RDH10 and ATRA sensitize triple-negative breast cancer to taxane-based chemotherapy through regulation of PIN1

Background: With few targeted therapies available for triple-negative breast cancer (TNBC), there is a critical need to identify biomarkers that can predict chemotherapy-response. Copy number amplications (CNAmps) have been implicated in tumorigenesis, progression, and treatment response. Previously, we performed a prospective neoadjuvant chemotherapy study (BEAUTY) and identied a 5 Mb CNAmp in TNBC patients associated with pathological complete response rates to taxane- and anthracycline-based chemotherapy. Here we further interrogated this amplicon and its role in chemotherapy response. Methods: Using siRNA screening of genes identied in the 5 Mb CNAmp, followed by cytotoxicity assays, we identied RDH10 signicantly associated with chemo-sensitivity in TNBC cell lines. Functional studies were performed using RDH10 knockdown or overexpression followed by cytotoxicity, cell proliferation, luciferase reporter assays, HPLC or western blot. Results: RDH10 is a rate-limiting step involved in the synthesis of all-trans retinoic acid (ATRA) - the active metabolite of retinoid metabolism that is frequently aberrant in cancer. Previous studies demonstrated ATRA to be a potent inhibitor of PIN1, a key regulator of several oncogenic signaling pathways often amplied in TNBC. We discovered that genetic manipulation of RDH10 affected ATRA intracellular concentrations, contributing to PIN1 degradation and taxane-response. Furthermore, co-treatment of TNBC cells with ATRA signicantly increased taxane-sensitivity. Conclusion: RDH10 amplication increases endogenous levels of ATRA and degrades PIN1 in TNBC. ATRA-mediated degradation of PIN1 sensitizes TNBC to chemotherapy, suggesting that RDH10 may be a novel biomarker of taxane response and the addition of ATRA to standard chemotherapy may improve chemo-sensitivity. mRNA levels were quantied by qRT-PCR using Power SYBR Green RNA-to-C T 1-Step Kit (Applied Biosystems by Thermo Fisher Scientic) and PrimeTime (IDT) pre-designed qPCR Primers. Gene expression analyses were performed using the 2 -ΔΔCt method and the average Ct value of b-actin and GAPDH was used as the internal reference genes. Three independent experiments were performed. drug treatment and absorbance was TECAN plate reader excitation 485 nm and emission detection at 520 nm. Results were normalized to DMSO control and plots were generated using the GraphPad PRISM software. For cell proliferation assay, 10 uL Cyquant was added to wells every 24 and absorbance was measured as above. Results were normalized to day 1 and were generated using GraphPad PRISM software.

overlap, generating putative CNA "hotspots" (7). Some "hotspots" have been reported to be strongly associated with clinical phenotypes resulting in the discovery of novel therapeutic options for many cancer types (8)(9)(10). While these regions encompass canonical oncogenes (e.g., MYC) or tumor suppressors (e.g., TP53, ATM, CDKN2A/B), many other genes within these regions may play a role in disease phenotype, such as drug response (11). Studying genes within CNAs identi ed from TNBC patient tumors may lead to the identi cation of novel treatment response biomarker(s) and provide the opportunity for therapeutic interventions for tumors resistant to standard chemotherapy.
