Low dose dimethyl sulfoxide driven gross molecular changes have the potential to interfere with various cellular processes

Dimethyl sulfoxide (DMSO) is a small molecule with polar, aprotic and amphiphilic properties. It serves as a solvent for many polar and nonpolar molecules and continues to be one of the most used solvents (vehicle) in medical applications and scientific research. To better understand the cellular effects of DMSO within the concentration range commonly used as a vehicle (0.1–1.5%, v/v) for cellular treatments, we applied Attenuated Total Reflectance (ATR) Fourier Transform Infrared (FT-IR) spectroscopy to DMSO treated and untreated epithelial colon cancer cells. Both unsupervised (Principal Component Analysis-PCA) and supervised (Linear Discriminant Analysis-LDA) pattern recognition/modelling algorithms applied to the IR data revealed total segregation and prominent differences between DMSO treated and untreated cells at whole, lipid and nucleic acid regions. Several of these data were supported by other independent techniques. Further IR data analyses of macromolecular profile indicated comprehensive alterations especially in proteins and nucleic acids. Protein secondary structure analysis showed predominance of β-sheet over α-helix in DMSO treated cells. We also observed for the first time, a reduction in nucleic acid level upon DMSO treatment accompanied by the formation of Z-DNA. Molecular docking and binding free energy studies indicated a stabilization of Z-DNA in the presence of DMSO. This alternate DNA form may be related with the specific actions of DMSO on gene expression, differentiation, and epigenetic alterations. Using analytical tools combined with molecular and cellular biology techniques, our data indicate that even at very low concentrations, DMSO induces a number of changes in all macromolecules, which may affect experimental outcomes where DMSO is used as a solvent.

. PCA is an exploratory and unsupervised pattern recognition method analysing the natural relationship of samples without any bias. Accordingly, the samples are unpredictably sorted in several different categories beyond their primary quality and grouping positions 9 . Figure 2 shows the 3D scores plots for DMSO-treated and untreated (UT) HCT-116, SW-480, and MCF-10A cells in the whole (4000-650 cm −1 ), lipid (3030-2800 cm −1 ) and nucleic acid (1250-1200 cm −1 ) IR spectral regions. Figure 2A,D and G represent clear differentiation of 0.5% and 1.5% DMSO treated groups from each other and from UT groups in whole spectral region for HCT−116, SW-480, and MCF-10A, respectively. A clear segregation was similarly observed in the lipid (Fig. 2B,E,H) and nucleic acid (Fig. 2C,F,I) regions for all three cell lines. The well-segregated scores plots in different spectral regions indicate remarkable dose-dependent molecular alterations induced by DMSO. Supervised Linear Discriminant Analysis (LDA) results in the whole spectral region further confirmed these findings (Supplementary Figs S2-4, for HCT-116, SW-480, and MCF-10A, respectively). As shown in LDA discrimination plots, and prediction and confusion matrices, 0.5% and 1.5% DMSO treated groups could be clearly discriminated from each other and from UT cells with 100% accuracy.
Treatment of cells with different doses of DMSO generated unique fingerprint IR spectra (1800-650 cm −1 ) in HCT-116 cells confirming the occurrence of major molecular alterations ( Supplementary Fig. S5). The spectrum consists of hundred or more bands, and it is not necessary to allocate most of them 10 . To get a better understanding of the qualitative and quantitative alterations occurring in individual macromolecules in the presence of DMSO, we analysed the positions of individual spectral bands and their intensities emerging from different functional groups of macromolecules. Of note, wavenumber and intensity of IR bands correspond to structural alterations and concentration of specific molecule, respectively 10, 11 . DMSO decreases total nucleic acid content and delays cell cycle progression. To determine any alterations in nucleic acid content induced by low doses of DMSO, we analysed the PO 2 antisymmetric band from the absorbance spectrum located in between 1242-1238 cm −1 that is assigned to total nucleic acids 12 . Additionally, the band located at 915 cm −1 and emerging from Ribose ring vibrations of total RNA and DNA 13,14 was analysed from the second derivative spectra. The intensities (%) of the PO 2 antisymmetric band are  Fig. 3A for HCT-116 cells. We observed a dose-dependent decrease in the intensity of this band in the presence of DMSO. This decrease was statistically significant in the 1.0% and 1.5% DMSO-treated cells with respect to the control (UT) cells (Fig. 3A). The intensities (%) of total RNA and DNA bands were statistically significantly reduced in a dose dependent manner in DMSO treated HCT-116 cells at all doses compared to the control cells (Fig. 3B). Similar results were obtained for these bands in SW-480 cells ( Supplementary Fig. S6A). On the other hand, when compared to HCT-116 and SW-480 CRC cells, a more pronounced decrease in the Representative histograms for cell cycle phase distributions of DMSO treated and untreated cells after release, and non-starved (non-synchronized) and starved (synchronized) cells are given. Percentages of cell cycle phases are given as mean ± SEM of three independent biological replicates each with two technical replicates. Statistical significance was calculated with respect to UT cells by using t test. (D) After 7 h release following 24 h starvation, expression levels of cell cycle proteins associated with G1 to S phase transition were examined in DMSO treated or untreated cells by western blot. β-Actin was used as loading control for normalization. Changes in the expression levels are given as fold change with respect to untreated cells.
