bioRxiv preprint doi: https://doi.org/10.1101/693010; this version posted July 5, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

1 SynToxProfiler: an approach for top drug combination selection based on integrated profiling of 2 synergy, toxicity and efficacy 3

1,2 1 1 1,2,3* 1* 4 Aleksandr Ianevski , Alexander Kononov , Sanna Timonen , Tero Aittokallio , Anil K Giri 5 1 Institute for Molecular Medicine Finland (FIMM), University of Helsinki, FI-00290 Helsinki, 6 Finland. 7 2 Helsinki Institute for Information Technology (HIIT), Aalto University, FI-02150 Espoo, Finland 8 3 Department of Mathematics and Statistics, University of Turku, Quantum, FI-20014 Turku, Finland 9 10 *Corresponding author 11 12 13 Abstract 14 Drug combinations are becoming a standard treatment of many complex diseases due to their 15 capability to overcome resistance to monotherapy. Currently, in the preclinical drug combination 16 screening, the top hits for further study are often selected based on synergy alone, without considering 17 the combination efficacy and toxicity effects, even though these are critical determinants for the 18 clinical success of a therapy. To promote the prioritization of drug combinations based on integrated 19 analysis of synergy, efficacy and toxicity profiles, we implemented a web-based open-source tool, 20 SynToxProfiler (Synergy-Toxicity-Profiler). When applied to 20 anti-cancer drug combinations tested 21 both in healthy control and T-cell prolymphocytic leukemia (T-PLL) patient cells, as well as to 77 anti- 22 viral drug pairs tested on Huh7 liver cell line with and without Ebola virus infection, SynToxProfiler 23 was shown to prioritize synergistic drug pairs with higher selective efficacy (difference between 24 efficacy and toxicity level) as top hits, which offers improved likelihood for clinical success. 25 26 Introduction 27 High throughput screening (HTS) of approved and investigational agents in preclinical model systems 28 has been established as an efficient technique to identify candidate drug combinations to be further 29 developed as safe and effective treatment options for many diseases, such as HIV, tuberculosis and 30 various types of cancers [1, 2]. Currently, the selection of top combinations for further development 31 often relies merely on the observed synergy between drugs, while neglecting their actual efficacy and 32 potential toxic effects, that are the other key determinants for the therapeutic success of drugs in the 33 clinics [3]. Notably, around 20% of drugs fail in the early development phase because of safety

1 bioRxiv preprint doi: https://doi.org/10.1101/693010; this version posted July 5, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

34 concerns (non-tolerated toxicity), and over 50% fail due to lack of sufficient efficacy [4]. Further, a 35 recent study argued that many clinically-used anticancer combination therapies confer benefit simply 36 due to patient-to-patient variability, not because of drug additivity or synergy [3], indicating that even 37 non-synergistic combinations may be beneficial for therapeutic purposes if they have a high enough 38 efficacy and low enough toxicity profiles. To make a better use of these various components of drug 39 combination performance already in preclinical HTS experiments, we implemented, to the best of our 40 knowledge, the first web-tool, SynToxProfiler, which enables users to profile synergy, toxicity and 41 efficacy of drug combinations simultaneously for the top hit prioritization and is also extendible for 42 multi-drug (3 or more drugs) combination screening. 43 44 Methods 45 SynToxProfiler workflow 46 The SynToxProfiler web-application is freely available at https://syntoxprofiler.fimm.fi, together with 47 example drug combination data, video tutorial and user instructions. SynToxProfiler enables ranking of 48 drug combinations based on integrated efficacy, synergy and toxicity profiles (Fig.1). Therefore, for 49 each drug combination, SynToxProfiler first calculates a normalized volume under dose-response 50 surface to quantify combination efficacy based on dose–response measurements on diseased cells, e.g. 51 patient derived primary cells (see Suppl. Fig. 1). Then, the combination synergy between each drug 52 pair is estimated using one of the synergy scoring models: Highest Single-Agent [5], Bliss 53 independence [6], Loewe additivity [7], or Zero Interaction Potency [8], as implemented in the 54 SynergyFinder R-package [9]. Normalized volume under the dose-synergy surface is utilized to 55 quantify final combination synergy score (Suppl. Fig. 1A). Next, using the measurements on control 56 cells, if available, the normalized volume under dose–response matrix is calculated to estimate 57 combination toxicity (Suppl. Fig. 1). Finally, SynToxProfiler ranks the drug combinations based on 58 integrated combination synergy, efficacy and toxicity (STE) score. Alternatively, if measurement on 59 control cells are not available, then the ranking of drug pairs can also be done based merely on 60 combination synergy and efficacy. As a result of the interactive analysis, SynToxProfiler provides a 61 web-based exportable report, which allows users to interactively explore their results (Fig. 1 and 62 Suppl. Fig 2). An interactive example of web-based report is given at 63 https://syntoxprofiler.fimm.fi/example. A more detailed description of the calculations and workflow is 64 provided in the technical documentation, https://syntoxprofiler.fimm.fi/howto. 65

