Galactose-modified duocarmycin prodrugs as senolytics

Senescence is a stable growth arrest that impairs the replication of damaged, old or preneoplastic cells, therefore contributing to tissue homeostasis. Senescent cells accumulate during ageing and are associated with diseases, such as cancer, fibrosis and many age-related pathologies. Recent evidence suggests that the selective elimination of senescent cells can be effective on the treatment of many of these senescence-associated diseases. A universal characteristic of senescent cells is that they display elevated activity of the lysosomal β-galactosidase this has been exploited as a marker for senescence (senescence-associated β-galactosidase activity). Consequently, we hypothesised that galactose-modified cytotoxic prodrugs will be preferentially processed by senescent cells, resulting in their selective killing. Here, we show that different galactose-modified duocarmycin (GMD) derivatives preferentially kill senescent cells. GMD prodrugs induce selective apoptosis of senescent cells in a lysosomal β-galactosidase (GLB1)-dependent manner. GMD prodrugs can eliminate a broad range of senescent cells in culture, and treatment with a GMD prodrug enhances the elimination of bystander senescent cells that accumulate upon whole body irradiation or doxorubicin treatment of mice. Moreover, taking advantage of a mouse model of human adamantinomatous craniopharyngioma (ACP), we show that treatment with a GMD pro-drug result selectively reduced the number of β-catenin-positive preneoplastic senescent cells, what could have therapeutic implications. In summary, the above results show that galactose-modified duocarmycin prodrugs behave as senolytics, suggesting that they could be used to treat a wide range of senescence-related pathologies.

Despite that the acute induction of senescence limits fibrosis and protects against cancer progression, the abnormal accumulation of senescent cells with age or in diseased tissues is detrimental (Munoz-Espin & Serrano, 2014). Interestingly, evidence drawn from genetic models has shown that eliminating senescent cells increases lifespan, improves healthspan and benefits the outcomes of a wide range of diseases (Baker et al., 2011Childs et al., 2016Childs et al., , 2017. These studies have led to a collective effort to identify "senolytics," drugs that selectively kill senescent cells. Several senolytics have been identified including dasatinib and quercetin (Zhu et al., 2015), piperlongumine , FOXO4-interfering peptides (Baar et al., 2017), HSP90 inhibitors (Fuhrmann-Stroissnigg et al., 2017), cardiac glycosides (Guerrero et al., 2019;Triana-Martinez et al., 2019) or the Bcl2 family inhibitors  (navitoclax) and ABT-737 (Chen et al., 2015;Yosef et al., 2016;Zhu et al., 2016). Currently, Bcl2 family inhibitors have become the gold standard on senolysis. Bcl2 family inhibitors eliminate a range of senescent cells in vivo and reproduce the effects observed in transgenic mice modelling senescence ablation (Ovadya & Krizhanovsky, 2018). However, ABT-263 causes severe thrombocytopenia and neutropenia, what might complicate its use on the clinic. Moreover, it is becoming evident that specific senolytics might be necessary to eliminate different types of senescent cells. Therefore, there is a need to identify additional drugs with senolytic properties.
An alternative strategy for targeting senescence is to exploit properties that differentiate senescent from normal cells. In this regard, the senescence-associated β-galactosidase activity (SA-β-gal) is one of the more conserved and defining characteristics of senescent cells. Senescent cells present an increased lysosomal mass (Kurz, Decary, Hong, & Erusalimsky, 2000). As a result, senescent cells display elevated levels of lysosomal enzymes such as β-galactosidase (encoded by GLB1 (Dimri et al., 1995)) or α-fucosidases (Hildebrand et al., 2013). Indeed, it has been shown that galacto-oligosaccharide encapsulated nanoparticles (GalNP) preferentially release their content on senescent cells (Agostini et al., 2012). Consequently, this GalNP can be used in combination with different cargos to either image or kill senescent cells (Munoz-Espin et al., 2018).
Galactose modification has been frequently used to improve the pharmacokinetic properties or the delivery of existing drugs. In addition, galactose modification can be used to generate prodrugs that rely on E. coli β-galactosidase for controlled activation (Melisi, Curcio, Luongo, Morelli, & Rimoli, 2011). When combined with antibody-linked β-galactosidase, this approach is known as antibody-directed enzyme prodrug therapy (ADEPT) (Bagshawe, 2006;Tietze & Schmuck, 2011). In ADEPT, a conjugate of a tumour-specific antibody and an enzyme, such as β-galactosidase, is combined with the application of a hardly cytotoxic prodrug. By means of the enzyme in the conjugate, the prodrug is selectively cleaved in cancer cells leading to the formation of a highly cytotoxic compound.
Here, we investigated whether galactose-modified prodrugs can preferentially kill senescent cells. We have assessed several GMD derivatives and confirmed their senolytic potential in cell culture, ex vivo and in vivo. Given the increasing list of senescence-associated diseases and the benefits of senolytic treatment, we propose that GMD derivatives and, more generally, galactose-modified prodrugs are a new class of senolytic compounds and they should be tested to assess their therapeutic potential.