The Breast Cancer Genome-Guided Therapy Study (BEAUTY) is a prospective neoadjuvant clinical study with a 132 patient cohort of women diagnosed with Stage I-III breast cancer (12). The objective of this study is to identify genomic alterations and perturbed pathways associated with patient response or resistance to chemotherapy. Therefore, Whole Exome Sequencing (WES) and RNA sequencing (RNAseq) were performed on patient tumor and germline biopsies at baseline, during, and following completion of neoadjuvant chemotherapy. Out of this cohort (n=132), approximately one-third (n=44) were diagnosed with TNBC. These patients received 12 weeks of taxane treatment, followed by 4 cycles of anthracycline/cyclophosphamide. Following surgery, patients (n=41) were strati ed based on pathological complete response (pCR) as either responders (achieving pCR; n=22) or non-responders (non-pCR; n=19) to the treatment. Association studies with outcomes were then performed using the sequencing data. Previously, we determined that 31.8% (n=7) of TNBC responders shared a 5 Mb copy number ampli cation (CNAmp) on chromosome 8q, containing 29 genes (12). Here we present a potential novel biomarker, Retinol Dehydrogenase 10 (RDH10) associated with response to paclitaxel in TNBC. Based on our understanding of RDH10 involvement in chemo-response, we also present a potential new therapeutic approach to increase sensitivity to taxane treatment in TNBC. RNA interference. For siRNA transfection, cells were forward transfected with the indicated siGENOME SMARTpool siRNA (Horizon Discovery) using Lipofectamine2000 (Life Technologies Invitrogen) for BT549 and Hs578t or TransfeX (ATCC) for MDA-MB-231 transfection reagents following the manufacturer's protocol. mRNA quanti cation. RNA was puri ed using the Zymo Research Quick-RNA MiniPrep Kit (Genesee Scienti c). For each reaction, 100 ng of total RNA was used for ampli cation of each target gene. mRNA levels were quanti ed by qRT-PCR using Power SYBR Green RNA-to-C T 1-Step Kit (Applied Biosystems by Thermo Fisher Scienti c) and PrimeTime (IDT) pre-designed qPCR Primers. Gene expression analyses were performed using the 2 -ΔΔCt method and the average Ct value of b-actin and GAPDH was used as the internal reference genes. Three independent experiments were performed.

Methods
Cytotoxicity and cell proliferation assay. For siRNA knockdown or overexpression studies, 1x10 5 and 2x10 5 cells/ well of BT549 and MDA-MB-231, respectively, were seeded in 6 well plates to achieve 80-90% con uence. The following day, cells were transfected with plasmids or siRNA using Lipofectamine2000 (for BT549) or TransfeX (for MDA-MB-231). Twenty-four hours after transfection, cells were trypsonized and reseeded at a cell density of 1000 cells/90 uL/well (for BT549, Hs578t) or 2000 cells/90 uL/well (for MDA-MB-231) in 96 well plates in triplicate. 10 uL of increasing concentrations of paclitaxel or epirubicin were added to wells the following day. For ATRA treatment studies, ATRA was dissolved in DMSO at a stock concentration of 100 mM and diluted in culture medium prior to use. Stocks were stored at -80 C for one week. Fresh stock of ATRA was created prior to media changes. ATRA at indicated concentrations were then added to BT549 and MDA-MB-231 when seeding in 96 well at-bottomed plates in triplicate. The following day, cells were treated with increasing concentrations of paclitaxel or epirubicin for 72 hours. For assessing cytotoxicity or proliferation, 10 uL Cyquant was added to the wells after appropriate drug treatment time and absorbance was measured using TECAN plate reader with excitation at 485 nm and emission detection at 520 nm. Results were normalized to DMSO control and plots were generated using the GraphPad PRISM software. For cell proliferation assay, 10 uL Cyquant was added to wells every 24 hours and absorbance was measured as described above. Results were normalized to day 1 and were generated using GraphPad PRISM software.
Western blot analysis. Cell lines were lysed in Pierce RIPA lysis buffer (Thermo Scienti c) containing phosphatase and proteasome cocktail inhibitors according to manufacturer's instructions. PDX tumors were mixed with NETN lysis buffer (100mM NaCl, 20mM Tris·HCl pH = 0.8, 0.5mM EDTA, NP-40) containing inhibitors and 0.5 mm RNase-free zirconium oxide beads in micro-centrifuge tubes and placed into a bullet blender for 10 minutes at 4 C. Lysates were then normalized via Bradford assay and mixed with Tris-Tricine SDS sample buffer containing 2% 2-ME and boiled for 10 minutes. Equal amounts of protein were subjected to electrophoresis in TGX Tris-Tricine SDS gels and were transferred to PVDF membranes. Membranes were blocked in TBST (1X TBS with 0.1% Tween-20) with 5% non-fat milk powder for one hour. Membranes were washed with TBST and then incubated overnight at 4 C with indicated primary antibodies, followed by anti-mouse IgG or anti-rabbit IgG secondary antibody (1:2000) for one hour at room temperature. All blots were visualized with Supersignal WestPico chemiluminescent ECL kits (Thermo Fisher) and image was documented by the Gel documentation system (Biorad).
