Differences between intrinsic and acquired nucleoside analogue resistance in acute myeloid leukaemia cells

SAMHD1 mediates resistance to anti-cancer nucleoside analogues, including cytarabine, decitabine, and nelarabine that are commonly used for the treatment of leukaemia, through cleavage of their triphosphorylated forms. Hence, SAMHD1 inhibitors are promising candidates for the sensitisation of leukaemia cells to nucleoside analogue-based therapy. Here, we investigated the effects of the cytosine analogue CNDAC, which has been proposed to be a SAMHD1 inhibitor, in the context of SAMHD1. CNDAC was tested in 13 acute myeloid leukaemia (AML) cell lines, in 26 acute lymphoblastic leukaemia (ALL) cell lines, ten AML sublines adapted to various antileukaemic drugs, 24 single cell-derived clonal AML sublines, and primary leukaemic blasts from 24 AML patients. Moreover, 24 CNDAC-resistant sublines of the AML cell lines HL-60 and PL-21 were established. The SAMHD1 gene was disrupted using CRISPR/Cas9 and SAMHD1 depleted using RNAi, and the viral Vpx protein. Forced DCK expression was achieved by lentiviral transduction. SAMHD1 promoter methylation was determined by PCR after treatment of genomic DNA with the methylation-sensitive HpaII endonuclease. Nucleoside (analogue) triphosphate levels were determined by LC-MS/MS. CNDAC interaction with SAMHD1 was analysed by an enzymatic assay and by crystallisation. Although the cytosine analogue CNDAC was anticipated to inhibit SAMHD1, SAMHD1 mediated intrinsic CNDAC resistance in leukaemia cells. Accordingly, SAMHD1 depletion increased CNDAC triphosphate (CNDAC-TP) levels and CNDAC toxicity. Enzymatic assays and crystallisation studies confirmed CNDAC-TP to be a SAMHD1 substrate. In 24 CNDAC-adapted acute myeloid leukaemia (AML) sublines, resistance was driven by DCK (catalyses initial nucleoside phosphorylation) loss. CNDAC-adapted sublines displayed cross-resistance only to other DCK substrates (e.g. cytarabine, decitabine). Cell lines adapted to drugs not affected by DCK or SAMHD1 remained CNDAC sensitive. In cytarabine-adapted AML cells, increased SAMHD1 and reduced DCK levels contributed to cytarabine and CNDAC resistance. Intrinsic and acquired resistance to CNDAC and related nucleoside analogues are driven by different mechanisms. The lack of cross-resistance between SAMHD1/ DCK substrates and non-substrates provides scope for next-line therapies after treatment failure.

