNAs were pooled, purified with SPRIselect beads (0.8x) and eluted in 12m1. Nine mΐ (typically 10 - 20 ng of cDNA) were used for tagmentation and Illumina library preparation as described before (Arguel et al., 2017).

Sequencing

Libraries were sequenced on an Illumina NextSeq500, on a High Output Flow Cell, 20 nt. index 1 read (8bp cell barcode and 12 bp UMI) and 71 nt. read 1 for cDNA sequencing, with custom sequencing primers as described in (Arguel et al., 2017).

Data analysis of isolated single-cell sequencing

Bioinformatics analyses

Mapping of cDNA reads was performed using STAR against human hgl9 build following encode RNA-seq recommendations. Drop-seq tools (http://mccarrolllab.com/, (Macosko et al, 2015)) was then used to tag SAM records with cell barcode (XC) and molecule identifier (XM) and produce the digital expression count matrix.

Statistical analyses

Exploratory analysis of single cell RNAseq data was performed for quality control and low quality cells were excluded from the final datasets. Library size normalization and highly variable genes selection (compared to ERCC spike ins) were performed using the R package scran. Cell cycle assessment was performed using the algorithm described before (Macosko et al, 2015), based on our curated gene sets (Revinski et al, 2018). Hierarchical clustering and heatmaps were generated using the R package pheatmap, using Euclidean distance and Ward.D method unless otherwise specified. Differential analysis was performed using the edgeR likelihood ratio test framework. P-values were adjusted for multiple testing using the Benjamini-Hochberg method. Gene set enrichment analyses were performed using Ingenuity Pathway Analysis (IP A, QIAGEN Inc).

Droplet based single-cell sequencing

Single cell preparation Clonal HeLa cells stably expressing the FRET-based initiator caspase reporter (ICRP) were cultured in 6-well plates for 24h and then treated with TRAIL. Right after treatments, cells from each treatment conditions, were detached with trypsin (filtered, counted) and incubated with Hashtag antibodies (Biolegend TotalSeq™, San Diego, CA) for 20min at 4°C. Live cells from each treatment conditions were then counted and merged into one sample according to the manufacturer’s recommendations (lOx Genomics, Netherlands). Cell Hashing protocol with Hashtag oligonucleotide (HTO) was then conduct with lOx Genomics single-cell 3' V2 chemistry to generate single-cell emulsion.

Single cell and Hashtag oligo sequencing

Single-cell RNA-seq was performed following manufacturer’s protocol (Chromium™ Single Cell 3' Reagent Kit, V2 Chemistry) to obtain single cell 3’ libraries for Illumina sequencing. PCR of 12 cycles was performed for cDNA amplification with HTO primer (Stoeckius et al, 2018). Final cDNA library was then pooled with 10% of HTO library in final pool for sequencing.

Libraries were sequenced with a NextSeq 500/550 High Output v2 kit (75 cycles) that allows up to 91 cycles of paired-end sequencing: Read 1 had a length of 26 bases that included the cell barcode and the UMI; Read 2 had a length of 57 bases that contained the cDNA insert; Index reads for sample index of 8 bases. Cell Ranger Single-Cell Software Suite v3.0.2 was used to perform sample demultiplexing, barcode processing and single cell 3' gene counting using standards default parameters and human build GRCh38. Hashtag library prep and analysis

For HTO library preparation, 10 cycles of PCR amplification (Stoeckius et al., 2018) followed by a size selection with Blue Pippin 3% agarose gel cassette (BDQ3010) Marker Q3 were necessary to select 180bp specific amplicon. Antibody hashtags counting was done using CITE-seq-Count (Stoeckius et al., 2017) using standards lOxgenomics parameters “-cbf 1 -cbl 16 -umif 17 -umil 26 -hd 2” and barcode whitelist” from Cell Ranger analysis.

Data analysis of droplet based single-cell sequencing

Hashtag demultiplexing and preprocessing Single cell RNAseq and hashtag data were analyzed using the R toolkit Seurat V3 and custom R scripts. Hashtag counts were normalized using the CLR method, then cell barcodes were demultiplexed and doublet and negative cells were excluded using the HTODemux function with parameter positive. quantile = 0.95. Low quality cells were excluded from the dataset based on the following criteria: number of genes detected < 2,000, percentage of UMI mapping to mitochondrial genes > 5% or number of UMIs <

10,000. The final number of cells included was: 1529 for control sample, 1481 for TRAIL 50mn 25ng/ml, 1631 for TRAIL 120mn lOng/ml, 1804 for TRAIL 120mn 25ng/ml and 2115 for TRAIL 120mn 40ng/ml. Cell cycle scores were calculated using Seurat CellCycleScoring function, based on previously published gene sets (Revinski et al, 2018). Each sample was first normalized to 10,000 UMIs, and variable features were selected using the vst method.