| A galactose-modified duocarmycin prodrug with senolytic properties
The natural antibiotic duocarmycin is a highly cytostatic compound (Boger & Johnson, 1995). A series of glycosidic derivatives of duocarmycin have been previously developed to be used as prodrugs in the context of antibody-directed enzyme prodrug therapy (ADEPT) (Tietze, Hof, Muller, Krewer, & Schuberth, 2010;Tietze et al., 2009). Given that senescent cells display elevated levels of SA-βgalactosidase activity, we hypothesized that galactose-modified cytotoxic prodrugs will be preferentially processed by senescent cells, resulting in their selective killing. To test this hypothesis, we took advantage of a galactose-modified duocarmycin (GMD) prodrug (referred as prodrug A, JHB75B) previously described (Tietze et al., 2009). We analysed the effects that a seco-duocarmycin analogue dimer (duocarmycin SA) and its galactose derivative (prodrug A) had on the survival of IMR90 ER:RAS cells, a model of oncogene-induced senescence (OIS). Activation of the ER:RAS fusion with 4-hydroxytamoxifen (4OHT) induces senescence in IMR90 ER:RAS cells (Georgilis et al., 2018). Treatment with duocarmycin SA was equally effective in killing normal and senescent cells, with the exception of a small selectivity towards senescent cells at the lower concentrations ( Figure 1a). In contrast, when we treated IMR90 ER:RAS cells with prodrug A (differing only in the addition of two galactose groups that inactivate it), we observed the preferential elimination of senescent cells (Figure 1b and Figure S1a). Duocarmycins are known to bind and alkylate DNA in AT-rich regions of the minor groove and induce cell death in a way dependent of DNA replication (Boger et al., 1994;Tietze et al., 2006Tietze et al., , 2009 We checked that senescent cells were growth arrested at the time of the drug treatment ( Figure S1b). This shows that the effect observed is not due to hyperreplication of cells during early stages of OIS and suggests that the prodrug might act by some of the alternative cytotoxic mechanisms described for duocarmycin dimers (Wirth et al., 2012). Treatment with prodrug A induced caspase 3/7 activity on senescent cells (Figure 1c), and the selective death of these cells was prevented with a pan-caspase inhibitor ( Figure 1d). The above results suggest that GMD prodrugs can behave as senolytics by selectively inducing apoptosis on senescent cells.