Cells incubated with DMSO in the range of 1.0-2.0%, v/v have been reported to exhibit cellular differentiation accompanied by growth inhibition and arrest at the G1 phase of the cell cycle [15][16][17][18] . Since the PO 2 antisymmetric and Ribose ring vibrations of total RNA and DNA band intensities revealed a decrease in nucleic acid content in DMSO treated cells, we investigated the effect of DMSO on cell cycle progression. We observed that even at the low dose of 0.5% (v/v), DMSO treatment resulted in the accumulation of cells at the G1 phase of the cell cycle ( Supplementary Fig. S7A). This effect of DMSO was more prominent when the cells were synchronized through serum starvation and released. 7 h after release, the percentages of the cells at the G1/S boundary of the cell cycle in DMSO treated cells was higher (Fig. 3C), along with an accumulation of p21, and decreased expression of Cyclin E, Cyclin D and Cyclin-Dependent Kinase 4 (CDK4). Additionally, phosphorylation of Retinoblastoma (Rb) was lower in DMSO treated cells compared to control cells (Fig. 3D). Similar results were also observed in SW-480 cells ( Supplementary Fig. S7B).
Although we did not address the mechanism through which DMSO treatment delayed the cell cycle progression in HCT-116 and SW-480 cells, we suggest that DMSO mediated decrease in cellular ROS may be responsible for the observed effect. This stems from the fact that mitogenic pathways can be stimulated in the presence of ROS; therefore, the entry of cells into the S phase may be regulated by controlling the activity of CDKs and phosphorylation of Rb protein 19 . Thus, it can be concluded that the growth inhibitory effect observed in DMSO treated HCT-116 and SW-480 cells (Fig. 1A) may have resulted from delayed progression through cell cycle phases.
DMSO alters DNA topology. The sub-bands of PO 2 antisymmetric band were analysed in detail from the second-derivative IR spectra. These are highly sensitive and marker bands that are assigned to the main B and main A conformations of DNA 12 . The absorbance of the main B and A-forms of DNA are located at 1221 cm −1 and 1240 cm −1 , respectively 20 . Particularly, we analysed the band located at 1066 cm −1 that is strongly enhanced specifically in the presence of Z-form DNA 21  Information about the relative concentration of a specific biomolecule in a system can be achieved by analysing the alterations in band intensity values 22 . When the ratio of the enhanced Z-DNA band intensity to the sum of intensities of A-DNA and B-DNA forms was calculated, we found that the content of Z-DNA increased significantly with DMSO treatment in these cells (Fig. 4B,E,H).
