Identification of Novel Rare ABCC1 Transporter Mutations in Tumor Biopsies of Cancer Patients

The efficiency of chemotherapy drugs can be affected by ATP-binding cassette (ABC) transporter expression or by their mutation status. Multidrug resistance is linked with ABC transporter overexpression. In the present study, we performed rare mutation analyses for 12 ABC transporters related to drug resistance (ABCA2, -A3, -B1, -B2, -B5, -C1, -C2, -C3, -C4, -C5, -C6, -G2) in a dataset of 18 cancer patients. We focused on rare mutations resembling tumor heterogeneity of ABC transporters in small tumor subpopulations. Novel rare mutations were found in ABCC1, but not in the other ABC transporters investigated. Diverse ABCC1 mutations were found, including nonsense mutations causing premature stop codons, and compared with the wild-type protein in terms of their protein structure. Nonsense mutations lead to truncated protein structures. Molecular docking and heat map analyses of ABCC1/MRP1 pointed out that Lys498* appeared in a separate cluster branch due to the large deletion, leading to a massive disruption in the protein conformation. The resulting proteins, which are nonfunctional due to nonsense mutations in tumors, offer a promising chemotherapy strategy since tumors with nonsense mutations may be more sensitive to anticancer drugs than wild-type ABCC1-expressing tumors. This could provide a novel tumor-specific toxicity strategy and a way to overcome drug resistance.


Introduction
ATP-binding cassette (ABC) transporters are drug efflux pumps hampering the effectiveness of cytotoxic anticancer drugs. ABC transporter overexpression is associated with multidrug resistance (MDR) in cancer cells, possibly leading to chemotherapy failure [1]. The overexpression of ABC transporters is also associated with shorter survival rates of cancer patients [2][3][4]. RNA-sequencing and other genome-and transcriptome-wide techniques paved the way to address the MDR phenomenon and the corresponding influence of ABC transporters in a more comprehensive fashion.
The advances in next-generation sequencing methodology have spurred researchers to study a large amount of genomic data to identify patients' tumor-specific mutations. Indeed, this knowledge will assist clinicians in making personalized treatment decisions to improve treatment strategies [5,6]. In recent years, medical oncologists have focused on targeted sequencing to identify genetic aberrations that may be predictive of response to anticancer therapies [7]. While most analyses focused on mutation profiling of the main cell populations of a tumor, the mutations in small subpopulations may be especially decisive for the refractoriness of tumors. Tumor cells with mutations in drug resistance genes may survive and grow if the predominant tumor population has been eradicated by chemotherapy. Tumor heterogeneity within a patient has been recognized as a major factor of drug resistance [8,9]. It is worth mentioning that the advancement of genome-wide DNA mutation analyses have enabled the identification of hundreds of low-frequency mutated genes in tumors and triggered scientists to unravel their impact on the prognosis and pathogenesis of cancer [10]. Structural comparisons of truncated forms of ABC transporters with wild-type proteins have shed light on the effect of these nonsense mutations. Nonsense mutations were clearly associated with truncated protein structures. Tumors carrying nonsense mutations in ABC genes possess nonfunctional proteins for the corresponding ABC transporter. The determination of truncated nonfunctional ABC transporters may imply a promising chemotherapy strategy. Tumors with nonsense mutations in ABC transporters may be sensitive to chemotherapy with ABC transporter substrates. While tumors carry a nonsense-mutated ABC transporter, this transporter is not mutated in normal tissues and is still intact. Hence, chemotherapy would preferentially affect tumor tissues with nonsense-mutated and nonfunctional ABC transporters rather than normal tissues. This strategy may trigger a novel tumor-specific chemotherapy strategy to overcome drug resistance. We analyzed low-frequency mutations in 12 ABC transporters associated with drug resistance (ABCA2, -A3, -B1, -B2, -B5, -C1, -C2, -C3, -C4, -C5, -C6, -G2) [11][12][13][14][15]. Novel ABCC1 transporter mutations, including nonsense mutations causing premature stop codons, were identified that have not been reported before.
In the present study, we performed RNA-sequencing in tumors from 16 patients with different tumor types at a late stage who had not responded to conventional chemotherapy and two leukemia patients' biopsies were collected during the initial diagnosis (n = 18 in total). We specifically focused on low-frequency mutations. Additionally, we identified novel nonsense and missense mutations in the ABCC1 gene and speculate that substrates of MDR-associated protein 1 (MRP1, encoded by the ABCC1 gene), such as doxorubicin, docetaxel, etoposide, and teniposide could be administered to patients with nonsense ABCC1 mutations. Furthermore, we selected three missense and one nonsense mutations, in order to evaluate the binding mode of MRP1 substrates and inhibitors. By applying heat map analyses, we compared the binding patterns with those of wild-type MRP1.