2 bioRxiv preprint doi: https://doi.org/10.1101/693010; this version posted July 5, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

66 67 Fig. 1. A schematic overview of SynToxProfiler. The dose-response data from drug combination 68 screening, measured in both diseased (e.g. patient-derived cells) and healthy control cells (e.g. 69 PBMCs), is provided as input to SynToxProfiler (left panel). Then, SynToxProfiler quantifies drug 70 combination efficacy and synergy (using combination responses in diseased cells) as well as toxicity 71 (using combination responses in control cells) for each drug pair (middle panel), and summarizes them 72 into integrated synergy, toxicity and efficacy (STE) score. The STE score is further used to rank and

73 visualize the drug pairs in 2D or 3D interactive plots (right panel). 74 75 Calculation of normalized volume 76 The normalized volume under the dose-response surface is calculated while quantifying combination 77 efficacy and toxicity based on measurements on diseased and control cells, respectively (Suppl. Fig. 1). 78 Synergy score was calculated based on measurements on diseased cells as normalized volume under 79 synergy matrix (excess matrix of combination responses over expected responses determined by one of 80 the synergy models, such as Bliss). For each combination AB of drugs, A and B, the normalized

81 volume under the dose-response surface VAB is calculated as: † ∑∑ͷ«Ÿ¶ ͷ«Ÿ¶ Ex, yΔc͵ΔcͶ ΦͰͷ ΧͰͷ† V «§¬ «§¬ . Eq. 1 ͵Ͷ ͵ ͵ Ͷ Ͷ lnCΛΏΦ/CΛΗΜ lnCΛΏΦ/CΛΗΜ A A 82 Here, c min and c max are the minimum and maximum tested concentrations of drug A, respectively, B B A B 83 and c min, and c max are those of drug B; Δc and Δc are the logarithmic increase in concentration of 84 drug A and drug B between two consecutive measurements of dose-response matrix; and E(x, y) is the 85 efficacy or toxicity levels at concentration x of drug A and at concentration y of drug B. The current

3 bioRxiv preprint doi: https://doi.org/10.1101/693010; this version posted July 5, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

86 approach for volume-based scoring normalizes for the different dose-ranges measured in different drug 87 combinations, as commonly occurring in HTS settings. The extension of formulation for volume - 88 based scoring of synergy, efficacy and toxicity profiles for multi-drug combinations (3 or more drugs) 89 is given in the supplementary file. 90 91 Ranking of drug combinations 92 SynToxProfiler ranks the drug combinations based on an integrative analysis of synergy, toxicity and

93 efficacy, quantified as STE score. First, the difference in efficacy (EAB) and toxicity volume scores

94 (TAB) is calculated for each drug combination to quantify a selective response in diseased cells, relative

95 to that of control cells. We defined this difference as a selective efficacy score (sEAB) of a drug 96 combination. This theoretical concept for selective efficacy has been adopted from the single drug 97 dose-response assays, where the difference in normalized areas under the curve (AUC) between 98 diseased and healthy cells is often used to calculate the patient-specific drug efficacies [10, 11]. The

99 final STE score is given by averaging two different ranks of (i) combination synergy score (SAB) (the

100 higher is the synergy, the higher is the rank), and (ii) selective combination efficacy (sEAB) (the higher 101 is selective efficacy, the higher is the rank): rankS ranksE STE ͵Ͷ ͵Ͷ , Eq. 2 ͵Ͷ 2N