In vivo, DNA is most frequently found as the canonical duplex right-handed helical structure known as B-DNA 23 . However, DNA is able to fit different conformations other than the familiar right-handed B-DNA double helix. Z-DNA is left-handed form of the helix in which all bases are in anti-conformation. The phosphates in the backbone follow a zigzag path; hence, it is called as Z-DNA 24 . Compared to B-DNA, Z-DNA is not favoured thermodynamically. Although Z-DNA was shown to be formed in purine-pyrimidine repeats (d(CG)n) as a result of decreased electrostatic repulsion under high salt concentrations 25 , more recently, anion species were also shown to play a critical role in the formation and stabilization of Z-DNA 26 . Moreover, negative supercoiling and chemical base modifications were claimed to stabilize Z-DNA [27][28][29] . To determine whether the binding of DMSO to Z-form of DNA versus B-form of DNA was energetically more favourable, docking reactions were carried using AutoDock. 30 docking files were obtained for each case, and the positions with the lowest energy are shown in Fig. 5. In both Z-DNA and B-DNA, DMSO interacted with the amine (NH 2 ) group of guanine base via hydrogen bond between DMSO's oxygen moiety and one of the hydrogens of NH 2 (Fig. 5A). Another important detail is that DMSO was bound to the minor groove of B-DNA (Fig. 5B), probably due to the orientation of guanine's amine group always facing the minor groove. Binding energy for DMSO docked to Z-DNA was −2.57 Kcal/mol, while that for B-DNA was −2.48 Kcal/mol, giving a difference of −0.09 Kcal/mol, which in physiological conditions would require an increase in temperature of 45 °C to occur spontaneously. This indicates that DMSO binds to Z-DNA more strongly than to B-DNA.
When the space occupied by DMSO in Z-DNA was analysed in detail, it was found that Z-DNA could hold a second DMSO molecule parallel to the first. As shown in Fig. 5C, the two sides of the structure were perfect mirror images of each-other. However, it was not the same for the B-DNA conformation as two molecules could not be fitted in the structure. To determine whether the binding of two DMSO molecules to the DNA was energetically favourable, a mirror image of the docked DMSO on the complementary strand of Z-DNA was created and energy calculations were carried out. Since AutoDock was unable to dock more than one DMSO molecule to the DNA at a time, DFT calculations for the relative energies of none, one, and two-DMSO bound Z-DNA structures were carried out using B3LYP/6-31 G(d) in Gaussian. The results showed that binding of a single DMSO to a Z-DNA stabilizes the system by ca. 50 Kcal/mol, and binding of two DMSO molecules to the same Z-DNA stabilizes the system by ca. 90 Kcal/mol compared to the energy of naked Z-DNA (Z-DNA with two non-interacting DMSO positioned far away). These energy differences are remarkably high and cannot be overcome at the normal physiological temperature. ATR-FTIR data indicated an increase in the enhanced Z-form of DNA in the presence of DMSO (Fig. 4); we think that this increase could have resulted from the DMSO induced stabilization of Z-DNA. The results demonstrate that Z-DNA bound to two DMSO molecules was the most stable structure, followed by Z-DNA bound to one DMSO and finally naked Z-DNA, which is known to be very unstable in physiologic conditions 30 .
Z-DNA formation in cells rely on sequence composition together with chromatin structure 31 . Besides, a variety of environmental factors such as a high degree of hydration, high ionic strengths, transition-metal complexes and spermine are known to stabilize the Z-form 32 . Additionally, negative supercoiling, Z-DNA binding proteins, bromination and cytosine methylation are among the biological factors to stabilize Z-DNA in cells [33][34][35] . DMSO Scientific RepoRTS | (2018) 8:14828 | DOI:10.1038/s41598-018-33234-z was shown to destabilize DNA double helix structure at low solvent concentrations by forming hydrogen bonds with nucleic acids 36 . This property is exploited to enhance PCR amplification reactions by inhibiting secondary structures in the DNA template or primers, especially during the synthesis of GC-rich gene fragments 37 . Prior studies showed that DMSO can denature duplex DNA because of its hydrophobic character 38 . While DMSO concentrations used in this study are not adequate to disrupt the duplex structure 38,39 , it may promote local conformational changes in DNA, as suggested by Kashino et al. 40 . By forming hydrogen bonds with DNA bases, DMSO may potentially help alter the helix conformation towards Z-form, in agreement with very early findings, which suggest that some regions of covalently-closed-circular DNA become more loose or more negatively supercoiled by the addition of 5.0% (v/v) DMSO 39 . Therefore, initiation and elongation of RNA synthesis may be enhanced at some locally relaxed regions of DNA 41,42 . In fact, there is a strong correlation between Z-DNA formation and active transcription 31 . There are several reports demonstrating that even at very low concentrations (0.05-2.5%, v/v), DMSO can affect gene expression [43][44][45][46][47][48][49][50] and induces differentiation of many cell types including leukemia cells, mesenchymal stem cells, embryonic carcinoma cells, and preosteoblasts 5,51-53 . DMSO is also known to epigenetically alter DNA methylation profile [54][55][56] . Addition of DMSO (0.02-1.0%, v/v) to mouse embryoid bodies (EBs) affected genome-wide DNA methylation status, by changing hypermethylation as well as hypomethylation of genes and intergenic regions in EBs. Methylation-induced Z-DNA formation and stabilization is a known phenomenon 23,[57][58][59] . Therefore, in the current study we suggest that gene expression changes and epigenetic modulations attributed to DMSO may be, at least in part, explained by the formation of Z-DNA. This in turn may affect In summary, our experimental results, supported by in silico analyses unambiguously demonstrate that DMSO treatment significantly affected DNA conformation. To the best of our knowledge, we have described here for the first time that low concentrations of DMSO can change DNA helix topology to the Z-form. Although further advanced structural and biological studies are needed to understand the dynamic process of this phenomenon, this important finding emphasizes how the use of DMSO as a solvent may influence experimental results by interfering with gene expression.