| Senolytic properties of prodrug A depend on the lysosomal β-galactosidase
We had hypothesized that GMD prodrugs could behave as senolytics due to the higher SA-β-galactosidase activity of senescent cells. To investigate whether the levels of β-galactosidase activity correlate with the sensitivity of senescent cells to GMD prodrugs, we induced senescence in IMR90 ER:RAS cells. Afterwards, we treated control or senescent cells with 2.5 µM prodrug A and used a fluorescent substrate (DDAO) to quantify SA-β-galactosidase activity at single- . IMR90 ER:RAS cells were treated with DMSO or with 4OHT (4-hydroxy-tamoxifen) for 6 days to induce OIS. Cells were treated with the indicated concentrations of seco-duocarmycin analog dimer for 72 hr. Cell numbers were quantified using DAPI staining, and percentage of survival cells are plotted (right) (n = 4). (b) Molecular structure of a galactose-modified prodrug derivative of seco-duocarmycin analog dimer (JHB75B, referred as prodrug A, left). Cells were treated with prodrug A for 72 hr as described before (n = 4). (c) Treatment of senescent cells with a GMD prodrug triggers caspase-3/7 activity. IMR90 ER:RAS were treated with 4OHT or vehicle (DMSO) for 6 days to induce senescence. 2.5 μM prodrug was then added together with NucLight Rapid Red Reagent for cell labelling and Caspase-3/7 reagent for apoptosis (IncuCyte). Caspase 3/7 activity was measured at 4-hr intervals. (d) After 6-day treatment with 4OHT or vehicle (DMSO), IMR90 ER:RAS were treated with 1 μM ABT-263 or 2.5 μM prodrug A for 72 hr in the presence or absence of the pancaspase inhibitor Q-VD-OPh (n = 4). All statistical significances were calculated using unpaired Student's t tests. All error bars represent mean ± s.d; n represents independent experiments.; ns, not significant; *p < .05; **p < .01; ***p < .001, ****p < .0001 Quantification of cell survival of senescent and control IMR90 ER:RAS infected with different shRNAs targeting GLB1 or an empty vector and treated with ABT-263 or prodrug A for 3 days (n = 3). Statistical significance was calculated using two-tailed, Student's t test. All error bars represent mean ± SD; n represents independent experiments; ns, not significant; *p < .05; **p < .01; ***p < .001, ****p < .0001 Figure S2), linking the senolytic selectivity of prodrug A with the SAβ-galactosidase activity.
The increase in β-galactosidase observed on senescent cells is due to an increase in lysosomal mass (Kurz et al., 2000) resulting in higher activity of the lysosomal β-galactosidase (encoded by GLB1) (Lee et al., 2006). To further prove that the senolytic activity of GMD prodrugs is dependent on SA-β-galactosidase, we took advantage of three independent shRNAs to knock down GLB1 (Figure 2d).

Knock-down of GLB1 in IMR90 ER:RAS cells resulted in decreased
SA-β-galactosidase activity, but it did not impact the growth arrest or the induction of p16 INK4a observed during OIS (Figure 2e,f and Figure S3a,b). Taking advantage of these cells, we observed that GLB1 knock-down did not affect the senolytic potential of ABT-263 but ablated the ability of prodrug A to selectively kill senescent cells ( Figure 2g and Figure S3c). In summary, our data suggest that GMD prodrugs trigger apoptosis of senescent cells in a GLB1-dependent manner.

| Galactose-modified duocarmycin prodrugs are broad-spectrum senolytics
To understand the extent to which GMD prodrugs behave as senolytics, we assessed the effect that prodrug A has on several types of senescent cells. To this end, we took advantage of IMR90 cells and induced senescence by etoposide or doxorubicin treatment, irradiation, or serial passage ( Figure S4a Figure S2. All statistical significances were calculated using unpaired Student's t tests. All error bars represent mean ± SD; n represents independent experiments; ns, not significant; *p < .05; **p < .01; ***p < .001 general concept (conversion of other cytotoxic drugs in galactosemodified prodrugs) was wider. To this end, we took advantage of two previously described GMD prodrugs, JHB76B and JHB35B (Tietze et al., 2009(Tietze et al., , 2010. Both drugs were also effective in selectively eliminating senescent cells (Figure 3f and Figure S5), suggesting that the generation of galactose-modified prodrugs might be a general route to design senolytic compounds.

| Prodrug A exerts a bystander effect
The above experiments suggest that senescent cells preferentially convert GMD prodrugs into their active duocarmycin derivatives.
Since duocarmycins have strong cytostatic properties (Boger & Johnson, 1995), we wonder whether the conversion of GMD prod-

| Prodrug A eliminates senescent cells in vivo
Chemotherapy and radiotherapy are amongst the most common anticancer treatments. Irradiation, chemotherapy and even some targeted anticancer drugs, all induce senescence (Schmitt et al., 2002;Wang et al., 2017). Although induction of tumour senescence explains the anticancer properties of these treatments, the generation of bystander senescent cells is responsible for their side effects (Demaria et al., 2017). To assess whether prodrug A could eliminate these senescent cells, we first irradiated mice and upon a latency period to allow for the accumulation of senescent cells, treated them with prodrug A, ABT263 or vehicle (Figure 5a). Treatment with prodrug A or ABT-263 resulted in a reduced presence of senescent cells in lung as assessed using SA-β-galactosidase activity (Figure 5b,c).
Furthermore, a similar trend was observed when we assessed the expression of Cdkn1a (that encodes for p21 Cip1 ) or the SASP components Il6 and Cxcl1 (Figure 5d).