Western blot quanti cation was performed using the volume tool in the Image Lab software. Adjusted volume values were normalized to the internal standard (Vinculin, a-Tubulin, or b-Actin) then normalized back to control (shNT or EV).
Generation of stable cell lines. To establish stable cell lines, cells were infected by lentivirus. RDH10 shRNA clones V2LHS_267926 (shRNA 1) and V2LHS_100112 (shRNA 2) (Open Biosystems) and PIN1 shRNA clones NM_006221.x-616s1c1 (shRNA 1) and NM_006221.x-420s1c1 (shRNA 2) (Sigma-Aldrich) were utilized in this paper. Virus was generated in HEK-293T using the pMD2G and psPAX2 viral packaging vectors. Supernatant concentrated using the Takara Lenti-X Concentrator were then added to MDA-MB-231, BT549 and Hs578t cell lines with 8 ug/mL polybrene for 72 hours followed by puromycin selection for 2 weeks. Individual clones were then isolated and knockdown e ciency was veri ed by western blot.
Generation of overexpression constructs. Human RDH10 cDNA (Gene ID: 157506) was purchased from Horizon Discovery. cDNA was ampli ed by PCR and sub-cloned into pENTR shuttle vector using pENTR/D-TOPO cloning kit (Thermo Fisher). pENTR-RDH10 construct was subcloned into pcDNA3. Chromatographic separation of all-trans retinoic acid was achieved on a LiChrosorb® RP 18-5 (Supelco, Bellefonte, PA) C18 reversed-phase column (250 mm × 4.6 mm, 5 uM). The mobile phase consisted of acetonitrile: methanol: 2% ammonium acetate: glacial acetic acid (79:2:16:3, v/v at a ow rate of 1.0 mL/min at 25 °C coupled with a Phenomenex Security Guard™ C18 guard column 4 × 3 mm (Phenomenex, Torrance, CA). The mobile phase was degassed as well as vacuum ltered through 0.22 um nylon membranes before use. The HPLC system was equipped with the diode array detector SPD-M20A (Shimadzu Corporation, Kyoto, Japan). 100 uL injections were used to analyze all-trans retinoic acid by UV absorption at 350nm. Data were acquired and processed with LabSolutions software version 5.85 (Shimadzu Corporation, Kyoto, Japan). Peaks from samples were then identi ed and quanti ed by comparison with the all-trans retinoic acid standard (Sigma, St. Louis, MI).

KM-Plotter database Association between the mRNA expression levels of DAPK1 and Relapse-Free
Survival (RFS) was analyzed using the online database, KM-plotter (kmplot.com) using the gene expression data and the survival information of patients with breast cancer using all available datasets (15). Cohorts of patients were split by median expression. For subtype speci cation, the restriction analysis were default ("all") with the following exceptions: ER+ subtype: ER status-ER-positive; HER2+ subtype: HER2 status-HER2 positive; TNBC subtype: ER status-ER-negative, PR status-PR negative, HER2 status-HER2 negative.
Statistical analysis Statistical tests were performed using GraphPad Prism Software; All gures presented have been replicated by 3 independent experiments and are reported as the mean ± SD from triplicate data of one experiment, followed by determining signi cant differences using the two-tailed student t-test where *P< 0.01, **P< 0.001, and ***P< 0.0001 unless otherwise stated. The Kaplan-Meier survival plots with the number at risk, hazard ratio (HR) and log-rank p-values were obtained using the KM plotter website. Log-rank P-values <0.05 was used to indicate a statistically signi cant difference.