Low DMSO concentration can influence protein secondary structure. To determine DMSO induced changes in the secondary structures of proteins, the second-derivative IR spectra of main secondary structural components in Amide I region (1680-1600 cm −1 ), considered as main protein region [61][62][63] , were examined. The two main α-helical structures located at 1657 and 1649 cm −1 and the two main β-sheet structures located at 1636 and 1629 cm −1 11,64 are shown in Fig. 6A.
Treatment of HCT-116 cells with DMSO resulted in a dose dependent change in the conformation of proteins from α-helical to a β-sheet state starting at doses as low as 0.1% (v/v). We found that 0.1% DMSO treatment converted nearly 15% of α-helices to β-sheet conformation while treatment with 0.5% and 1.5% (v/v) resulted in conversion rates of 30% and 40%, respectively (Fig. 6B). Similar results were observed for SW-480 cells ( Supplementary Fig. S8). Protein aggregates generally exhibit enhanced β-sheet content and diminished α-helicity compared to the native state, probably due to intermolecular contacts 65 . The formation of intermolecular β-sheets appears to be quite extensive in DMSO treated cells 66 . This calls for caution and awareness and the necessity of ensuring the physiochemical integrity as well as the biological activity of the analysed protein, since DMSO has the potential to cause protein denaturation, aggregation, or degradation, and even change the binding properties of the proteins 67 .
To the extent of our knowledge, this is the first study in which the impact of low doses of DMSO treatment on cellular proteins has been demonstrated. A large body of studies has consistently reported that DMSO can influence the properties of proteins in various ways. However, most of these studies were performed at very high DMSO concentrations (10-100%, v/v) with purified proteins 66,68-70 . Tjernberg et al. showed the effects of lower amounts of DMSO (0.1-3%, v/v) on two protein systems. These authors reported that DMSO could change the biochemical and binding properties of proteins in solution 67 . Functional and structural stability of proteins require a fine balance between the favourable internal interactions among residues of the protein and with solvent molecules 68 . It is suggested that DMSO destabilizes protein structure by perturbing the structure of water around the protein molecule which leads to DMSO induced protein unfolding 66 13 . The intensity (%) of this band was dose dependently reduced in DMSO treated cells compared to the untreated control (UT) cells. However, we did not observe any differences in wavenumber values between untreated and DMSO treated samples indicating that DMSO does not alter molecular structure of cholesterol (Fig. 7A,B).
The thickness of membranes is related to their cholesterol and cholesteryl esters content, with thinner membranes containing less cholesterol 72 . Decreased membrane thickness with increased doses of DMSO may enhance membrane permeability and facilitate the transport of active molecules 73 . Thus, membrane associated effects attributed to DMSO, like enhancement of permeability, facilitation of membrane fusion, and enabling the cell membrane to accommodate osmotic and mechanical stresses during cryopreservation 74 , may be related to its ability to change the lipid composition and dynamics of biological membranes.