Conclusion: Intrinsic and acquired resistance to CNDAC and related nucleoside analogues are driven by different mechanisms. The lack of cross-resistance between SAMHD1/ DCK substrates and non-substrates provides scope for next-line therapies after treatment failure.
Keywords: Leukemia, Acute myeloid leukemia, Acute lymphoblastic leukemia, CNDAC, Sapacitabine, SAMHD1, DCK, Intrinsic resistance, Acquired resistance Background Drug resistance is a main obstacle in the successful treatment of cancer [4,9,31]. Resistance can be either intrinsic or acquired. Intrinsic resistance means that a therapynaïve cancer does not respond to treatment right from the start. In acquired resistance, there is an initial therapy response, but resistance develops over time [31,40].
Intrinsic and acquired resistance are conceptually different. Intrinsic resistance is a collateral event during carcinogenesis not influenced by treatment. In contrast, acquired resistance is the consequence of a directed evolution driven by therapy. In agreement, discrepancies have been detected between drug resistance mechanisms in the intrinsic and the acquired resistance setting [31,36,40,44].
Clonal sublines were generated by limiting dilution. Cells were plated at a density of 1 cell per well on a 96well plate and grown for 1-2 weeks. Wells with only one visible cell colony were identified and the respective clones were expanded. SAMHD1-deficient THP-1 (THP-1 KO) cells and control cells (THP-1 CTRL) were generated using CRISPR/ Cas9 approach as previously described [36,41,47]. THP-1 cells were plated at a density of 2 × 10 5 cells/ mL. After 24 h, 2.5 × 10 6 cells were suspended in 250 μl Opti-MEM, mixed with 5 μg CRISPR/Cas plasmid DNA, and electroporated in a 4-mm cuvette using an exponential pulse at 250 V and 950 mF in a Gene Pulser electroporation device (Bio-Rad Laboratories, Feldkirchen, Germany). We used a plasmid encoding a CMV-mCherry-Cas9 expression cassette and a human SAMHD1 gene specific gRNA driven by the U6 promoter. An early coding exon of the SAMHD1 gene was targeted using the following gRNA construct: 5′-CGGAAGGGGTGTTTGAGGGG-3′.
Cells were allowed to recover for 2 days in 6-well plates filled with 4 ml medium per well before being FACS sorted for mCherry-expression on a BD FACS Aria III (BD Biosciences, Heidelberg, Germany). For subsequent limiting dilution cloning, cells were plated at a density of 5, 10, or 20 cells per well of nine round-bottom 96-well plates and grown for 2 weeks. Plates were scanned for absorption at 600 nm and growing clones were identified using custom software and picked and duplicated by a Biomek FXp (Beckman Coulter, Krefeld, Germany) liquid handling system. DCK-expressing MV4-11rAra-C 2μg/ml and MOLM-13rAra-C 2μg/ml cells were established by lentiviral transduction and designated as MV4-11rAra-C 2μg/ml -pWPI+ DCK and MOLM-13rAra-C 2μg/ml -pWPI+DCK (or MV4-11rAra-C 2μg/ml -pWPI and MOLM-13rAra-C 2μg/ml -pWPI for control cells transduced with the empty vector). To generate the pWPI+DCK plasmid, the dCK gene was PCR-amplified from pDNR-Dual_dCK (DNAsu HsCD00000962) using Pfu DNA polymerase (Promega, Germany) and gene-specific primers (Eurofins Genomics, Germany) and subcloned into pWPI IRES puro via BamHI/SpeI. The plasmid was verified by Sanger sequencing (Eurofins Genomics, Germany). For the generation of lentiviral vectors 293 T cells were cotransfected with pWPI+DCK (or pWPI as control), Addgene packaging plasmid pPAX, an envelope plasmid encoding VSV-G and pAdVAntage (Promega). Four days after transfection, lentiviral vectors were harvested and concentrated by ultracentrifugation. For lentiviral transduction MV4-11rAra-C 2μg/ml and MOLM-13rAra-C 2μg/ml cells were seeded at 5 × 10 5 cells/ well of a 96well-plate and spinoculated with the lentiviral vectors. 24 h after transduction, successfully transduced cells were selected with 3 μg/ml puromycin (Sigma-Aldrich) and DCK expression was monitored by Western Blot.

Primary AML samples
Peripheral blood or bone marrow samples derived from AML patients between 2018 and 2020 were obtained from the UCT Biobank of the University Hospital Frankfurt. The use of peripheral blood and bone marrow aspirates was approved by the Ethics Committee of Frankfurt University Hospital (approval no. SHN-03-2017). All patients gave informed consent to the collection of samples and to the scientific analysis of their data and of biomaterial obtained for diagnostic purposes according to the Declaration of Helsinki.

Viability assay
The viability of AML and ALL cell lines treated with various drug concentrations was determined by 3-(4,5dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay modified after Mosman [32], as previously described [37]. Cells suspended in 100 μL cell culture medium were plated per well in 96-well plates and incubated in the presence of various drug concentrations for 96 h. Then, 25 μL of MTT solution (2 mg/mL (w/v) in PBS) were added per well, and the plates were incubated at 37°C for an additional 4 h. After this, the cells were lysed using 100 μL of a buffer containing 20% (w/v) sodium dodecylsulfate in 50% (v/v) N,N-dimethylformamide with the pH adjusted to 4.7 at 37°C for 4 h. Absorbance was determined at 570 nm for each well using a 96-well multiscanner (Tecan Spark, Tecan, Crailsheim, Germany). After subtracting of the background absorption, the results are expressed as percentage viability relative to control cultures which received no drug. Drug concentrations that inhibited cell viability by 50% (IC 50 ) were determined using CalcuSyn (Biosoft, Cambridge, UK) or GraphPad Prism (San Diego, CA, USA).
For AML blasts viability assays were performed using the CellTiter-Glo (Promega, Walldorf, Germany) assay according to the manufacturer's protocol. Briefly, cells were seeded at 5000 cells per well in 96-well plates and treated for 96 h. Luminescence was measured on a Tecan Spark (Tecan). IC 50 values were calculated using GraphPad Prism.