Analysis of control samples

Analysis of the control sample alone was performed using the standard Seurat workflow based on a number of PC = 10 and clustering resolution parameter res=0.3. Integration of control and treated samples

The inventors used the Seurat V3 data integration pipeline to analyze all 5 samples together. Integration was performed based on the first 20 dimensions, and a regression of the G2M and S scores was included to remove the effect of cell cycle. PCA was used as dimensionality reduction technique, and the number of PCs to include in downstream analyses was set based on inspection of the elbow plot. A resolution parameter res=0.1 was used for the clustering. Significance of differential expression analysis between clusters or between experimental conditions was performed using Wilcoxon Rank Sum test using the FindMarker function.

Pseudo-bulk analysis Two pseudo-bulk samples were constructed from each single cell samples by randomly selecting 700 cells (without replacement) among all cells passing quality filter, then, the inventors calculated the UMI count value for each gene by adding the UMIs counts corresponding to this gene from these 700 cells. Statistical analysis of bulk samples was performed using the R package edgeR. Library size differences normalization factors were calculated using the TMM method. Differential analysis was performed using the likelihood ratio test. Pvalues were adjusted for multiple testing using the Benjamini- Hochberg method. Heatmaps were generated using the R package pheatmap. Gene set enrichment analysis was performed using the package fgsea.

Transcriptional noise analysis

Raw single cell count data was used for this analysis. The following preprocessing was applied in order to account for differences in sequencing depth for each cell and for differences in the number of single cell captured for each sample. First, we started by randomly selecting 1000 cells among all cells with at least 20,000 UMIS in each sample. Then, for each cell, the number of UMIs was downsampled to 20,000 using the DownSample function in Seurat. For each sample, the transcriptional noise was measured as 1- rho, where rho is the Spearman correlation coefficient between 2 cells, for all possible pairs of cells in this sample (Angelidis et ak, 2019. Nat Commun. 10(1):963).

2. Results

Drug response phenotype switching evidenced by clonal expansion and drug selection.

To show the interconversion property of the two cell-response phenotypes of the clonal cell population (the capacity to survive or to commit to cell death after treatment), the inventors performed several rounds of clonal expansion and drug selection of HeLa cell cultures. HeLa cells were either sub-cloned into new cultures, or treated with TRAIL at around the IC50 (25ng/ml for 24hrs, Fig. 1 A). The surviving cells from the first treatments were allowed to regrow for 72 hours. Once the surviving cells and the sub-clones had regrown to new cultures (72 hours and one week, respectively), they were all treated with TRAIL for 24hrs at several doses, and their responses were compared to the parental cultures. As evidenced by the dose-response curves, neither clonal expansion nor drug selection did affect the proportioning of responding and non-responding cells to TRAIL treatments; both subculture strategies produced cell populations of similar drug-response profiles as the parental cells (Fig. 1A). These experiments show that each single-cell can give rise to a seemingly homogenous cell population, which invariably exhibits the same drug response heterogeneity.

Even isogenic populations contain cells in different states, such as the different phases of the cell cycle. Cell cycle genes are predominant marker of cell states, therefore, current single-cell analyses have regressed out these genes to uncover other gene signatures that distinguish cells types or discover sub-clones (Buettner et al., 2015. Nat Biotechnol. 33(2): 155-60). However, there is still very limited knowledge on different cell states signatures within the same cell type that could explain the fractional response observed here. Other studies have suggested that fractional killing arises from pre-existing difference in levels of apoptotic proteins, linked to stochasticity in gene expression, as cell death time decorrelates rapidly in daughter cells, unless translation is inhibited (Flusberg et al, 2013. Mol. Boil. Cell. 24(14):2186-200; Spencer et al., 2009. Nature. 459(7245):428-32).

Given that the fractional response observed by the inventors, is a stable property of the cell population, the inventors first sought to determine by single-cell RNA sequencing (scRNAseq), whether the transcriptomic profiles of the resting cells could reveal the presence of groups in relevant cell states (pro-, anti-apoptotic). The inventors profiled over a thousand untreated HeLa cells using the 10X Genomics platform. The inventors first performed analyses using nonlinear dimensionality-reduction techniques which showed that single-cell clusters were primarily driven by the cell cycle genes as observed in other single-cell studies (Buettner et al., 2015. Nat Biotechnol. 33(2): 155-60). After regressing out the cell cycle component using the inventors cell cycle gene set, the inventors could no longer find robust clustering of cell states, that would predict of a cellular response relevant to TRAIL signaling.