The reaction of free radicals with the olefinic bonds of polyunsaturated lipids results in lipid peroxidation, which in turn causes the loss of unsaturation in the lipid structure together with the breakdown of longer chains into smaller lipid acyl chains 14 . The olefinic band located at 3009 cm −1 in the lipid region of IR spectra assigned to unsaturated lipids was analysed to measure the level of unsaturation and lipid peroxidation end-products as a result of DMSO treatment. An increase in intensity of this unsaturation marker band 13 is related to elevated concentrations of unsaturated acyl chains of lipids and loss of lipid-peroxidation end-products [75][76][77] . Specifically, intensities (%) of olefinic band and the ratio of olefinic to total lipids (olefinic/total lipid) were analysed to monitor the acyl chain unsaturation level of the phospholipids 14 (Fig. 7C,D). In this, the sum of the intensities (%) of olefinic (3009 cm −1 ), CH 2 antisymmetric (2921 cm −1 ), CH 2 symmetric (2852 cm −1 ) and cholesteryl ester (1172 cm −1 ) bands were calculated as total lipids. The intensities (%) of olefinic band and olefinic/total lipid ratios (%) increased statistically significantly in 0.1% and 0.5% DMSO treated HCT-116 cells; however, higher DMSO doses reverted the effect (Fig. 7C). This may be caused by a negative feedback mechanism to maintain membrane phospholipid composition 78,79 . On the other hand, the intensities (%) of olefinic band and olefinic/total lipid ratios (%) were significantly increased in 1.0% and 1.5% DMSO treated SW-480 cells (Fig. 7D). It is possible that a similar reversion may occur in SW-480 at even higher doses of DMSO. Of note, the olefinic intensity (%) was similar to the olefininc/total lipid ratio, indicating that rather than inducing any alteration in lipid biosynthesis or degradation, the effect was on the peroxidation of existing lipid molecules. Thus, while low doses of DMSO were quite enough to elevate the unsaturation level and decrease lipid peroxidation end-products in HCT-116, higher doses of DMSO were needed for this purpose in SW-480 cells.
The relative susceptibility of individual fatty acids to peroxidation depends on the membrane fatty acid composition 80 . Tapiero et al. reported that the effects of DMSO on membranes were related to the structure and/or composition of the cell membranes that are different in various cell types 81 . Therefore, the different oxidative effects of DMSO in HCT-116 and SW-480 cells may be related to distinct membrane lipid characteristics of these cells. To demonstrate these differences, we ran unsupervised PCA and supervised LDA to compare HCT-116 and SW-480 cells, in lipid (3030-2830 cm −1 ) (Fig. 7E) and whole (4000-650 cm −1 ) IR regions (Fig. 7F). The PCA results are given in the left and the LDA results are presented in the right panel of Fig. 7. PCA 3D scores plots (Fig. 7E,F; left panel) revealed a clear-cut segregation between HCT-116 and SW-480 cells, confirming our hypothesis of inherent differences between these cell lines. LDA discrimination plots, and prediction and confusion matrices (Fig. 7E,F; right panel) similarly revealed a well-defined discrimination between HCT-116 and SW-480 cells with 100% accuracy. Moreover, the obvious spectral differences between these two cell types were prominent in the lipid (4000-2800 cm −1 ) and fingerprint (1800-650 cm −1 ) IR spectra as demonstrated in Supplementary Fig. S9A,B, respectively. To further examine whether discrimination between SW-480 and HCT-116 cells with such high accuracy was due to differential gene expression related to lipid metabolism, we examined a publicly available microarray dataset (GSE41445) that reported gene expression of wild type HCT-116 and SW-480 cells. We observed that nearly double the number of genes (61 genes in SW-480 cells versus 37 genes in HCT-116 cells) related to the gene ontology (GO) term LIPID_METABOLISM were statistically significantly (p < 0.05) differentially expressed with fold change of 1 between the two cell lines (Supplementary Fig. S10). Thus the effects of DMSO on the lipid component of the two cell lines examined may have resulted from genotypic differences between the cell lines.