102 where SAB and sEAB are the synergy and selective efficacy scores, respectively, for a combination of 103 drug A and B, calculated using the normalized volume under the dose-response surface; and N is the 104 total number of drug combinations being tested. However, since calculation of STE score using the 105 whole dose-response matrix may miss some of the top hit drug combinations with a narrow synergistic 106 dose window, SynToxProfiler also offers the users a possibility to rank combinations based on the 107 selective efficacy and synergy scores calculated only at the most synergistic area of the drug 108 combination matrix (defined as the 3x3 concentration window with the highest synergy in the dose- 109 response matrix), instead of the default full matrix calculation. 110 111 Data submission and reporting 112 The default input of SynToxProfiler is a text or xlsx file that comprises annotations of each drug 113 combination dose–response matrix, including drug names, concentrations, cell types (e.g. sample or 114 control), and phenotypic responses (e.g. relative inhibition). The number of drug combinations 115 provided in the input file is unrestricted. More information on the input file format is given in the 116 website documentation (https://syntoxprofiler.fimm.fi/howto/). As the result, SynToxProfiler provides 117 an interactive visualization of STE scores using bar charts, as well as 2- and 3-dimensional scatter

4 bioRxiv preprint doi: https://doi.org/10.1101/693010; this version posted July 5, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

118 plots. Publication-quality figures (e.g. heatmap for dose-response and synergy matrix, 2D and 3D 119 scatter plot for different scores) can be exported in PDF files, as well as all the calculated scores can be 120 downloaded in an xlsx file. 121 122 Drug combination assay 123 The in-house drug combination testing was carried out at Institute for Molecular Medicine Finland 124 (FIMM), in peripheral blood mononuclear cells (PBMCs) of a patient with T-cell prolymphocytic 125 leukemia (T-PLL) and a healthy volunteer were used in accordance with the regulations of Finnish 126 Hematological Registry and biobank (FHRB). The written informed consents were obtained from both 127 participants and the study was carried in accordance with the principles of Helsinki declarations. 128 Twenty combinations of drugs with different mechanisms of actions (see Supplementary Table S1) 129 were tested on the PBMCs in 8x8 dose-response matrix assay as described previously [12, 13]. Briefly, 130 20 microliters of cell suspension along with compounds (in 8 different concentrations including zero 131 dose) and their combinations were plated on clear bottom 384-well plates (Corning #3712), using an 132 Echo 550 Liquid Handler (Labcyte). The concentration ranges were selected for each compound 133 separately to investigate the full dynamic range of dose-response relationships. After 72 hours 134 incubation at 37°C and 5% CO2, cell viability of each well was measured using the CellTiter-Glo 135 luminescent assay (Promega) and a Pherastar FS (BMG Labtech) plate reader. As positive (total 136 killing) and negative (non-effective) controls, we used 100 μM benzethonium chloride and 0.1 % 137 dimethyl sulfoxide (DMSO), respectively, for calculating the relative efficacy (% inhibition). 138 The published dataset of 78 antiviral drug combinations was tested at the Integrated Research 139 Facility, National Institutes of Allergy and Infectious Diseases (NIAID), in the Huh7 liver cells 140 infected with Makona isolate, Ebola virus/H.sapiens-tc/GIN/14/WPG-C05, as described in the original 141 study [14] (data available at https://matrix.ncats.nih.gov/matrix-client/rest/matrix/blocks/6323/table 142 and https://matrix.ncats.nih.gov/matrix-client/rest/matrix/blocks/6324/table). Briefly, drugs in 50-µL of 143 Dulbecco’s modified Eagle’s medium were transferred to the Huh7 cells seeded in black, clear- 144 bottomed, 96-well plates 1 hour prior to inoculation with EBOV/Mak. After 48 hours of viral 145 inoculation, drug combination efficacy was measured in triplicates with a 6 × 6 dose-response matrix 146 design using CellTiter-Glo assays (Promega). The EBOV/Mak virus was detected using mouse 147 antiEBOV VP40 antibody. For the toxicity measurements, the same CellTiter-Glo assay was 148 performed on non–virus-infected Huh7 cells with 3 replicates for each drug concentration, and the 149 assay was repeated at least twice for confirmation. We utilized 77 out of the 78 combinations for the 150 present analysis, as colchicine-colchicine pair was removed because the inhibition levels were 100% 151 for all the tested concentrations for the drug combination.