| Galactose-modified prodrugs eliminate preneoplastic senescent cells
OIS is primarily considered as a tumour suppressive mechanism (Collado et al., 2005),  Here, we add galactose-modified duocarmycin (GMD) prodrugs as a new class of senolytic agents. These GMD prodrugs are converted to their corresponding duocarmycin drugs in a manner dependent on processing by β-galactosidase. Since senescent cells display elevated levels of lysosomal β-galactosidase (encoded by GLB1), GMD selectively affects senescent cells. In this manuscript, we present evidence showing that GMD prodrugs can eliminate multiple types of senescent cells, what is consistent with SA-βgalactosidase being a universal marker of senescence. Our preliminary results suggest that GMD prodrugs could also be capable of

| D ISCUSS I ON
eliminating bystander senescent cells arising from anticancer therapies and preneoplastic senescent cells in mouse models. This needs further investigation.
Given the promise that senolytics present for the treatment of age-related disease, and their associated benefits over healthspan and lifespan, we believe that this study provides the basis to specifically assess the potential benefits of GMD on ageing.
Previously, the potential to harness the elevated β-galactosi- In summary, we have described that galactose-modified duocarmycin prodrugs are a new class of broad-spectrum senolytic agents. We have characterized their ability to eliminate different types of senescent cells in culture and carried out preliminary experiments in vivo. Given the increasing list of diseases that are associated with senescence, galactose-modified duocarmycin prodrugs have the potential to be used in the context of anticancer therapies and to treat different age-related diseases. The present study should provide the basis for investigating the potential of galactose-modified prodrugs as senolytics to treat senescence-associated diseases.

| Drugs
The following compounds were used in this study: ABT-263

| Antibodies
The following primary antibodies were used in this study: mouse

| Cell lines
IMR90 cells were obtained from ATCC. IMR90 ER:RAS were generated by retroviral infection of IMR90 cells and have been described elsewhere Barradas et al., 2009) F I G U R E 6 Galactose-modified duocarmycin prodrug eliminates preneoplastic senescent lesions. (a) Experimental design for the senolytic experiment in the Hesx1 Cre/+ ;Ctnnb1 lox(ex3)/+ mouse model of adamantinomatous craniopharyngioma (ACP). Tumoural pituitaries from 18.5dpc Hesx1 Cre/+ ;Ctnnb1 lox(ex3)/+ embryos were cultured in the presence of prodrug A at the indicated concentrations or vehicle (DMSO) and processed for analysis after 72 hr. (b) Immunofluorescence staining against β-catenin (green) and synaptophysin (red) is shown. Synaptophysin is a marker of the normal hormone-producing cells in the pituitary gland. Scale bar, 50μm. (c) Quantification of β-cateninaccumulating cells after treatment with different concentrations of prodrug A or vehicle (n = 6-12). (d) Quantification of synaptophysinpositive cells after treatment with different concentrations of prodrug A or vehicle (n = 6-12). (e) Quantification of β-catenin-accumulating cells positive for cleaved caspase-3 after treatment with different concentrations of prodrug A or vehicle (n = 6-12). All statistical significances were calculated using nonparametric ANOVA with Dunn's post hoc comparison. All error bars represent mean ± SD; n represents number of pituitaries; ns, not significant; *p < .05; **p < .01; ***p < .001
To generate IMR90 ER:RAS-expressing shRNAs against GLB1, lentiviral infections were carried out as described before (Aarts et al., 2017). Briefly, HEK293T cells were transfected with the lentiviral and packaging vectors using PEI (PEI 2500, Polysciences). Two days after transfection, HEK293T viral supernatants were collected, filtered (0.45 μM), diluted 1/4, supplemented with 4μg/ml polybrene and added to IMR90 ER:RAS cells plated the day before at a density of 1 million cells per 10-cm dish. Four hours later, lentivirus-containing media was replaced with fresh media. Three days after infection, cells were passaged and cultured for three days in the presence of 1μg/ml puromycin (InvivoGen) to select for infected cells.

| BrdU incorporation
BrdU incorporation assays were performed as previously described (Georgilis et al., 2018). Briefly, for BrdU incorporation assays, the cells were incubated with 10 μM BrdU for 16-18 hr before being fixed with 4% PFA (w/v). BrdU incorporation was assessed by Immunofluorescence and High Content Analysis microscopy.