Conclusion
DMSO is a widely used organic solvent for biological applications because of its capability to dissolve numerous polar or nonpolar compounds. In high throughput screening assays for drug discovery, most small molecule compounds, including potential anti-cancer agents, are dissolved in DMSO 82,83 . For the first time, in the present study the effects of low, non-toxic doses of DMSO on colon cancer epithelial cells, at concentrations that are widely used in biological research, were investigated in detail by Infrared (IR) spectroscopy coupled with pattern recognition techniques and confirmed by molecular biology techniques and in silico analyses. We observed The observed changes in gross macromolecular structures (Fig. 8) can address many of the applications of DMSO in cell biology as well highlight some of the pitfalls of using DMSO as a solvent. Thus, it is the hope of the authors that this report will bring researcher's attention to exhibit caution when interpreting data and conclusions in experiments where DMSO is used as a solvent, since even low concentrations of DMSO may interfere with important cellular processes as demonstrated in this study.
IR spectroscopy measurements and data analyses. The cell culture medium was aspirated and the wells washed with cell culture grade Phosphate-buffered saline (PBS). Cells were then detached using Trypsin-EDTA (0.25%) solution. The trypsin was neutralized by adding fresh medium. The cells were centrifuged, the pellet was washed in PBS, followed by centrifugation at 400 × g for 7 min, to remove the PBS. The cells were resuspended in PBS, counted with a haemocytometer and adjusted to 5 × 10 5 cells/10 µL of PBS for ATR-FTIR.
The IR spectra of untreated and DMSO treated cells were obtained using Spectrum 100 FTIR spectrometer (PerkinElmer, Waltham, Massachusetts, USA) equipped with a Universal ATR accessory. The spectrum of air was used as a reference. 10 µL of sample, corresponding to 5 × 10 5 cells in PBS, was placed on a diamond/ZnSe crystal plate (PerkinElmer) and dried with a mild nitrogen flux for 2 min. Purging samples with non-invasive N 2 was applied in order to remove the overlapping free water bands from the samples (while keeping the bound water in the system), a common procedure in ATR-FTIR studies 84 . PBS, which was used during the sample preparation step, was scanned under identical experimental conditions as the samples. These spectra were manually subtracted from the sample spectra using the Spectrum 100 software (PerkinElmer, Waltham, Massachusetts, USA). During the subtraction process, the free water band located around 2125 cm −1 was flattened using the same software in order to eliminate the effect of PBS. The difference spectra obtained were used for all further spectral analyses 11 . For HCT-116 and SW-480 cells, three biological replicates each with three technical replicates, and for MCF-10A cells, two biological replicates each with four technical replicates were analysed. The samples were scanned over the spectral range 4000 to 650 cm −1 , at room temperature, with a resolution of 4 cm −1 and spectra were collected as an average of 64 scans. Pre-processing and/or standardization of FT-IR data are essential for accurate interpretation. Vector normalization and mean-centering by using second derivative spectra in the full and/or restricted fingerprint regions (ideal for quantitative analyses) is commonly used and is considered to be a reliable method for pre-processing of FT-IR data. In vector normalization, each spectrum is divided by the Euclidean norm and is usually applied to second derivative spectra because they contain both positive and negative values (below zero). Min/max normalization is another method in which the standardization is carried out on a single and stable peak (Amide I and II). Compared with vector normalization, min/max normalization method is often used for visual and qualitative purposes, since it can only process spectra with positive values 85 .
During data analyses, the second derivative and vector-normalized IR spectra were used in order to increase the accuracy of band positions. However, for the determination of absolute intensities of Amide bands (proteins) and PO 2 antisymmetric and symmetric bands (total nucleic acids), vector-normalized absorbance IR spectra were used. Opus 5.5 software (Bruker, Billerica, Massachusetts, USA) was utilized for all data pre-processing. IR analysis of the absolute band intensities of untreated groups were normalized to 100%, and DMSO treated groups were calculated accordingly and expressed as percent intensity.

Principal component analysis (PCA).
PCA is an unsupervised modelling technique that transforms a group of equivalent variables into a smaller set of distinct variables called principal components (PCs), while preserving the vast majority of actual information. Thus, a reduced size PC model can be employed to rapidly identify and differentiate abnormalities/deviations in the original system 86 . There are different criteria for selecting the number PCs, since there is no universally accepted and unconditional methodology 87 . Higher PCs (PC1 and PC2) describe the predominant sources of variations in the dataset and generally explain the gross spectral changes that may mask smaller ones 20,88 . This strategy is preferable when performing robust analysis of the bulk data. Another strategy is the analysis of subsequent minor PCs that should be considered for the detection of much smaller but more distinctive and itemized alterations occurring in macromolecules 89 .