Reasoning that detectable changes could arise within the first hours after treatment, the inventors repeated the experiments with HeLa cells treated with TRAIL (10, 25 and 40ng/ml) for 1 and 2 hours (as significant cell death was observed afterwards, precluding scRNAseq analyses). Here again, no significant groups emerged after cell cycle genes regression (Fig. IB), and no significant differentially expressed (DE) genes was detected after 1-hour treatment (Fig. 1C). It required 2 hours to start detecting some DE genes in HeLa cell treated with TRAIL, with only the TNF signaling hallmark pathway showing significant enrichment. Further single-cell analyses revealed that treatment time and doses marginally affected gene expression noise, suggesting that pre-existing cell-to-cell variability is not changed shortly after treatment.

Therefore, gene expression noise greatly impairs single-cell approaches from the finding of the molecular determinants of cell decision in isogenic populations, even though their underlying cell states have been phenomenologically evidenced by drug response phenotype switching and ultimately fractional killing (by us and others (Inde and Dixon, 2018. Crit. Rev. Biochem. Mol. Biol. 53(1):99-114; Mitchell et al, 2018. Proc. Natl. Acad. Sci. U S A. 115, E2888-E2897)).

Apart from cell cycle genes, one hypothesis for the undetected (or measurable) differences in seemingly homogeneous cell populations is that they contain cells in a wide variety of possible cell states that predisposed them to a number of responses or functions such as cell death, confounding pathway enrichment analyses. As a consequence, the measured variability of isogenic cells has so far been referred as gene expression noise (unstructured variability (Eling et al., 2019. Nat. Rev. Genet. 20(9):536-548)) instead of an observable signal, because of the lack of prior knowledge on cell types or on the actual cell state (for the studies focusing on clonal cell populations).

To recover the molecular determinants of clonal cells response to cancer drugs from the measured gene expression variability, the inventors assembled a same-cell functional pharmacogenomics approach: it couples the prior knowledge of the cell state (predicted drug response) to the transcriptomic profile of the same cell. The workflow is based on live-cell microscopy and laser capture microdissection, coupled to single-cell RNA Sequencing (LCM2Seq, Fig. 2).

Phenotype-coupled same-cell profiling, LCM2Seq

The inventors have previously shown that during the course of TRAIL treatment, the response phenotype of a single-cell was correlated with its initiator caspase (caspase-8) maximal activity (hereafter maxC8), with more than 80% accuracy (Roux et al, 2015. Mol Syst Biol. 11(5):803). However, the time of maxC8 varied greatly between cells, and for the responding ones, it occurred at cell death (which defied its predictive value and only allowed the recovery of non-responding cells).

By evaluating the caspase-8 activation rates (time derivative of caspase-8 biosensor FRET ratio (Roux et al., 2015. Mol Syst Biol. 11(5):803)) early after drug addition and associating them to the response phenotypes of each cell at the end of the experiment (dying or surviving cell), the inventors found, a significant predictive value caspase-8 activity for cell fate determination at 45-50 minutes after treatment (Fig.3A-B). This prediction could be done for -25% of the cell population (most differing cells) and was of particular interest because (1) it allowed to determine accurately the response phenotype of a cell before a transcriptional response could induce major confounding changes (as shown in Fig 1C) and because (2) it allowed to recover both responding and non-responding cells. The inventors used this predictive metric (time derivative of the FRET ratio at 45-50 minutes, Fig. 2A) to choose cells based on their predictive response phenotype, predicted responding and predicted non-responding (Fig. 2B). Each chosen cell had been tracked and labelled so to be isolated in a tube (one cell per tube) by laser-capture microdissection, and then, processed for single-cell RNA sequencing on isolated cell (Fig. 2C). Single-cell RNAseq was performed on cells for which the phenotype is predicted with an accuracy greater than 95%. It is important to note here that a single measure of FRET at 45 minutes would not be sufficient to determine the cell fate, as the predictive metric is based on the time derivative of the FRET ratio and requires at least two consecutive measures of the same cell in time (and proper normalization to the same cell at time 0). So far, features of cell signaling heterogeneity could help distinguishing drug response phenotypes between different cell types (Singh et al, 2010. Mol Syst Biol. 6:369), now this workflow enables the transcriptomic analysis within an isogenic population, of each cell response profile independently.