Caspase 3/7 assay
To determine Caspase 3/7 activity in THP-1 SAMHD1 KO and CTRL cells the Caspase-Glo 3/7 assay (Promega, Walldorf, Germany) was used according to the manufacturer's protocol. Briefly, cells were seeded at 5000 cells per well in white 96-well plates, treated with different concentrations of CNDAC and incubated for 24, 48 and 72 h at 37°C in a humidified 5% CO 2 incubator. After incubation an equal volume of Caspase-Glo 3/ 7 reagent was added, mixed for 30 min and luminescence was measured on a Tecan Spark (Tecan).

Determination of population doubling time (PDT)
To generate a growth curve, cells were seeded at 2000 cells per well in a white 96-well plate in 100 μl culture medium and incubated for 0, 1, 2, 3, 4 and 7 days at 37°C in a humidified 5% CO 2 incubator. Cell viability was detected using the CellTiter-Glo assay (Promega) according to the manufacturer's protocol. Growth curves were created and the population doubling times calculated using the following formula: PDT ¼ cultivation period h ½ Â log 10 2 ð Þ log 10 final cell count ð Þ − log 10 starting cell count ð Þ

Western blot analysis
Whole-cell lysates were prepared by using Triton-X sample buffer containing protease inhibitor cocktail from Roche (Grenzach-Wyhlen, Germany). The protein concentration was assessed by using the DC Protein assay reagent obtained from Bio-Rad Laboratories. Equal protein loads were separated by sodium dodecyl sulfatepolyacrylamide gel electrophoresis and proteins were transferred to nitrocellulose membranes (Thermo Scientific, Dreieich, Germany). The following primary antibodies were used at the indicated dilutions: SAMHD1

SAMHD1 promoter methylation
The SAMHD1 promoter contains five HpaII sites surrounding the transcription start site [7]. To measure methylation of the SAMHD1 promoter genomic DNA was treated with the methylation-sensitive HpaII endonuclease as described previously [7,36]. Methylation of the HpaII sites in the SAMHD1 promoter prevents digestion by HpaII and the intact sequence serves then as a template for PCR amplification using SAMHD1 promoter-specific primers that flank the HpaII sites: PM3.fwd: TTCCGCCTCATTCGTCCTTG and PM3.rev: GGTTCTCGGGCTGTCATCG were used as SAMHD1 promoter-specific primers. A single PCR product (993bp) corresponding to the SAMHD1 promoter sequence was obtained from untreated genomic DNA and treated DNA from cells with methylated but not from cells with unmethylated SAMHD1 promoter. To serve as input control, a 0.25-kb fragment of the GAPDH gene lacking HpaII sites was PCR-amplified using the same template DNA.

Manipulation of cellular SAMHD1 levels using siRNA or Vpx-VLPs
For siRNA-mediated silencing, AML cells (1 × 10 6 ) were transfected with 2.5 μM ON-TARGET plus human SAMHD1 siRNA SMART-pool obtained from Dharmacon (Munich, Germany, L-013950-01-0050) in resuspension electroporation buffer R (Invitrogen, Dreieich, Germany) using the Neon transfection system (Invitrogen) according to the manufacturer's recommendation. Additionally, ON-TARGET plus Non-targeting Control Pool obtained from Dharmacon (D-001810-10-50) was transfected in parallel. The electroporation was performed with one 20 msec pulse of 1700 V and analysed 48 h after transfection by western blotting and a cell viability assay. For Vpx virus-like particle (VLP)-mediated SAMHD1 degradation, cells were spinoculated with VSV-G pseudotyped virus-like particles carrying either Vpx or Vpr as control from SIVmac251. VLPs carrying Vpx or Vpr were produced by co-transfection of 293 T cells with pSIV3 + gag pol expression plasmids and a plasmid encoding VSV-G as previously described [36,41]. For viability assays cells were preincubated with VLPs for 24 h before the studied compounds were added.