Results siRNA cytotoxicity screen reveals RDH10 association with taxane response in TNBC We previously identi ed a 5 Mb copy-number ampli cation on chromosome 8q (chr8q13.1-.3 and chr8q21.11-.12) in TNBC tumors associated with pCR (12). This region contains 29 genes. We rst removed any RNA-coding genes and genes without any known function, leaving 20 protein-coding genes ( Figure 1A). We next utilized RNA sequencing data available in TNBC cell lines to eliminate genes with low baseline expression using an RPKM threshold of 0.1, which corresponds to a false-discovery rate (FDR) and false-negative rate of 5% (16)(17)(18), narrowing our list to 17 genes ( Figure 1A). Finally we eliminated any genes that had poor knockdown e ciency (≤50%) in both TNBC cell lines BT549 and MDA-MB-231 as determined by qRT-PCR, leaving 10 genes for functional study. To identify predictive biomarkers of treatment response to chemotherapy, siRNA cytotoxicity screening was performed in TNBC cell lines, BT549 and MDA-MB-231, with taxane (paclitaxel) or anthracycline (epirubicin)-two classes of drugs used in the BEAUTY study (12) as part of the neoadjuvant chemotherapy regimen ( Figure 1B). Following this screening, four genes showed a more resistant phenotype to paclitaxel compared to control (p-value< 0.05) in either BT549 or MDA-MB-231 cells ( Figure 1C, Supplementary Table 1). Only RDH10 showed a consistent phenotype in both cell lines.

RDH10 expression regulates cell proliferation, taxane sensitivity and intracellular ATRA in TNBC
We rst utilized RNAseq of TNBC cell lines to select three cell lines with varying expression of RDH10 (high (BT549), medium (Hs578t), low (MDA-MB_231)), con rming their expression levels using qRT-PCR and western blot (Figure 2A). To study the role of RDH10 in chemo-response, we generated stable knockdowns in TNBC cell lines BT549 and Hs578t, using two individual shRNA constructs targeting RDH10 ( Figure 2B, C). In addition, we performed transient overexpression of RDH10 in BT549 and MDA-MB-231 ( Figure 2D, E). Hs578t was utilized for knockdown studies and MDA-MB-231 was utilized for overexpression studies as each had higher and lower endogenous levels of RDH10, respectively. We utilized these lines to determine TNBC response to paclitaxel and epirubicin. We found that loss of RDH10 only increased resistance to paclitaxel in both BT549 and Hs578t with no effect observed with epirubicin ( Figure 2B, C). Conversely, overexpression of RDH10 only increased sensitivity to paclitaxel, with no effect with epirubicin ( Figure 2D, E). We also determined that loss of RDH10 increased cell proliferation ( Figure 2B, C) while overexpression of RDH10 decreased cell proliferation ( Figure 2D, E).
RDH10 is the rate limiting step involved in the synthesis of all-trans retinoic acid (ATRA) -the active metabolite in retinoid metabolism ( Figure 3A). Since RDH10 is one of over 20 RDH isoforms with known catalytic activity (19), we sought to determine the speci c impact RDH10 had on the intracellular concentrations of the RDH precursor, retinol (ROL), and the active metabolite ATRA. We utilized HPLC-UV detection to quantify the endogenous levels of ROL and ATRA in whole cell lysates. We found that following knockdown of RDH10 in BT549 and Hs578t, intracellular ROL increased, while ATRA was reduced by 40-60% ( Figure 3B, C), while overexpressing RDH10 in BT549 and MDA-MB-231 resulted in a decrease in ROL and a four-fold increase in intracellular ATRA ( Figure 3D, E). These results were further con rmed by determining the ATRA effect on gene transcription through retinoid receptor using a luciferase reporter assay in cells transfected with a luciferase reporter construct with an upstream retinoic acid response element (RARE-luc) (14).