Same-cell Functional Pharmacogenomics reveals drug sensitivity signature within gene unstructured variability

In order to evaluate the gene expression state at the origins of the opposing cell responses to TRAIL, the set of sequenced single cells was composed of high responders (predicted responding cells) and low responders (predicted non-responding cells), all from the same isogenic population of HeLa cells briefly stimulated with the same TRAIL treatment (25ng/ml). In this framework, it is important to note that the cells are being compared among themselves based on their predicted phenotype and they are not compared to an untreated control as it is classically done. To illustrate this point and to verify that this set of isogenic cells treated equally, are indeed homogenous in their transcriptomic profile, the inventors first analyzed their transcriptome without considering their predicted phenotype. First, the inventors could confirm by principal component analysis and unsupervised hierarchical clustering that predicted responding and non-responding cells do not cluster in separate regions of the first two principal components landscape (Fig. 4 A), and that the clustering was not linked to the presence of confounding factors such as batches of single cell collections, cell cycle phase prediction, percentage of dropouts and even predicted phenotype (Fig. 4B,E). Importantly these results indicate that although the cell fate decision is already determined for these chosen cells (highly accurate prediction), their potential expression difference could be confounded within gene expression noise. To further verify this finding, the inventors performed a filtering of highly variable genes (HVG) within these isogenic cells treated identically, based on their distance relative to ERCC spike-ins (Fig. 4C), as described before (Lun et al, 2016. FlOOORes. 5:2122). Then, gene expression correlation for the genes with the most significant biological variation was assessed and again showed that no feature could recover gene covariations between the two predicted cell response phenotypes (Fig. 4D). These results demonstrate that in the case of isogenic cell populations, unstructured variability also characterizes the variability observed in the cells that committed to a phenotypic response (high and low responders).

The inventors next tested whether grouping cells by their predicted response phenotype would uncover structured variability. The inventors found 135 genes with statistically significant differential expression between the predicted responding and non-responding cells (Fig. 6 A-C). Interestingly now, pathway enrichment analyses significantly ranked functions such as “Cell death and Survival”, indicating that the inventors increased the meaningful expression signal. Hierarchical clustering of the grouped cells with the most significant genes highlighted the heterogeneity of anticancer drug sensitivity information contained in identical cells treated together and revealed a TRAIL- sensitivity gene signature of HeLa cells treated equally (Fig. 5).

The inventors derived a short list of target genes intersecting on a couple of statistical criteria (minimal false discovery rate (FDR), minimal dropout rate, maximal fold change and with a well-balanced expression level, Fig. 7A-B). Although they exhibit a relatively high Fano factor (expression variability), the inventors found that the most discriminating genes were not the most variable genes, suggesting that once the cells are committed to a cellular function, the expression of the genes related to that function, stabilizes.

Target genes functional validation To further validate the function of the DE genes between the two predicted cell responses, the inventors engineered cell lines to overexpress the short listed genes that were found highly expressed in predicted responding cells when compared to predicted non responding cells (BLOC1S1, PRSS56, SIVA1, SLC25A1, UBE2D4 and UQCC3 “Cl lorf83”). Using live-cell microscopy, the inventors compared each cell line to a control cells (established in parallel but overexpressing a fluorescent protein only) by monitoring their caspase-8 activation, the time and fraction of cell death after TRAIL treatment.

The inventors also measured their capacity to regrow colonies after TRAIL treatment. The inventors found that overexpression of PRSS56, SIVA1, SLC25A1, UBE2D4 and UQCC3 significantly increased caspase-8 activity and cell sensitivity to TRAIL (except BLOC1S1 which effect was more variable between experiments), and that all significantly reduced the growth of non-responding colonies (Fig. 8); effects that have not been reported before. Other genes that were either further down the list of DE genes or had previously described interactions with caspase-8 were used as controls. The inventors next used inhibitor drugs to test whether the top gene target DNM1L, which was found to be highly expressed in predicted non-responding cells would increase cell sensitivity to TRAIL.

The inventors found that both Dynasore and Mdivi-1 could significantly increase TRAIL- induced caspase-8 activity (so far unreported), and consequently cell death response to TRAIL (Fig. 9). The inventors concluded that phenotype-coupled single-cell profiling successfully identified a set of target genes that could increase cell response to TRAIL.

Unexpectedly, by evaluating the shortest path between the target genes and caspase-8, the inventors found that at equal distance: the noisier was the expression in parental cells (highest Fano factor), the larger effect a gene overexpression had on caspase-8 activity and ultimately, cell death, supporting evidences that signaling networks utilize noise to increase population-level information transfer and regulate the overall cell population response (Suderman et al., 2017. Proc Natl Acad Sci U S A. 144(22):5755-5760).