5 bioRxiv preprint doi: https://doi.org/10.1101/693010; this version posted July 5, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

152 Results 153 SynToxProfiler prioritizes clinically useful drug combination as top hits for T-PLL cancer patient and 154 Ebola virus infection 155 To demonstrate the performance of SynToxProfiler in prioritizing therapeutically-relevant synergistic 156 combinations, we applied it to in-house drug screening data involving 20 drug combinations tested in 157 one control and one T-PLL patient-derived cells. The T-PLL case study revealed that ranking of 158 combinations based on the STE score successfully prioritizes both effective and safe drug pairs. For 159 example, Cytarabine-Daunorubicin pair was identified as the top hit out of the tested combinations 160 (Table 1, Additional File 1); this combination is widely used as approved induction therapy for acute 161 myeloid leukemia treatment [15,16]. Ibrutinib-Navitoclax was ranked as the third-best combination for 162 further study; this combination has shown promising results in phase II clinic trail (NCT02756897) for 163 chronic lymphocytic leukemia (CLL), and recently suggested as first-line treatment for CLL [17]. 164 165 Table 1: Ranking of 20 in-house measured combinations based on STE scores calculated from the 166 most synergistic area of dose-response matrix in T-PLL and healthy control cells

Drug1 (concentration range in Drug2 (concentration range Synergy score Efficacy Toxicity Selective efficacy STE

nM) in nM) (SAB) score (EAB) score (TAB) score (EAB) score

Cytarabine (0 -100) Daunorubicin (0 - 1000) 6.701 58.727 20.864 37.863 0.825

Trametinib (0 -100) S-63845 (0 - 25) 4.202 48.4 12.427 35.973 0.825

Ibrutinib (0 -1000) Navitoclax (0 - 100) 1.529 71.686 19.922 51.764 0.825

Quizartinib (0 -100) S-63845 (0 - 100) 0.69 56.379 27.43 28.949 0.75

Omacetaxine (0 -1000) Ipatasertib (0 - 1000) 0.188 70.096 11.361 58.735 0.725

Gefitinib (0 -1000) Omacetaxine (0 - 1000) 0.693 69.355 17.071 52.284 0.725

Clofarabine (0 -1000) Idarubicin (0 -100) 2.075 73.876 27.528 46.348 0.725

Omacetaxine (0 -1000) Alpelisib (0 - 1000) -0.723 70.87 10.947 59.923 0.675

Clofarabine (0 -1000) Prexasertib (0 - 1000) 1.59 19.695 8.899 10.796 0.675

Gefitinib (0 -25) Trametinib (0 - 1000) 0.402 10.254 4.513 5.741 0.55

Buparlisib (0 -100) Ibrutinib (0 - 1000) 0.937 4.449 1.333 3.116 0.55

Ibrutinib (0 -100) Doxorubicin (0 - 100) 0.84 9.822 1.734 8.088 0.525

Vinorelbine (0 -1000) Clofarabine (0 - 1000) -1.143 66.463 21.399 45.064 0.5

Clofarabine (0 -1000) Omacetaxine (0 - 1000) -4.266 20.336 0.585 19.751 0.375

Dexamethasone (0 -1000) Clofarabine (0 - 1000) -16.368 0 0 0 0.35

Dasatinib (0 -1000) Ipatasertib (0 - 100) 0.237 0.672 2.077 -1.405 0.25

Carboplatin (0 -1000) Dexamethasone (0 - 1000) -0.507 0 0 0 0.2

Ipatasertib (0 -1000) ASP3026 (0 - 1000) -0.009 0 0 0 0.175

Idarubicin (0 -100) Ibrutinib (0 - 100) -0.893 0.689 2.28 -1.591 0.175

Trametinib (0 -100) Dasatinib (0 - 25) -2.049 0 1.232 -1.232 0.1

6 bioRxiv preprint doi: https://doi.org/10.1101/693010; this version posted July 5, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