Endogenous ATRA levels can also be regulated through intracellular metabolism by auto-inducible cytochrome P450s (CYPs) (20). While CYP3A4, CYP2C8, CYP26A1 and CYP26B1 contribute to systemic ATRA clearance (21), CYP26A1 and CYP26B1 are also expressed in extrahepatic tissues and in uence tissue-speci c intracellular ATRA concentrations (20,22,23). To determine if ATRA produced by RDH10 overexpression affected its metabolism, we quanti ed the expression of ATRA-metabolizing CYPs in TNBC cell lines. Utilizing a strict Ct-value cut-off of 30 (24), we found only CYP26A1 and CYP26B1 to be expressed in BT549 and MDA-MB-231 cells, respectively ( Figure 3F). Interestingly, overexpression of RDH10 signi cantly down-regulated transcripts of CYP26A1 in BT549, and CYP26B1 in MDA-MB-231 ( Figure 3F). PIN1 loss increases taxane sensitivity and decreases cell proliferation similar to RDH10 overexpression in TNBC ATRA was previously identi ed through a large-scale, small-molecule screen to potently bind to and cause the ablation of peptidyl-prolyl cis/trans isomerase, NIMA interacting-1 (PIN1) (25). PIN1 is often ampli ed in aggressive forms of cancer, such as TNBC, and is involved in the cis-trans isomerization of phosphorylated serine/threonine-proline residues-activating over 50 oncogenes and inhibiting more than 20 tumor suppressors (26). To further study the mechanism of the RDH10 effect on chemo response, we sought to test the role of PIN1 in this process. First, we knocked down PIN1 in two TNBC cell lines (Supplementary Figure 1A), followed by treatment with increasing doses of paclitaxel or epirubicin. We found that PIN1 knockdown decreased cell proliferation (Supplementary Figure 1C) and increased sensitivity of BT549 and MDA-MB-231 to paclitaxel, but not epirubicin (Supplementary Figure 1B); a phenomenon that was very similar to that of RDH10 overexpression ( Figure 2D, E).

RDH10 affects TNBC taxane response and cell proliferation through PIN1
To determine if the ability of RDH10 to modulate intracellular ATRA concentration and response to chemotherapy is through PIN1, we determined endogenous PIN1 levels after knocking down RDH10. We found that when RDH10 was knocked down and intracellular ATRA levels were low ( Figure 3B, C), PIN1 levels increased in BT549 and Hs578t ( Figure 4A, B), while overexpression of RDH10 increased intracellular ATRA levels ( Figure 3D, E) and decreased PIN1 levels ( Figure 4C, D); consistent with previous ndings that ATRA can degrade PIN1 (25). Furthermore, using TNBC PDX tumor lysates obtained from the BEAUTY patients with either copy number ampli cation on chromosome 8q or those without, we found that tumors with 8q ampli cation exhibited higher levels of RDH10 and lower levels of PIN1 compared to those without 8q ampli cation ( Figure 4E).
To further determine if RDH10 effect on cell proliferation and chemo-response is through PIN1 regulation and if PIN1 activity is required for this process, we generated PIN1 stable knockdown cell lines (Supplementary Figure 2) and then introduced either ag-tagged wild-type PIN1 or catalytically inactive S71D PIN1 (27), followed by overexpressing RDH10 (Figure 5). In contrast to WT PIN1, overexpression of the catalytic inactive S71D PIN1 cannot rescue cell proliferation decreased by endogenous PIN1 knockdown. Furthermore, in our experiment, we found the S71D mutation is insu cient to inhibit ATRA from binding and degrading PIN1 which is differs from observations reported previously (25). Additional studies are required to better understand the function of this mutation. Our results, nevertheless, showed that RDH10 decreased PIN1 levels, leading to decreased cell proliferation and increased TNBC sensitivity to paclitaxel ( Figure 5; Supplementary Table 2). This suggests that the cell proliferative and cytotoxic effects of RDH10 act through regulation of PIN1 and the effect is also dependent on PIN1 activity.