LC-MS/MS analysis
AML or ALL cells were seeded at 2.5 × 10 5 cells per well in 24 well plates, treated with 10 μM CNDAC and incubated at 37°C in a humidified 5% CO2 incubator for 6 h.
Subsequently, cells were washed twice in 1 ml PBS, pelleted and stored at − 80°C until measurement. The concentrations of canonical dNTPs and CNDACtriphosphate in the samples were analysed by liquid chromatography-electrospray ionization-tandem mass spectrometry, as previously described for canonical dNTPs [43]. Briefly, the analytes were extracted by protein precipitation with methanol. An anion-exchange HPLC column (BioBasic AX, 150 × 2.1 mm, 5 μM, Thermo Scientific) was used for the chromatographic separation and a 5500 QTrap (Sciex, Darmstadt, Germany) was used as analyser, operating as triple quadrupole in positive multiple reaction monitoring (MRM) mode. CNDAC-TP was quantified using 2deoxycytidine-13 C 9 , 15 N 3 -triphosphate ( 13 C 9 , 15 N 3 -dCTP) as internal standard (IS). The precursor-to-product ion transition used as quantifier was m/z 493.1 → 112.1 for CNDAC-TP. Owing to the lack of commercially available standards for CNDAC-TP, relative quantification was performed by comparing the peak area ratios (analyte/IS) of the differently treated samples.

Crystallization and data collection
Purified SAMHD1 protein in buffer (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 5 mM MgCl2, and 0.5 mM TCEP) was mixed with 1 mM GTP, 0.1 mM dATP, and 10 mM CNDAC. All crystals were grown at 25°C using the microbatch under-oil method by mixing 1 μL of protein (3 mg/mL) with 1 μL of crystallization buffer (100 mM succinate-phosphate-glycine (SPG) buffer, pH 7.4, 25% PEG 1500; Qiagen). Crystals were improved by streak seeding. Crystals were cryoprotected in paratone oil and frozen in liquid nitrogen. Diffraction data were collected at Advanced Photon Source beamline 24-ID-E. The data statistics are summarized in Table 1.

Structure determination and refinement
Using the previously published SAMHD1 tetramer structure (PDB ID code 4BZB), with the bound nucleotides removed, as the search model, the structure was solved by molecular replacement using PHASER [29,45,46]. The model was refined with iterative rounds of restrained refinement using Refmac5 [33], followed by rebuilding the model to the 2Fo-Fc and the Fo-Fc maps using Coot [8]. Refinement statistics are summarised in Suppl. Table 5. Coordinates and structure factors have been deposited in the Protein Data Bank, with accession codes listed in Suppl. Table 5.

Enzymatic assay
In vitro SAMHD1 activity was measured as described [42].

Statistics
Statistical data analysis was performed using GraphPad Prism. Pearson's correlation coefficient was used to compute correlations between variables, using a t-test to assess significance of the correlation. Group comparisons were performed using Student's t-test.

SAMHD1 levels correlate with leukaemia cell sensitivity to CNDAC
Initially, we characterised a panel of 13 human AML cell lines for the levels of SAMHD1 and deoxycytidine kinase (DCK) (Fig. 1A). DCK phosphorylates and activates cytidine analogues in a rate-limiting step [15,28,48] and may, hence, determine cell sensitivity to a nucleoside analogue like CNDAC anticipated to be a SAMHD1 inhibitor [14]. We detected varying SAMHD1 and DCK levels (Fig. 1A, Suppl. Figure 1), varying CNDAC concentrations that reduced cell viability by 50% (IC 50 ) (Fig.  1B, Suppl. Figure 2, Suppl. Table 1), and varying CNDA C-TP levels (Fig. 1C) across the investigated cell lines. However, the CNDAC IC 50 s did not correlate with the cellular levels of DCK (Fig. 1D), indicating that DCK is not a critical determinant of CNDAC activity in our cell line panel.
In contrast, the CNDAC IC 50 s correlated with the cellular SAMHD1 levels (Fig. 1E), suggesting that SAMHD1 may cleave and inactivate CNDAC-TP, although CNDAC had been proposed to be a SAMHD1 inhibitor [14]. Also, there was no correlation between cellular CNDAC-TP and DCK levels (Fig. 1F), but an inverse correlation of the CNDAC-TP levels with SAMHD1 (Fig. 1G). This further supports the notion that SAMHD1 but not DCK critically determines CNDA C phosphorylation and activity. Notably, SAMHD1 promoter methylation (Fig. 1H) did not always indicate cellular SAMHD1 levels (Fig. 1A), showing that multiple mechanisms are involved in regulating the cellular abundance of this protein.
The CNDAC IC 50 s also correlated with the cellular SAMHD1 levels in acute lymphoblastic leukaemia (ALL) cells (Fig. 1I, Suppl. Table 2). In agreement with previous findings [39], T-cell ALL (T-ALL) cells were characterised by lower SAMHD1 levels than B-ALL cells (Fig.  1J). This was reflected by higher CNDAC sensitivity (Fig.  1K) and higher CNDAC-TP levels (Fig. 1L) in T-ALL cells than in B-ALL cells. Taken together, these findings suggest that CNDAC is a SAMHD1 substrate and that SAMHD1 but not DCK critically determines CNDAC phosphorylation and activity in AML and ALL cells.
Results from the enzymatic assay confirmed that SAMHD1 hydrolyses CNDAC-TP only in the presence of dGTP (Fig. 4A). This indicates that CNDAC-TP is a SAMHD1 substrate but not able to activate the enzyme by binding to A1 and A2.