RNA Sequencing and Mutation Analysis
The ABC transporter mutations in our dataset of 18 patients with various cancer types were identified by RNA sequencing. Informed consent was collected from all patients. The procedure of RNA sequencing has been described previously [16]. The clinical data of the patients is described in Table 1. Considering frequent mutations, none of the patients possess nonsense mutations. In order to identify the low frequent mutations, Strand NGS 3.4 software (Strand Life Sciences Pvt. Ltd., Bangalore, India) was used. Twelve ABC transporters together with their chromosomal position were selected and imported as a gene list. As a first step, the patients' ".vcf" files and a ".bam" file as a reference human genome alignment were imported. Then, by using the "filter by region list" option, read lists (aligned reads) and region lists (patient data) were selected to generate a further read list. The next round of "filter by region list" was performed by selecting the read list from the previous step and the imported ABC transporter gene list as the region list. This final read list was used to perform low-frequency SNP detection by clicking on "SNP detection" and "perform low frequency SNP detection" with default options. Default lower threshold of the base quality range for the binomial test iteration is 20 and default upper threshold of the base quality range for the binomial test iteration is 30 for low-frequency SNP detection. Detailed explanation for low-frequency SNP detection is listed at the user manual Cells 2020, 9, 299 3 of 20 Section 11.5.4 of Strand NGS software. We took the same threshold for low-frequency mutations. Afterwards, SNP effect analysis was performed, and the gene lists and the mutations were exported.

Homology Modeling, Protein Structure Preparation
A human MRP1 homology model was created by using bovine Mrp1 (PDB ID: 5UJ9) as template. The Modeller 9.23 algorithm embedded in UCSF Chimera software was used as previously described [17]. Structural comparisons were performed for human MRP1 focusing on nonsense mutations which cause premature stop codons. Corresponding protein structure files were prepared by using the human MRP1 homology model.

Molecular Docking and Clustering
The modeled wild-type MRP1, three missense mutations (T550A, T556A, and V1101F) and one nonsense mutation (Lys498*) were subjected to molecular docking analyses against a panel of MRP1 substrates (zoledronic acid, irinotecan, etoposide, epirubicin, doxorubicin, docetaxel, dactinomycin, 7-ethyl-10-hydroxycamptothecin, and camptothecin) and MRP1 inhibitors (reversan, mk-571, glibenclamide, pak104p, sulfinpyrazone, indomethacin, probenecid, quercetin, genistein, and diltiazem) [18][19][20][21][22][23][24][25][26][27][28]. The two missense mutations, T550A and T556A, were hotspot mutations in the transmembrane domain. Thr550 and Thr556 were located in the inner leaflet region of transmembrane domain 10 (TM10), playing a vital role in determining the drug resistance profile [29]. The V1101F mutation was also observed in COSMIC database, present in the HCT15 colon cancer cell line. Lys498* was the nonsense mutation observed in our patients leading to the largest deletion in MRP1. A total of four human MRP1 homology models, 9 classical anticancer drugs that are MRP1 substrates, and 10 MRP1 inhibitors have been subjected to an automated and comprehensive virtual molecular docking campaign. Each molecular docking was based on at least three independent dockings, each consisting of 2,500,000 calculations. The AutoDock 4 algorithm was used for the defined molecular docking calculations on the drug-binding pocket of MRP1 as previously described [30].
Hierarchical clustering was performed with the CIM miner software (https://discover.nci.nih. gov/cimminer/home.do) (Bethesda, MD, USA) as previously described by applying ward cluster algorithm [31,32]. The heat maps were visualized in order to classify the compounds and the mutant models in terms of their binding energy distributions.

ABC Transporters Mutation Analysis
Low-frequency missense and nonsense mutations were only observed for ABCC1, but not for the other 11 ABC transporters investigated. These low-frequency nonsense mutations in ABCC1 are listed in Table 2. Low-frequency missense and deletion/insertion mutations in ABCC1 are listed in Table 3. All identified nonsense mutations in our patient dataset are new and were not listed in the COSMIC database (https://cancer.sanger.ac.uk/cosmic/gene/analysis?ln=ABCC1#variants). Therefore, they can be considered as novel mutations.         For visualization of these low-frequency nonsense mutations, their read values and the corresponding alignments are depicted in Figure 1. For visualization of these low-frequency nonsense mutations, their read values and the corresponding alignments are depicted in Figure 1. Figure 1. Low-frequency nonsense ABCC1 mutation sequence reads and the corresponding alignments. Reference genome sequence (gray) is on top of each screenshot. Sequence reads from the tumor biopsy with different clusters are labeled in green. Read numbers of each cluster sequence for that region are shown on the right side of each screenshot. On the bottom right side, the coverage percentage for the shown cluster reads is depicted for that region. One patient for each mutation was selected for visualization.