ATRA has previously been used in the clinic in combination with chemotherapy for the treatment of acutepromyelocytic leukemia (APL), with complete remission rates greater than 90% (28). To determine if ATRA would increase the chemo-sensitivity in TNBC, we rst tested the effect of ATRA alone, and found no evidence for cytotoxic effects on BT549 or MDA-MB-231 until concentrations reached as high as 100 uM ( Figure 6A). We next determined the effect of ATRA on PIN1 levels with the assumption that PIN1 level is a pharmacodynamic marker for ATRA drug effect. When treated with 1 uM and 10 uM ATRA in BT549 and MDA-MB-231, PIN1 levels decreased ( Figure 6B). We next performed HPLC on whole cell lysates of cells treated with 1 uM and 10 uM ATRA. We found with 1 uM ATRA treatment, BT549 exhibited three times greater endogenous levels of ATRA compared to MDA-MB-231. Previous reports have shown that epithelial cells treated with retinoid-receptor agonists, including ATRA, can upregulate gene expression of RDH10 (29). We observed RDH10 to also be upregulated in BT549 with a lesser effect in MDA-MB-231 ( Figure 6B), suggesting ATRA-induced RDH10 expression could contribute to the observed differences in intracellular ATRA concentrations and PIN1 levels with 1 uM ATRA treatment between cell lines. However, both cell lines achieved similar endogenous ATRA concentrations with 10 uM ATRA regardless of RDH10 expression. We also noted that 10uM ATRA treatment resulted in similar intracellular ATRA levels that were achieved when RDH10 was overexpressed in both cell lines ( Figure 3D, E).
To determine if ATRA treatment could increase sensitivity of TNBC to chemotherapy, we co-treated BT549 and MDA-MB-231 with various concentrations of ATRA and paclitaxel or epirubicin. As expected, addition of 1 or 10 uM ATRA had limited effect on epirubicin sensitivity ( Figure 6E); However, there was a dose dependent sensitization to paclitaxel and a decrease in cell proliferation, consistent with that of PIN1 knockdown (Supplementary Figure 1C) or RDH10 overexpression ( Figure 2D, E).

Discussion
Aberrant retinoid metabolism has been suggested to be a key factor driving tumorigenesis (30). Compared to healthy surrounding tissues, dramatically decreased gene expression of retinoid metabolizing enzymes and transporters corresponding with signi cantly lower endogenous ATRA concentrations has been observed in non-small cell lung (31), ovarian (13), breast (32,33), colon (34), and prostate cancers (35). However, few studies have demonstrated the role of aberrant retinoid metabolism in drug-response. Previously, Guo et al. showed that elevated expression levels of several genes involved in alcohol metabolism that also play a role in retinoid metabolism were associated with decreased mortality in gastric cancer patients receiving 5-FU-based chemotherapy (36). Our study demonstrates for the rst time in TNBC cell lines that the rate-limiting enzyme in retinoid metabolism, RDH10, signi cantly affects the intracellular concentration of ATRA (Figure 3), leading to altered cell proliferation and sensitivity to taxane chemotherapy ( Figure 2). These ndings were further con rmed through the addition of ATRA to cell culture media ( Figure 6). Interestingly, Resetkova et al. revealed an association between gene expression of another retinoid metabolizing enzyme, ALDH1, in the tumor microenvironment of TNBC patients and outcome after chemotherapy treatment. This study noted that while ALDH1 gene expression in tumor cells was not associated with overall or disease-free survival in TNBC patients, high stromal ALDH1 expression was signi cantly associated with outcome (37). Since ATRA is inherently lipophilic, we hypothesize with our in-vitro ndings (Figure 6), that ATRA produced in the microenvironment could have diffused into the surrounding tumor cells, contributing to TNBC chemoresponse for these patients. Of course future work is needed to con rm this hypothesis.