Crystal structure of CNDAC-TP bound to SAMHD1
To investigate the interaction of CNDAC-TP and SAMHD1 further, we crystallised the catalytically inactive HD domain (residues 113-626; H206R, D207N) of SAMHD1 in the presence of GTP, dATP, and excess CNDAC-TP as previously described [21] and collected diffraction data to 2.8 Å. SAMHD1 crystallised as a tetramer with GTP and dATP occupying A1 and A2, respectively, and CNDAC-TP bound to the catalytic site (Fig. 4B, Suppl. Table 5).
Previous studies investigating the binding of triphosphorylated nucleoside analogues to SAMHD1 showed that modifications at the 2'ribose (R) position are major determinants of interaction with the catalytic SAMHD1 site [21]. Analogues with 2'R modifications abrogate binding to SAMHD1, while 2′S stereoisomers are more permissive. Furthermore, the catalytic site tolerates larger 2′S modifications, whereas analogue binding at the A2 site is either impaired or fully blocked by 2′S fluorination or hydroxylation of the sugar ring, respectively [21].
Consistent with these observations, the CNDAC-TPbound SAMHD1 adopts the same conformation as the canonical nucleotide-bound form (overall RMSD: 0.30 Å vs PDB ID 4BZB). The ribose 2′S nitrile modification of CNDAC-TP (Fig. 4C) protrudes outward from the catalytic pocket without affecting canonical nucleotide contacts with active site residues. CNDAC-TP is therefore easily accommodated in the catalytic site to serve as a substrate for SAMHD1 triphosphohydrolase activity. However, the large nitrile group of CNDAC-TP prevents binding to the more restrictive A2 site. Thus, CNDAC alone is insufficient for SAMHD1 activation.