167 Synergy scores were calculated using ZIP [8] model (default option in SynToxProfiler). 168 169 To further illustrate the wide applicability of SynToxProfiler also in non-cancer combinatorial screens, 170 we used a published dataset of 77 drug combinations tested as anti-viral agents where the drug 171 combinations’ efficacy and toxicity were tested in Ebola-infected and non–virus-infected Huh7 liver 172 cells, respectively. SynToxProfiler ranked established combinations (e.g. clomifene-sertraline and 173 sertraline-toremifene) that inhibit EBOV fusion to cell surface as top hits for further study (Additional 174 File 2). All the three drugs (clomifene, sertraline and toremifene) showed survival benefit in in-vivo 175 murine Ebola virus infection model [18], indicating that SynToxProfiler prioritizes drug pairs with a 176 strong potential to be rapidly advanced towards clinical settings and used as therapeutic interventions. 177 178 Top hits selected by SynToxProfiler based on integrated scoring are synergistic drug pairs with higher 179 selective efficacy 180 We compared the synergy and selective efficacy level of the top hits prioritized based on the STE 181 score, synergy score and selective efficacy scores, using the 77 combinations in the Ebola dataset. The 182 top combinations identified by STE scores had a notably higher selective efficacy as well as higher 183 synergy (shown by arrow in Fig. 2A), indicating that STE score represents a proper balance between 184 high selective efficacy and synergy. Additionally, we observed a marked overlap (65%) between the 185 top-10% of analyzed combinations prioritized based on STE score and synergy score, as well as based 186 on the STE score and selective efficacy score (50% overlap), as shown in Fig. 2B. In contrast, there 187 was a smaller overlap (41%) between the top-10% hits selected based on selective efficacy and 188 synergy scores. Further, a low Pearson correlation (r=0.22) between selective efficacy and synergy was 189 observed. These results indicate that synergy and selective efficacy are independent drug combination 190 components, which cannot be used alone to prioritize potent and less toxic synergistic drug 191 combinations.

7 bioRxiv preprint doi: https://doi.org/10.1101/693010; this version posted July 5, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

192 193 194 Fig. 2. STE score considers both synergy and selective efficacy when prioritizing potent drug 195 combinations. (A) 3D surface shows increase in STE score with increasing synergy and selective 196 efficacy scores across 77 antiviral combinations measured in Huh7 liver cell line infected with Ebola 197 virus (the arrow marks the gradient of the increase in STE score). The 3D surface is fitted by a 198 generalized additive model with a tensor product smooth, implemented in mgcv R package. (B) Scatter 199 plot showing the overlap in the top hits selected on the basis of different scores (the dotted vertical line 200 denotes the overlap between the top 10% combinations selected based on any of the three scores). 201 202 A more detailed analysis revealed that SynToxProfiler ranks lower the toxic drug pairs despite their 203 higher synergy (e.g. clomifene-colchicine and toremifene citrate-apilimod). For example, 204 SyntoxProfiler ranked clomiphene citrate and sertraline HCl combination (STE=0.96) as the top hit 205 (Fig.3), despite its lower synergy as compared to more synergistic toremifene citrate and apilimod pair 206 (STE=0.86). This is due to a higher toxicity (13.30 vs 24.60) of latter, although both of the drug 207 combinations have similar efficacy scores (70.88 vs 68.20). The lower ranking of combinations 208 involving cilchicine and apilimod is in accordance with their observed extreme toxicity in the clinic 209 [19,20]. This case study indicates that SynToxProfiler can identify safe top hits with high selective 210 efficacy and synergy that have increased potential for clinical success, as compared to hits selected 211 based on synergy alone. 212

8 bioRxiv preprint doi: https://doi.org/10.1101/693010; this version posted July 5, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

213 214 Fig. 3. SynToxProfiler penalizes for toxicity of drug pairs while ranking top hits. (A) The efficacy, 215 toxicity, and synergy matrices for the top drug pairs selected based on the highest STE score 216 (clomiphene citrate and sertraline HCL, upper panel) and the highest synergy score (toremifene citrate 217 and apilimod, lower panel). The synergy was calculated using the ZIP model implemented in 218 SynergyFinder. The square with dotted line denotes the 3x3 concentration range with the most 219 synergistic area in the dose-response matrix. 220 221 Discussion and conclusions 222 The primary motivation for the use of synergistic drug combinations in the clinic is to achieve higher 223 efficacy (by means of drug interaction) with reduced toxicity (by decreasing the drug doses). 224 Therefore, the HTS screening aims to discover drug pairs that are more effective than the individual 225 single drugs, and, at the same time, show less toxicity for the patients. Hence, the assessment of 226 synergistic efficacy along with toxicity is critical for the selection of candidate drug pairs for further 227 study, as there exists a fundamental trade-off between clinical efficacy and tolerable toxicity. 228 229 To the best of our knowledge, there are currently no methods to provide the global view in terms of 230 synergy, efficacy and toxicity of drug pairs in an HTS setting. In this respect, SynToxProfiler offers an 231 important advancement into the current practice for drug combination selection, as it provides an easy- 232 to-use platform for in-vitro or ex-vivo assessment of the three critical aspects of drug combinations 233 that are necessary for success in the clinics. Furthermore, SynToxProfiler facilitates the identification