Structural Comparison of the Nonsense MRP1 Mutant Models
Low-frequency ABCC1 nonsense mutation caused truncated MRP1 protein structures. A structural comparison of each low-frequency nonsense mutant with the wild-type MRP1 structure is visualized in Figure 2. Reference genome sequence (gray) is on top of each screenshot. Sequence reads from the tumor biopsy with different clusters are labeled in green. Read numbers of each cluster sequence for that region are shown on the right side of each screenshot. On the bottom right side, the coverage percentage for the shown cluster reads is depicted for that region. One patient for each mutation was selected for visualization.

Structural Comparison of the Nonsense MRP1 Mutant Models
Low-frequency ABCC1 nonsense mutation caused truncated MRP1 protein structures. A structural comparison of each low-frequency nonsense mutant with the wild-type MRP1 structure is visualized in Figure 2.

Molecular Docking Clustering
Hierarchical clustering and heat maps are visualized for MRP1 inhibitors in Figure 3A and substrates in Figure 3B. For both inhibitors and substrates, Lys498* appeared in a separate cluster branch as expected due to the large deletion, leading to a massive disruption in the conformation. T550A and T556A mutants clustered together on the heat map for both inhibitors and substrates, indicating that both mutants have the same pattern of conformation. Interestingly, the V1101F mutant clustered with the wild type for inhibitors and clustered in close distance for substrates, suggesting this mutant has little effect on the MRP1 transport function. The inhibitor Mk-571 showed higher affinity to the mutants than the wild type. Glibenclamide and pak104p revealed lower binding energies for the T550A and T556A mutants than for the wild type and V1101F mutant. Intriguingly, Figure 2. Influence of nonsense mutations on the ABCC1 gene and structural comparison with wild-type MRP1 protein structure. Wild-type MRP1 protein (green) and mutant MRP1 homology models (red).

Molecular Docking Clustering
Hierarchical clustering and heat maps are visualized for MRP1 inhibitors in Figure 3A and substrates in Figure 3B. For both inhibitors and substrates, Lys498* appeared in a separate cluster branch as expected due to the large deletion, leading to a massive disruption in the conformation. T550A and T556A mutants clustered together on the heat map for both inhibitors and substrates, indicating that both mutants have the same pattern of conformation. Interestingly, the V1101F mutant clustered with the wild type for inhibitors and clustered in close distance for substrates, suggesting this mutant has little effect on the MRP1 transport function. The inhibitor Mk-571 showed higher affinity to the mutants than the wild type. Glibenclamide and pak104p revealed lower binding energies for the T550A and T556A mutants than for the wild type and V1101F mutant. Intriguingly, epirubicin showed higher affinity to the nonsense-mutated and truncated MRP1 than the wild type and the other mutants. This could be attributed to a unique binding site that might be reachable for epirubicin in the deleted form of MRP1. Cells 2020, 9,299 12 of 16 epirubicin showed higher affinity to the nonsense-mutated and truncated MRP1 than the wild type and the other mutants. This could be attributed to a unique binding site that might be reachable for epirubicin in the deleted form of MRP1.

Discussion
The development of MDR phenotypes is a major obstacle in the treatment of malignancies as it leads to therapeutic failure. Overexpression of ABCC1 is considered as one of the mechanisms that elicits MDR development in cancer cells due to its capacity to expel the chemotherapeutic drugs from the cells [33]. MRP1 localizes primarily at the plasma membrane of the cells and in the membranes of sub-cellular organelles such as mitochondria, endoplasmic reticulum and endocytic vesicles [34]. ABCC1 mRNA overexpression has been reported in small-cell lung carcinoma [35], prostate [36], neuroblastoma [37], acute lymphoblastic leukemia [38], breast [39], and glioblastoma [25].
In the present study, we performed rare mutation analyses for 12 ABC transporters related to drug resistance (ABCA2, -A3, -B1, -B2, -B5, -C1, -C2, -C3, -C4, -C5, -C6, -G2) in a dataset consisting of RNA sequencing data from 18 cancer patients. Novel rare mutations were found only in ABCC1. Low-frequency ABCC1 mutations have been identified, which reflect tumor heterogeneity and the presence of various missense and nonsense mutations at the ABCC1 coding region for each patient. Cases with a large deletion as result of a nonsense mutation were characterized by dramatic structural changes. These kind of nonsense mutations presumably led to aberrant structures and thus