Impact of CNDAC on cellular levels of physiological nucleoside triphosphates and the activity of SAMHD1 substrates
The finding that CNDAC-TP is itself a substrate of SAMHD1 does not exclude the possibility that it also exerts inhibitory effects on SAMHD1, as previously suggested [14]. Hence, we investigated the effects of CNDA C on the levels of physiological desoxynucleoside triphosphates (dNTPs) and the activity of cytarabine, the triphosphate of which is known to be a SAMHD1 substrate [41].
CNDAC did in contrast to VPX-VLPs, which served as a positive control for suppressing SAMHD1 activity, not increase the levels of physiological dNTPs (Fig. 4D). Moreover, CNDAC did not increase the activity of cytarabine (Fig. 4E). Thus, these findings do not suggest a pharmacologically relevant activity of CNDAC as SAMHD1 inhibitor in AML cells.
There was a significant correlation between SAMHD1 protein levels (but not the DCK protein levels) and the CNDAC IC 50 s (Fig. 5E), and siRNA-mediated SAMHD1 depletion resulted in increased CNDAC (but not daunorubicin) sensitivity in three selected clones displaying differing SAMHD1 levels (Fig. 5F, Suppl. Figure 8). The different effects of SAMHD1 on CNDAC-and daunorubicinmediated toxicity suggest that SAMHD1 interferes with CNDAC activity predominantly by cleaving CNDAC-TP and not by generally augmenting DNA repair.
Differences in cellular SAMHD1 levels may affect cell proliferation [10,22,23,49], but there was no significant correlation between the SAMHD1 (or DCK) levels of the MV4-11 clones and their doubling times (Fig. 5G).
Taken together, these findings confirm that the response to CNDAC is primarily driven by the SAMHD1 levels in CNDAC-naïve AML cells. Acquired resistance to CNDAC is associated with decreased DCK levels To investigate the role of SAMHD1 in acquired CNDAC resistance, we established twelve CNDAC-resistant sublines of each of the AML cell lines HL-60 and PL-21, which are characterised by low SAMHD1 levels (Fig.  1A) and high CNDAC sensitivity (Fig. 1B). Interestingly, none of the 24 resulting CNDAC-resistant sublines displayed increased SAMHD1 levels but all showed reduced, virtually non-detectable DCK levels (Fig. 6A). Among twelve single cell-derived clones of HL-60 and PL-21, none displayed similarly low DCK levels (Fig.  6A). Then, we determined resistance profiles in the CNDAC-resistant HL-60 and PL-21 sublines and the clonal HL-60 and PL-21 sublines to a set of cytotoxic (CNDAC, sapacitabine, cytarabine, clofarabine, cladribine, fludarabine, gemcitabine, decitabine, azacytidine, 6-thioguanine, daunorubicin) and targeted (venetoclax, vismodegib, olaparib, ganetespib, volasertib, gedatolisib, molibresib) drugs (Fig. 6B, Suppl. Table 6). In addition to resistance to CNDAC and its prodrug sapacitabine, all CNDAC-adapted sublines also consistently displayed a markedly reduced sensitivity to the nucleoside analogues clofarabine, cladribine, fludarabine, gemcitabine, and decitabine, whose activation critically depends on monophosphorylation by DCK (Fig. 6B, Suppl. Table 6). In contrast, there was no crossresistance to the nucleoside analogues azacytidine and 6thioguanine that are no DCK substrates and to the anthracycline daunorubicin. This suggests that the reduced DCK expression is the predominant acquired resistance mechanism in our panel of CNDAC-adapted AML cell lines.
This notion was also confirmed by the general lack of cross-resistance to targeted drugs with a range of different targets, including the smoothend receptor (vismodegib), PARP1 (olaparib), HSP90 (ganetespib), PLK1 (volasertib), and PI3K/mTOR (gedatolisib). There was some level of resistance to the BET inhibitor molibresib among the CNDAC-adapted sublines (Fig. 6B, Suppl. Table 6). However, some level of resistance to these drugs was also detected among the clonal HL-60 and PL-21 sublines (Fig. 6B, Suppl. Table 6), which may suggest that this molibresib resistance may rather be the consequence of clonal selection processes during resistance formation and not part of the acquired CNDAC resistance mechanisms.
The Bcl-2 inhibitor venetoclax was the only targeted drug against which the CNDAC-adapted sublines displayed an increased level of resistance that was not detectable in the clonal sublines (Fig. 6B, Suppl. Table 6). This may indicate a generally increased resistance to apoptosis in the CNDAC-adapted sublines (Fig. 6B, Suppl. Table 6), which may reflect that apoptosis induction is anticipated to be part of the anti-cancer mechanism of action of CNDAC [27].
Taken together, our findings suggest that DCK downregulation is the major acquired CNDAC resistance mechanism in AML cells, potentially complemented by a generally reduced potential to undergo apoptosis.