PCA was applied to ATR-FTIR spectral data in order to perform an exploratory analysis of the studied experimental groups. For this method, unit vector-normalized and mean-centered spectra of each group were imported into The Unscrambler ® X 10.3 (CAMO Software AS, Norway) multivariate analysis (MVA) software. PCA was applied in the 4000-650 cm −1 , 3030-2830 cm −1 and 1250-1200 cm −1 IR regions using Nonlinear Iterative Partial Least Squares (NIPALS) algorithm and the results were presented as three-dimensional (3D) scores plots. Hotelling's T2 statistics were used in scores plots to monitor major variations within the PCA model.

Linear discriminant analysis (LDA). LDA is a supervised modelling technique that involves linearly trans-
forming n-dimensional feature samples into an m-dimensional space (m < n). In this way, members of the same class can be categorized together while members of different classes cluster separately, maximizing the differences between the predefined classes, with respect to the new variable 90 .
LDA was performed using The Unscrambler ® X 10.3 (CAMO Software AS, Norway) multivariate analysis (MVA) software. The modelling was based on PC scores. The category variable column was included into a data matrix and subsequently, randomly chosen spectra among the different sample categories was used to build a training set. After defining the data for modelling, Mahalanobis and/or Linear classifiers methods were used for LDA. The results are presented as discrimination plots, as well as prediction and confusion matrices.

Western blot analysis. Cells were lysed in M-PER Mammalian Protein Extraction Reagent (Thermo Fisher
Scientific, Waltham, Massachusetts, USA) containing protease inhibitor cocktail (Roche, Switzerland) and phosphatase inhibitor (Roche, Switzerland) according to manufacturer's instructions. Proteins (20-50 µg) were separated in 10% SDS-PAGE gels at 100 V and transferred to polyvinylidene difluoride (PVDF) membranes using standard techniques. After 1 h of blocking in 5% skim-milk in TBS-T (Tris-buffered saline, 0.1% Tween 20), the membranes were incubated in primary antibodies (Supplementary Table S1) overnight at 4 °C, followed by 1 h incubation with HRP-conjugated secondary antibody at room temperature. Bands were visualized by using Clarity ECL Substrate (Bio-Rad, Hercules, California, USA) in a ChemiDoc MP Imaging System (Bio-Rad).

MTT assay.
To evaluate the effect of DMSO on cell viability, Vybrant ® MTT (3-(4,5-dimethylthiazol-2-yl)-2, 5-diphenyltetrazolium bromide) Cell Proliferation Assay (Thermo Fisher Scientific) was used according to the manufacturer's instructions. Briefly, cells were seeded as 5 × 10 3 cells/well of a 96-well plate and allowed to attach overnight. The day after, the cells were incubated with different doses of DMSO (0.1-1.5%) or left untreated. After 24 h of incubation with DMSO, the medium was replaced with 100 µL of complete culture medium containing 1.2 mM of MTT. Following 4 h of incubation at 37 °C, 100 µL of the SDS-HCl solution (1 g of SDS in 10 mL of 0.01 M HCl) was added to each well and mixed thoroughly by pipetting. The plates were incubated at 37 °C for 16 h in a humidified atmosphere at 37 °C. The absorbances were measured at 570 nm in a microplate reader (Thermo Fisher Scientific), and relative cell viability (%) was expressed as a percentage relative to the untreated control cells.
Cell cycle and DNA content analysis. To determine the effect of DMSO on cell cycle distribution, cells were treated with indicated DMSO doses for indicated times. The cells were synchronized at the G1 phase by incubating the cells in serum free medium. After 24 h, the serum free medium was removed, and the cells were incubated with complete growth medium containing 10% FBS for 7 h to release from G1.
For cell cycle analysis, DMSO treated and untreated cells were detached in trypsin/EDTA, and washed once in ice cold PBS (Phosphate-buffered saline), before fixation in 70% cold ethanol. The fixed cells were washed with PBS two times, resuspended in a staining solution consisting of 0.1% Triton X-100, 20 μg/mL RNase A (Thermo Fisher Scientific) and 20 µg/mL Propidium Iodide (Sigma-Aldrich Chemie GmbH, Germany) in PBS Scientific RepoRTS | (2018) 8:14828 | DOI:10.1038/s41598-018-33234-z and incubated for 30 min in the dark at room temperature. The percentage of cells at different stages of cell cycle was determined by using Accuri C6 Flow Cytometer (BD Biosciences, Franklin Lakes, New Jersey, USA) in the FL-3 channel.