Discussion
The development of MDR phenotypes is a major obstacle in the treatment of malignancies as it leads to therapeutic failure. Overexpression of ABCC1 is considered as one of the mechanisms that elicits MDR development in cancer cells due to its capacity to expel the chemotherapeutic drugs from the cells [33]. MRP1 localizes primarily at the plasma membrane of the cells and in the membranes of sub-cellular organelles such as mitochondria, endoplasmic reticulum and endocytic vesicles [34]. ABCC1 mRNA overexpression has been reported in small-cell lung carcinoma [35], prostate [36], neuroblastoma [37], acute lymphoblastic leukemia [38], breast [39], and glioblastoma [25].
In the present study, we performed rare mutation analyses for 12 ABC transporters related to drug resistance (ABCA2, -A3, -B1, -B2, -B5, -C1, -C2, -C3, -C4, -C5, -C6, -G2) in a dataset consisting of RNA sequencing data from 18 cancer patients. Novel rare mutations were found only in ABCC1. Low-frequency ABCC1 mutations have been identified, which reflect tumor heterogeneity and the presence of various missense and nonsense mutations at the ABCC1 coding region for each patient. Cases with a large deletion as result of a nonsense mutation were characterized by dramatic structural changes. These kind of nonsense mutations presumably led to aberrant structures and thus nonfunctional or only partially functional MRP1 transporters. As a result, MRP1 substrates could not be recognized by the transporter anymore and could remain within the tumor cells without being expelled. This implies a novel personalized chemotherapy approach in those cases, where nonsense mutations in ABCC1 are determined. Tumors with nonfunctional/partially functional MRP1 due to the presence of premature stop codon formations may be more sensitive towards anticancer substrates of MRP1, but normal tissues expressing wild-type ABCC1/MRP1 are not sensitized at the same time, because these nonsense mutations occur as somatic mutations only in tumor cells, but not as germline mutations in normal tissues. The identification of ABCC1/MRP1 nonsense mutations, therefore, could be critical in this regard and could take place on an individual basis by sequencing each tumor transcriptome.
All identified nonsense mutations in our patient dataset are novel. These results considerably expand upon the findings of Yin et al. [40] in analyzing the role of MRP1 polymorphisms in drug resistance, toxicity, and prognosis prediction. They identified only 14 non-synonymous polymorphisms with very low frequencies, but no nonsense polymorphism has been found [40]. Single nucleotide polymorphisms occur in healthy tissues at variable high percentages and may therefore also occur in tumors derived from these normal tissues. The results we obtained in our analysis are somatic mutations in tumors. Saito [43]. In our study, we focused on rare missense, del/ins, and nonsense mutations. We identified 14 novel, rare ABCC1 nonsense mutations (Table 2) and 301 novel, rare ABCC1 missense and del/ins mutations ( Table 3).
As shown by our molecular docking results, the truncated ABCC1/MRP1 forms revealed different patterns of drug binding than the wild type and other mutants for specific MRP1 inhibitors and substrates, because they clearly appeared as a separate cluster. Importantly, the known MRP1 inhibitor, Mk-571, showed higher affinity to missense mutations than wild-type MRP1. Combining MRP1 chemotherapeutic substrates with Mk-571 might be a favorable therapeutic option for patients whose ABCC1 gene bears these missense mutations.
The two polar residues, Thr550 and Thr556, located in the inner leaflet region of TM10, play a vital role in determining the drug resistance profile [29]. Both residues were mutated and led to an amino acid exchange and placement of alanine. Increased resistance to vincristine and decreased resistance to doxorubicin and etoposide were observed in the T550A and T556A mutations which was attributed to the smaller size of the Ala side chain that favors transport of the larger drug [29]. Moreover, our heat map analyses based on the molecular docking results clearly showed that these two missense mutations were clustered together and had the same pattern of substrate/inhibitor bindings. To the best of our knowledge, this is the first time to report those two missense mutations in clinical samples. Since both mutations are determinants for drug resistance, this warrants further care in the selection of chemotherapeutic agents for patients bearing these mutations.
In conclusion, a common assumption is that the low-frequency-mutation analyses may have more prognostic significant values for driver mutations [44]. Hence, mutations in driver genes have been mainly in the spotlight. In the present investigation, we intentionally did not focus on driver mutations that are important for carcinogenesis and tumor progression. Instead, we concentrated on ABC transporters as important mediators of chemotherapeutic drug resistance. In fact, we found that pointing to nonsense mutations in ABC transporters may offer a unique opportunity for precision medicine. Nonsense mutations in ABC transporters lead to truncated and nonfunctional proteins offering unique chances to sensitize tumor patients to chemotherapy. RNA-sequencing of tumor transcriptomes on an individual patient-to-patient basis may, therefore, unravel opportunities to successfully treat otherwise refractory cancer patients with drugs that are substrates of specific ABC transporters. Furthermore, rare mutations contribute to intratumoral heterogeneity and sub-clonal diversity for primary and metastatic tumors. Therefore, identification of rare mutations in ABC transporters, coupled with knowledge of protein structure and functional insight, may allow the design of novel and appropriate strategies for individualized cancer treatment.