| Immunofluorescence staining of cells
Cells were grown in 96-well plates, fixed with 4% PFA (w/v) and stained as previously described (Georgilis et al., 2018).

| Cytochemical SA-β-galactosidase assay
Cells were grown on 6-well plates, fixed with 0.5% glutaraldehyde For SA-β-galactosidase staining in tissues, frozen sections (6 μm) were fixed in ice-cold 0.5% glutaraldehyde (w/v) solution for 15 min, washed with 1mM MgCl 2 /PBS (pH 6.0) for 5 min and then incubated with X-Gal staining solution for 16-18 hr at 37°C as previously described (Georgilis et al., 2018). After the staining, the slides were counterstained with eosin, dehydrated, mounted and analysed by phase-contrast microscopy. SA-β-Gal tissue staining was quantified using ImageJ software (NIH) by measuring the percentage of stained area in each section and multiplying it by its mean intensity value as described before (Tordella et al., 2016). To exclude the luminal spaces in the lung sections, the percentage of SA-β-Gal-positive area was divided by the total lung area, as determined by eosin-positive area using ImageJ (NIH).

| Fluorescent SA-β-galactosidase assay
Cells were grown on 96-well plates. After 8 days, cells were treated with 100 μM DDAOG (D-6488, Thermo Fisher Scientific) 2 hr before fixation with 4% PFA for 15 min. Nuclei were stained with DAPI (1 μg/ml) for 15 min and images acquired the same day using a IN Cell Analyzer 2000 (GE Healthcare).

| High content analysis (HCA)
IF imaging was carried out using the automated high-throughput fluorescent microscope IN Cell Analyzer 2000 (GE Healthcare) with a 20× objective. Multiple fields within a well were acquired in order to include a minimum of 1,000 cells per sample-well. HCA of the images was processed using the INCell Investigator 2.7.3 software as described previously (Herranz et al., 2015). Briefly, DAPI served as a nuclear mask hence allowed for the segmentation of cells with a Top-Hat method. To detect cytoplasmic staining in cultured cells, a collar of 7-9 μm around DAPI was applied. In samples of cultured cells, a threshold for positive cells was assigned above the average intensity of unstained or negative control sample unless otherwise specified.

| Gene expression analysis
Total RNA was extracted using Trizol reagent (Invitrogen) and the RNeasy isolation kit (Qiagen). cDNA was generated using random hexamers and SuperScript II reverse transcriptase (Invitrogen).
Quantitative real-time PCR was performed using SYBR Green PCR master mix (Applied Biosystems) in a CFX96 real-time PCR detection system (Bio-Rad). Rps14 expression was used for normalization.
Mouse primer pairs are:

| Mouse models and drug treatments
For induction of senescence, C57BL/6J mice at 8-12 weeks of age were exposed to a sublethal dose (6 Gy) of total body irradiation. Eight weeks after, mice were injected with 50 nmols of prodrug A (i.v.) or vehicle for four consecutive days. Mice were killed 24 hr after the last injection.
Mice lungs were harvested for RNA extraction, paraffin embedded for immunohistology, or frozen in OCT/Sucrose 15% (1:1) solution for cryosectioning and SA-β-gal stains. The mice used for all experiments were randomly assigned to control or treatment groups. Both sexes were used throughout the study.
Immunofluorescence staining was performed as previously described (Gonzalez-Meljem et al., 2017). The proportion of β-cateninaccumulating and p21-positive cells was calculated as an index out of the total DAPI-stained nuclei. The proportion of β-cateninaccumulating, cleaved caspase-3 and p21-positive cells was calculated as an index out of the total DAPI-stained nuclei. Over 300,000 DAPI nuclei were counted from ten histological sections per sample, in a total of twelve neoplastic pituitaries.

| Statistical analysis
GraphPad Prism 7.0 was used for statistical analysis. Two-tailed Student's t tests were used to estimate statistically significant differences between two groups. Two-way ANOVA with Tukey's post hoc comparison was used for multiple comparisons. Values are presented as mean ± SD unless otherwise indicated. Asterisks (*) always indicate significant differences as follows: ns = not significant, *p < .5, **p < .01, ***p < .001, ****p < .0001.
For in vivo studies, mice were randomly assigned to treatment groups. All replicates in this study represent different mice.

ACK N OWLED G M ENTS
We are grateful to members of J. Gil's laboratory for reagents,

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.