Role of SAMHD1 and DCK in CNDAC cross-resistance of AML cell lines adapted to drugs from different classes
In contrast to the CNDAC-adapted AML cell lines introduced here, which displayed reduced DCK expression as main acquired resistance mechanism, AML cell lines adapted to the SAMHD1 substrates cytarabine or decitabine were characterised by a combination of increased SAMHD1 levels and decreased DCK levels [36,41].
CNDAC-adapted AML sublines displayed pronounced cross-resistance to nucleoside analogues that are activated by DCK but not to anti-leukaemia drugs with other mechanisms of action (Fig. 6). In a reversed setting, we next investigated CNDAC in a panel consisting of the AML cell line HL-60 and its sublines adapted to the nucleoside analogues cytarabine, Ara-G, azacytidine, and fludarabine, the purine antagonist 6mercaptopurine, the Bcl-2 inhibitor venetoclax, the PARP inhibitor olaparib, and the polo-like kinase 1 inhibitor volasertib.
The nucleoside analogue-resistant HL-60 sublines displayed increased SAMHD1 and/ or decreased DCK levels (Fig. 7A) and pronounced CNDAC resistance (Fig.  7B, Suppl. Figure 9), while little or no CNDAC resistance was detected in the remaining sublines (Fig. 7A,  Fig. 7B, Suppl. Figure 9). Moreover, cellular SAMHD1 levels directly and cellular DCK levels inversely correlated with the CNDAC IC 50 s (Fig. 7C), indicating that enhanced SAMHD1 levels and reduced DCK levels contribute to cross-resistance to CNDAC. VPX-VLPmediated SAMHD1 depletion sensitised nucleoside analogue-adapted HL-60 sublines to CNDAC to various extents (Fig. 7D), which probably reflects the relative importance of SAMHD1 and DCK levels for CNDAC resistance in these cell lines.
Next, we used cytarabine-adapted MV4-11 and MOLM-13 sublines to further study the role of SAMHD1 and DCK in cross-resistance of nucleoside analogue-adapted AML cells to CNDAC (Fig. 7E). In both cell lines, VPX-VLPmediated SAMHD1 depletion resulted in reduced CNDAC IC 50 s, which further decreased upon forced DCK expression. Similar findings were obtained with regard to the cytarabine resistance in these two cell lines (Fig. 7E). This confirms that, in principle, cellular SAMHD1 and DCK levels are involved in determining AML cell sensitivity to CNDAC (and cytarabine), although, as shown in this study, intrinsic and acquired CNDAC resistance differ in AML cells in that intrinsic CNDAC resistance is predominantly driven by high SAMHD1 levels and acquired CNDAC resistance by a reduction in DCK.

Discussion
The findings of this study indicate that in AML cells intrinsic CNDAC resistance is predominantly driven by SAMHD1, whereas acquired CNDAC resistance is primarily caused by reduced DCK levels. This difference is of potential clinical significance, because SAMHD1 is a candidate biomarker for predicting CNDAC sensitivity in therapy-naïve patients, while DCK is a candidate biomarker for the early detection of resistance formation. SAMHD1 is known to interfere with the activity of a range of anti-cancer nucleoside analogues as hydroxylase that cleaves the activated nucleoside analogue triphosphates [ [13,21,36,39,41]; Xagorias et al., 2021]. The finding that SAMHD1 levels critically determine AML (and ALL) cell sensitivity to CNDAC is nevertheless somewhat unexpected, as CNDAC had originally been suggested to be a SAMHD1 inhibitor [14].
However, data from a large range of cell line models (including clonal AML sublines characterised by varying SAMHD1 levels) and patient samples demonstrated that high SAMHD1 levels are associated with reduced CNDA C sensitivity and that CRISPR/Cas9-, siRNA-, and VPX-VPL (virus-like particles carrying the lentiviral VPX protein)-mediated SAMHD1 depletion increase cellular CNDAC-TP levels and sensitise AML cells to CNDAC. In agreement, enzymatic assays and crystallisation studies showed that CNDAC-TP is cleaved by SAMHD1, but can in contrast to some other nucleoside analogues [14,16,17,21,36] not activate SAMHD1 via binding to the A2 site.
Moreover, the determination of physiological dNTPs in the presence of CNDAC and combination experiments with the SAMHD1 substrate cytarabine did not provide evidence that CNDAC may function as pharmacological SAMHD1 inhibitor in leukaemia cells.
Although cellular SAMHD1 levels, but not those of DCK that is critical for CNDAC phosphorylation and activation [ [15,28]; Wu et al., 2021], predominantly determined CNDAC sensitivity in CNDAC-naïve cells, the establishment of 24 CNDAC-resistant AML sublines unanimously resulted in a loss of DCK but not in an increase of SAMHD1. This differs from acquired resistance mechanisms against the nucleoside analogues cytarabine and decitabine that were found to include both increased SAMHD1 levels and decreased DCK levels [36,41]. Two previously established CNDACadapted cancer cell lines had been shown to display reduced DCK levels but a contribution of SAMHD1 had not been investigated [34,35].
CNDAC-adapted AML sublines consistently displayed cross-resistance to other nucleoside analogues known to be activated by DCK but no pronounced crossresistance to other drugs with various mechanisms of action, further indicating that loss of DCK is the crucial resistance mechanism in CNDAC-adapted cells. Moreover, these data also show that drugs, which do not depend on DCK for activation, remain viable treatment options after resistance acquisition to CNDAC.
Similarly, among AML sublines adapted to a range of different anti-cancer drugs, only nucleoside analogues that displayed increased SAMHD1 and/ or decreased DCK levels were less sensitive to CNDAC. Thus, acquired resistance to a range of different anti-leukaemic drugs is unlikely to affect the efficacy of CNDAC.
Cytarabine-and decitabine-adapted AML cell lines are characterised by a combination of increased SAMHD1 levels and/ or reduced DCK levels as demonstrated previously [36,41]. Although acquired CNDAC Values are depicted as fold changes in drug concentrations that reduce cell viability by 50% (IC 50 s) between the respective parental AML cell line (shown in red) and the resistant cell lines or clones. Points closer to the centre than red lines indicate higher sensitivity to drugs in CNDAC-resistant sublines or clonal sublines than in parental cell lines, while points lying outside red lines indicate reduced sensitivity to the respective drug. IC 50 fold changes are shown as means from three independent experiments. Numerical values are provided in Supplementary Table 6 resistance was mediated by decreased DCK levels, both increased SAMHD1 levels and decreased DCK levels contributed to cross-resistance of cytarabine-adapted cells to CNDAC. In the future, it will be interesting to investigate why acquired resistance mechanisms differ between CNDAC-adapted cells on the one hand and cytarabine-and decitabine-adapted cells on the other hand.