9 bioRxiv preprint doi: https://doi.org/10.1101/693010; this version posted July 5, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

234 of therapeutic window range at which the drugs show highest synergy, high efficacy and lowest 235 toxicity by visualization of the dose-response surfaces. Since, SynToxProfiler uses the normalized 236 volume-based scoring for synergy, efficacy and toxicity levels (see methods and supplementary file), 237 the SynToxProfiler framework can be easily utilized to prioritize synergistic drug combinations with 238 high selective efficacy for multi-component (3 or more drugs) drug combination screening. Since 239 limited number of tools and methodology are available to analyze and interpret either synergy, efficacy 240 or toxicity of multi-component drug combinations, SynToxProfiler will be valuable resource for 241 screening of such combinations. 242 243 In this work, we showed how SynToxProviler prioritized cytarabine-daunorubicin as the top drug pair 244 out of 20 anticancer combinations for T-PLL case study (Table 1), and clomifene-sertraline for anti- 245 viral case study (Additional file 2). The identification of clinically established drug pairs as top hit 246 suggests that ranking based on all the three parameters can help to identify combinations that have 247 more chance to success in the clinic. These effective and safe combinations would have been otherwise 248 missed if combinations were selected merely based on their synergy scores. 249 250 In conclusion, we have developed SynToxProfiler, an interactive tool for top hit prioritization that 251 ranks drug pairs based on their combined synergy, efficacy and toxicity profile, and which can be 252 applied to any HTS drug combination screening project. We showed how this tool enable identification 253 of clinically established drug pairs as top hits and many more drug pairs with a translational potential. 254 We foresee SynToxProfiler will allow for more unbiased and systematic means to evaluate the pre- 255 clinical potency of drug combinations toward safe and effective therapeutic applications. 256 257 Authors' contributions

258 AI and AKG developed and tested the integrated scoring. AI implemented the platform and AKG 259 helped in designing and testing of the platform. ST performed the in-house drug combination screening 260 in T-PLL case study. AI prepared figures for manuscript and finalized with AKG. TA helped in 261 designing of the project and writing of the manuscript. AKG, TA, AI and AK conceptualized the study 262 and wrote the manuscript. All authors have read and approved the final manuscript.

263 Acknowledgements 264 We thank Prof. Satu Mustjoki for her valuable suggestions about the clinical use of SynToxProfiler, 265 Prof. Krister Wennerberg for many discussions regarding synergy, toxicity and efficacy scoring

10 bioRxiv preprint doi: https://doi.org/10.1101/693010; this version posted July 5, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

266 approaches for drug combinations, and Andrea Cremaschi for valuable discussions and suggestions on 267 volume-based combination scoring. 268 269 Conflict of Interest 270 Authors declared no conflict of interest. 271 Availability and requirement 272 The SynToxProfiler web-application is publicly available at https://syntoxprofiler.fimm.fi, together 273 with drug combination example data, user instructions, and the source code. The source code is also 274 available at https://github.com/IanevskiAleksandr/SynToxProfiler .

275 Funding 276 This work was supported by Academy of Finland (grants 292611, 279163, 295504, 310507, 326238), 277 European Union's Horizon 2020 Research and Innovation Programme (ERA PerMed JAKSTAT- 278 TARGET), the Cancer Society of Finland (TA) and the Sigrid Jusélius Foundation (TA). 279 280 References