Dihydroethidium (DHE) assay. To measure intracellular ROS levels in DMSO treated and untreated cells,
HCT-116 and SW-480 cells were incubated with indicated doses of DMSO for 24 or 48 h, detached by trypsinization and collected by centrifugation at 400 × g for 7 min. After washing in PBS, cells were incubated in DHE (Thermo Fisher Scientific) staining solution (3.2 μM in PBS) for 20 min at 37 °C, and immediately analysed by Accuri C6 Flow Cytometer (BD Biosciences) in the FL-2 channel.
Auto dock. In order to assess the interactions between DNA and DMSO, molecular docking analyses were carried out with AutoDock4.2 software 91 . To determine whether the binding affinities of DMSO to Z-DNA versus B-DNA were different, high resolution (1.6 Å for B-DNA and 1.0 Å for Z-DNA) X-ray structures containing a complete helical turn with the sequences TACGCCCACGC (11-mer) for B-DNA (PDB ID 1AAY) 92 and CGCGCG (6-mer) for Z-DNA (PDB ID 1DCG) 30 were chosen from the PDB repository. The coordinates of DMSO were downloaded in.sdf format from PubChem Public database (PubChem CID 679) and converted into pdbqt format by OpenBabel2.4.1 software 93 . Docking was performed as described previously 94,95 . Briefly, protein components, ions and water were removed from the PDB DNA structures. The chosen parameters were as follows: in the gridding step, the entire sequence (6 mer, CGCGCG) of the Z-DNA was included whereas for the B-DNA, only the GC-rich region (CGCCC) was included. The choice of sequence of B-DNA was based on preliminary optimizations by starting with the whole molecule and then narrowing down to this region for better comparison with the Z-DNA sequence. For docking, the Genetic Algorithm was used for searching with 30 runs and Lamarkian GA to write the output.dpf file. The parameter chosen for best position was "Binding Energy" and the values calculated from empirical free energy scoring function for all 30 structures are shown in Supplementary Table S2. The images were generated with UCSF Chimera software version 1.11.2 96 . Gaussian. The most stable Z-DNA-DMSO complex, determined from molecular docking analysis, was used for further structural analysis to determine the stability of two DMSO molecules docked in the Z-DNA structure relative to one molecule of DMSO and none. DFT computational results were obtained by using B3LYP/6-31 G(d) functional/basis set combination in Gaussian 09, Revision A.02 software package 97 at TÜBİTAK ULAKBİM, High Performance and Grid Computing Centre (TRUBA) and visualization of the output files was carried out using Chemcraft software (https://www.chemcraftprog.com). The initial structure of Z-DNA with two DMSO molecules was obtained from a stable Z-DNA-DMSO complex by adding a second DMSO molecule to the opposite strand mirroring the coordinates of the first. Two and one extra DMSO for none and one-DMSO cases, respectively, were added to the calculations and positioned far away from the Z-DNA in order to ensure no interaction between the Z-DNA and DMSO. That way equal number of atoms was achieved during the calculations for the three cases in order to make the resulted energies comparable. Prior to energy calculations the structures were optimized with PM3. During the optimization, the Z-DNA structure was frozen while DMSO molecules were left free to move to obtain energetically the most stable structure.
Bioinformatic data analysis. Raw data for GSE41445 was downloaded using GEOquery package from bioconductor 98 . The data was normalized using gcrma 99 and annotated using hgu133plus2.db 100 packages, respectively. Differential gene expression analysis for HCT-116 and SW-480 were performed using limma 101 package and the genes related to lipid metabolism (GO:0006629) were extracted using stringr 102 package based on the Affymetrix annotation file. All analyses were performed on R Language Software version 3.5.1 103 .

Data Availability
All data generated during and/or analysed during the current study are available from the corresponding author on reasonable request. The full length western blot images are shown in Supplementary Fig. S11.