Conclusion
Intrinsic AML cell response to CNDAC critically depends on cellular SAMHD1 levels, whereas acquired CNDAC resistance is predominantly mediated by reduced DCK levels. This adds to data demonstrating differences between intrinsic and acquired resistance mechanisms [31,36,40,44]. These findings also indicate that SAMHD1 is a candidate biomarker predicting . GAPDH served as loading control. (B) CNDAC concentrations that reduce cell viability by 50% (IC 50 s) in drug-adapted HL-60 sublines. Horizontal lines and error bars represent means ± SD of three independent experiments, each performed in three technical replicates. p-values were determined by two-tailed Student's t-tests (*p < 0.05; **p < 0.01; ***p < 0.001). (C) Correlation of CNDAC IC 50 values with cellular SAMHD1 or DCK protein levels, quantified using the nearinfrared Western blot image shown in (A) to determine the ratio SAMHD1/GAPDH or DCK/GAPDH. (D) CNDAC dose-response curves in drug-adapted HL-60 sublines in the absence or presence of VPX virus-like particles (VPX-VLPs, cause SAMHD1 depletion) or VPR virus-like particles (VPR-VLPs, negative control). Each symbol represents the mean ± SD of three technical replicates of one representative experiment out of three. Concentrations that reduce AML cell viability by 50% (IC 50 s) ± SD and Western Blots showing SAMHD1 degradation by VPXVLPs are provided. (E) CNDAC or cytarabine (Ara-C) dose-response curve in cytarabineadapted MV4-11 or MOLM-13 cells (characterised by loss of DCK expression) stably transduced with either DCK (pWPI+DCK) or an empty vector (pWPI) in the absence or presence of VPX virus-like particles (VPX-VLPs), or VPR virus-like particles (VPR-VLPs). Each symbol represents the mean ± SD of three technical replicates of one representative experiment out of three. IC 50 s (mean ± SD) and Western Blots showing successful transduction with DCK and SAMHD1 degradation by VPX-VLPs are provided CNDAC response in the intrinsic resistance setting, while DCK plays a potential role as biomarker indicating therapy failure early in the acquired resistance setting. Moreover, CNDAC-adapted cells displayed no or limited cross-resistance to drugs whose activity is not influenced by DCK or SAMHD1. Similarly, CNDAC was still effective in cells adapted to drugs that are not affected by DCK or SAMHD1. These findings indicate treatment options after therapy failure.