ATRA, in combination with chemotherapy, has been shown in randomized trials to improve long term survival and is the current standard treatment in the clinic for acute promyelocytic leukemia (APL), with complete remission rates of 90% (28). While it is widely thought ATRA binds to and degrades the PML-RARα fusion protein in APL, affecting downstream targets Cyclin D1 and NF-kB, inducing cellular differentiation and treating APL, the use of a pan-RAR agonist failed to mimic ATRA's effects. Work conducted by Wei et al. identi ed a novel non-canonical target for ATRA-PIN1 (25). PIN1 is highly ampli ed in aggressive forms of cancer, such as TNBC, and is involved in the cis-trans isomerization of phosphorylated serine/threonine-proline residues-activating over 50 oncogenes and inhibiting more than 20 tumor suppressors (26). PIN1's ability to bind to and isomerize these proteins can affect their structure and function, acting as an "on-off" switch for their activity. Low levels of PIN1 have been associated with favorable clinical outcome in many human cancers, including breast (38), and PIN1 has been suggested as an attractive non-toxic target to simultaneously block multiple cancer-driving pathways (39). While PIN1 ablation by ATRA has been shown to inhibit TNBC tumor growth in-vivo (25), its role in chemoresponse in TNBC has not been identi ed. Activation of NF-kB (40,41), EGFR (42)(43)(44), Mcl-1 (45)(46)(47), AKT (48), Notch-1/3/4 (49)(50)(51) pathways have all previously demonstrated to play a role in TNBC chemoresistance, and many of these genes are regulated and stabilized by PIN1 (25). Therefore PIN1 loss could restore chemo-response. In our studies, we observed PIN1 levels to be signi cantly lower in TNBC BEAUTY PDX tumors obtained from patients that exhibited taxane treatment response compared to TNBC PDX resistant to taxane therapy ( Figure 4E). We also found that increasing the endogenous levels of ATRA, either through overexpressing RDH10 ( Figure 3D, E) or ATRA supplementation ( Figure 6C) decreased PIN1 levels ( Figure 5, 6B) and increased sensitivity to taxane treatment in TNBC cells ( Figure 5, While PIN1 expression appears to be important to chemo-sensitivity in TNBC, we also showed PIN1 catalytic activity is equally important ( Figure 5). Work conducted by Lee et al. showed that Death-Associated Protein Kinase 1 (DAPK1) directly phosphorylates PIN1 at Ser71, rendering PIN1 catalytically inactive (27). DAPK1 is a well-known tumor suppressor often lost in cancer tissues (52)(53)(54); however high expression of DAPK1 only in the TNBC subtype is associated with greater relapse-free survival (Supplementary Figure 3). Our cytotoxicity results with catalytically inactivate S71D PIN1 suggest that the phosphorylation status of Ser71 PIN1 in addition to PIN1 level is important for taxane sensitivity in TNBC. Future work should include con rming the role of DAPK1 in TNBC taxane sensitivity.