281 1. Ramón-García S, Ng C, Anderson H, Chao JD, Zheng X, Pfeifer T, Av-Gay Y, Roberge M, 282 Thompson CJ (2011) Synergistic drug combinations for tuberculosis therapy identified by a 283 novel high-throughput screen. Antimicrob Agents Chemother, 55 (8):3861-3869. 284 2. Lopez JS and Banerji U (2017) Combine and conquer: challenges for targeted therapy 285 combinations in early phase trials. Nat Rev Clin Oncol, 14(1): 57-66. 286 3. Palmer AC and Sorger PK (2017) Combination Cancer Therapy Can Confer Benefit via 287 Patient-to-Patient Variability without Drug Additivity or Synergy. Cell 171 (7): 1678-1691. 288 4. Arrowsmith J (2011) Trial watch: Phase II failures: 2008-2010. Nat Rev Drug Discov 10 289 (5):328-329. 290 5. Berenbaum MC (1989) What is synergy. Pharmacol Rev 41 (2):93–141. 291 6. Bliss CI (1939) The toxicity of poisons applied jointly. Ann Appl Biol. 26:585–615. 292 7. Loewe S (1953). The problem of synergism and antagonism of combined drugs. 293 ArzneimiettelForschung 3 (6):286–290. 294 8. Yadav B, Wennerberg K, Aittokallio T, Tang J. (2015) Searching for Drug Synergy in 295 Complex Dose-Response Landscapes Using an Interaction Efficacy Model. Comput Struct 296 Biotechnol J, 13:504-513.

11 bioRxiv preprint doi: https://doi.org/10.1101/693010; this version posted July 5, 2019. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

297 9. Ianevski A, He L, Aittokallio T, Tang J (2017) SynergyFinder: a web application for analyzing 298 drug combination dose-response matrix data. Bioinformatics, 33(15), 2413-2415. 299 10. Yadav B, Pemovska T, Szwajda A, Kulesskiy E, Kontro M, Karjalainen R, et al. (2014). 300 Quantitative scoring of differential drug sensitivity for individually optimized anticancer 301 therapies. Sci Rep. 5(4):5193. 302 11. Andersson EI, Pützer S, Yadav B, Dufva O, Khan S, He L et al (2018) Discovery of novel drug 303 sensitivities in T-PLL by high-throughput ex vivo drug testing and mutation profiling. 304 Leukemia. 32(3):774-787. 305 12. He L, Tang J Andersson EI, Timonen S, Koschmieder S (2018) Patient-Customized Drug 306 Combination Prediction and Testing for T-cell Prolymphocytic Leukemia Patients. Cancer Res, 307 78 (9):2407-2418. 308 13. Gautam P, Karhinen L, Szwajda A, Jha SK, Yadav B (2016) Identification of selective 309 cytotoxic and synthetic lethal drug responses in triple negative breast cancer cells. Mol Cancer, 310 15(1):34.

311 14. Dyall J, Nelson EA, DeWald LE, Guha R, Hart BJ, Zhou H, et al (2018) Identification of 312 Combinations of Approved Drugs With Synergistic Activity Against Ebola Virus in Cell 313 Cultures. J Infect Dis. 218: S672-S678. 314 15. Murphy T and Yee KWL (2017) Cytarabine and daunorubicin for the treatment of acute 315 myeloid leukemia. Expert Opin Pharmacother, 189(16):1765-1780. 316 16. Nazha A, Kantarjian H, Ravandi F, Huang X, Choi S, Garcia-Manero G, et al (2018) 317 Clofarabine, idarubicin, and cytarabine (CIA) as frontline therapy for patients ≤60 years with 318 newly diagnosed acute myeloid leukemia. Am J Hematol. 88(11):961-6. 319 17. Jain N, Keating M, Thompson P, Ferrajoli A, Burger J, Borthakur G, et al (2019). Ibrutinib and 320 Venetoclax for First-Line Treatment of CLL. N Engl J Med. 2019 380(22):2095-2103. 321 18. Johansen LM, DeWald LE, Shoemaker CJ, Hoffstrom BG, Lear-Rooney CM, Stossel A et al 322 (2015) A screen of approved drugs and molecular probes identifies therapeutics with anti-Ebola 323 virus activity. Sci Transl Med. 7(290):290ra89. 324 19. Finkelstein Y, Aks SE, Hutson JR, Juurlink DN, Nguyen P, Dubnov-Raz G, et al (2010) 325 Colchicine poisoning: the dark side of an ancient drug. Clin Toxicol (Phila) 48(5):407-14 326 20. Gayle S, Landrette S, Beeharry N, Conrad C, Hernandez M, et al (2017) Identification of 327 apilimod as a first-in-class PIKfyve kinase inhibitor for treatment of B-cell non-Hodgkin 328 lymphoma. Blood 129(13):1768-1778.

12