Therapeutic doses of ATRA, as seen in the treatment of APL, result in a peak plasma concentration of 1 uM (55). However, we only observed low PIN1 levels and improved taxane sensitivity at this concentration in BT549. At this concentration, RDH10 levels were found to be three times higher ( Figure 6B) resulting in three times greater intracellular ATRA concentrations ( Figure 6C) in BT549 compared to MDA-MB-231 cells. This suggests that ATRA-induced RDH10 expression could have contributed to intracellular ATRA concentrations to levels su cient to degrade PIN1. Nevertheless, higher concentrations of ATRA resulted in increased taxane cytotoxicity for both BT549 and MDA-MB-231, regardless of RDH10 induction ( Figure  6D). It should be noted that rapid ATRA metabolism has limited its application in the treatment of many oncological malignancies, with the notable exception of APL. This may be due in part to the differences in drug penetrance of liquid and solid tumors. However, preliminary in-vivo work using TNBC cell linederived tumors found ATRA to decrease PIN1 levels at tolerable concentrations (no change in mouse bodyweight) (25). This is promising; however in-vivo work should be conducted to determine if these concentrations of ATRA can be tolerated in combination with taxane treatment. Nevertheless, much effort is on-going to improve ATRA bioavailability including novel drug delivery mechanisms, such as ATRAconjugated aptamers and nanoparticles (56). Furthermore in a phase I clinical trial, patients treated with liposomal ATRA (Atragen) achieved peak plasma concentrations of 25 uM (57). These efforts to improve ATRA bioavailability could bene t NAC treatment response for TNBC patients. Another reason contributing to the differences in ATRA levels between cell lines could be due to differences in cell line metabolism such as ATRA-dependent upregulation of CYPs, reducing the intracellular concentrations of ATRA. Interestingly, we observed that overexpression of RDH10 in-vitro had a signi cant down-regulation in CYP26A1 and CYP26B1 mRNA expression ( Figure 3F), compared with exogenous ATRA treatment (22).
Therefore we hypothesize patients with RDH10 ampli cation not only increase endogenous levels of ATRA, but also inhibit its intracellular metabolism. This nding not only provides a potential rationale for TNBC patient response, but also supports a growing body of evidence of CYP expression-regulating genes that can signi cantly in uence response to drug therapy (58).
Furthermore, retrospective genomic analysis of TNBC patient tumors and correlation with MRI tumor response following NAC revealed that 100% of patients with RDH10 ampli cation achieved pathological complete response compared 44.1% with wild-type levels of RDH10 (Supplementary Table 3). This nding supports RDH10 ampli cation as a strong candidate biomarker for predicting TNBC tumor response to neoadjuvant chemotherapy, and should be validated with additional cohorts.
In conclusion, here we demonstrated the bene t of screening individual genes from CNAmps associated with pCR following chemotherapy in TNBC patients to identify biomarkers and to elucidate additional treatment strategies. Nonetheless, most CNAs are large, containing many genes, and their association with treatment outcome may not be well justi ed by gene-centric approaches (59,60). For example, "passenger" genes may in uence the magnitude of the "driver" gene's effect on drug-response (61,62). Work conducted by Xue et al., demonstrated that simultaneous suppression of genes within chromosome 8p appeared to have a synergistic effect in promoting tumor growth compared to the loss of individual genes (63), suggesting large-scale deletions could create phenotypes that may fundamentally differ from those arising from loss of a single gene. Therefore systematically studying large-scale copy number aberrations in tumors would be a next step to predicting tumor behavior and treatment response, especially in highly heterogeneous diseases such as TNBC. Future generation of cancer cell line models to determine the effect of CNAs will require methods of whole-chromosome manipulation.  Gene selection and cytotoxicity screening of candidate genes in TNBC cell lines. (A) First, all genes with no-known-function, pseudogenes, and non-protein coding genes (e.g. lncRNA, miRNA, etc.) were removed from the 29 genes identi ed within the 5 Mb copy number ampli cation identi ed in BEAUTY. Next, RNA sequencing data of available cell lines were utilized to select the two TNBC cell lines that expressed the majority of the 20 protein-coding genes. Any genes with low baseline cell line expression (RPKM< 0.1)(16-18) were removed from our list, leaving 17 genes. Genes with ≤50% knockdown e ciency determined by qRT-PCR were removed, leaving 10 genes. (B) These genes were knocked down using siRNA in TNBC cell lines (mean +/-SD; *P <0.001) followed by treatment with increasing concentrations of epirubicin or paclitaxel in BT549 and MDA-MB-231. (C) IC50 ratio comparing IC50 in cells with speci c siRNAs targeting genes of interest to siNEG control for paclitaxel response. P-value was calculated using student t-test; Grey bars represent P <0.05; black bars represent P >0.05.