Drug resistance to targeted therapeutics is widespread and the need to identify mechanisms of resistance–prior to or following clinical onset–is critical for guiding alternative clinical management strategies. Here, we present a protocol to couple derivation of drug-resistant lines in vitro with sequencing to expedite discovery of these mechanisms.
Although targeted therapies are initially effective, resistance inevitably emerges. Several methods, such as genetic analysis of resistant clinical specimens, have been applied to uncover these resistance mechanisms to facilitate follow-up care. Although these approaches have led to clinically relevant discoveries, difficulties in attaining the relevant patient material or in deconvoluting the genomic data collected from these specimens have severely hampered the path towards a cure. To this end, we here describe a tool for expeditious discovery that may guide improvement in first-line therapies and alternative clinical management strategies. By coupling preclinical in vitro or in vivo drug selection with next-generation sequencing, it is possible to identify genomic structural variations and/or gene expression alterations that may serve as functional drivers of resistance. This approach facilitates the spontaneous emergence of alterations, enhancing the probability that these mechanisms may be observed in the patients. In this protocol we provide guidelines to maximize the potential for uncovering single nucleotide variants that drive resistance using adherent lines.
Detailed molecular characterization of tumor genomes by robust sequencing technologies and improved data analytical tools has led to the discovery of key genetic alterations in specific cancer types1,2. Development of targeted therapies aimed at these genetic lesions, such as HER2, BCR-ABL, EGFR and ALK, have significantly improved quality of life for patients1,2. However, despite the specificity of this approach, the clinical response to most single therapies has been sub-optimal as resistance ultimately emerges. Recently, significant progress has been made in understanding the molecular underpinnings of resistance to targeted therapeutics. Intriguingly, it is becoming evident that a prominent mechanism of resistance involves persistent target/pathway activity. As a case in point, androgen receptor (AR)-directed enzalutamide treatment of prostate cancer leads to enrichment of activating mutations in AR itself, maintaining AR-signaling output in the presence of the inhibitor3-5. This knowledge has led to an aggressive campaign to 1) develop third generation antagonists that can continue to suppress both WT and mutant-AR function in enzalutamide-resistant PCa3 and 2) identify potential downstream nodes of AR signaling that may be targeted for therapeutic intervention. Similarly, resistance to other classes of inhibitors such as those targeting EGFR, BRAF and ABL often lead to mutations that reactivate the original addicting kinase pathway1.
With the knowledge that resistance inevitably emerges in most patients, developing approaches to expeditiously bring these mechanisms to light will allow development of effective follow-up therapies. One approach that is being widely used is to analyze the genomics of clinical refractory tumors relative to treatment-naïve or -sensitive tumors to identify enrichments/depletions in genetic lesions that may be amenable to drug discovery. Despite its promise, there are two major liabilities of this approach that hamper quick discovery. Firstly, gaining timely access to tumor material for genomic interrogation may serve as a significant hurdle in moving from therapy to cure. Secondly, deconvolution of the myriad of genetic lesions in the resistant setting may be challenging since tumors can present significant intra-tumoral heterogeneity 6,7.
In light of these challenges, there has been increased reliance on preclinical discovery of resistance mechanisms. This approach may allow identification of prominent resistance mechanisms prior to the clinical trials1 that may guide alternative clinical management strategies in those patients that bear these mechanisms either prior to therapy or following onset of resistance.
One such preclinical discovery tool that is being widely used is to apply unbiased functional RNAi screens. For example, Whittaker and colleagues applied a genome-scale RNAi screen to identify that NF1 loss mediates resistance to RAF and MEK inhibitors through sustained MAPK pathway activation8. These findings were found to be clinically relevant as loss-of-function mutations in NF1 were observed in BRAF-mutant tumor cells that are intrinsically resistant to RAF inhibition and in melanoma tumors resistant to vemurafenib activation8. However, despite the success of this approach, many clinically relevant targets are often not identified, presumably due to the loss-of-function bias of this approach.
In contrast, a less biased tool for preclinical discovery of resistance mechanisms involves the generation of resistant cell lines through prolonged exposure to compound of interest coupled with NGS-based genomic or transcriptomic profiling. This approach has been successfully implemented by several groups to identify spontaneous recurrent single nucleotide variants or expression alterations that enable resistance5,9,10. For example, a recurrent F876L mutation in AR was recently discovered in resistant clones in vitro3-5 and in xenograft tumors in vivo5 prior to identification of this mutation in the clinic4. Very recently, Bhang and colleagues (2015)11 used ClonTracer in two clinically relevant models to show that majority of resistant clones that arise during prolonged drug exposure were part of a pre-existing subpopulation suggesting that most functionally relevant mutations are likely already pre-existing that become selected for during selection11.
In contrast to the genomic profiling of tumors discussed earlier, this approach benefits from less heterogeneity as 'homogenous' resistant clones are used for analysis facilitating more accurate genetic dissection of potential drivers. Furthermore, excitingly, in addition to the potential for uncovering mechanisms of resistance, this method can also be applied to identify the cellular mechanisms of action and targets of bioactive small molecules for which this information is unknown10. Given the clear advantages and the multiple uses of this approach, here we present a protocol detailing successful implementation of such a preclinical screen to maximize the potential for clinically meaningful discoveries.
1. Assessing GI50 for Compound(s) of Interest
2. Setting up Drug Resistance Assays
3. Isolating Single Cell Clones
4. Assessing Degree of Resistance of Isolated Clones
5. Next-generation Sequencing
6. Bioinformatics analysis of samples (whole-exome sequencing)
To maximize the potential for discovering key functional drivers of resistance, select single cell clones for expansion, phenotypic testing and sequencing. As illustrated in Figure 4A, HCT116 cells treated for a prolonged period with cytotoxic compound #1 led to the spontaneous emergence of resistant clones that continued to grow during treatment (dashed black circles). These clones were picked using approach #1 highlighted in Figure 3 and subsequently expanded for phenotypic analysis. As shown in Figure 4B, resistant clones 1-3 all showed significant resistance to compound #1 and its close analog compound #2 (greater viability/growth), whereas all clones showed sensitivity to an unrelated cytotoxic compound velcade. Following confirmation of phenotypic resistance, gDNA was isolated and submitted for whole-exome sequencing analysis. Bioinformatics tools were applied to narrow in on those structural variants that are 1) recurrent and 2) have the potential for functional impact (Figure 5A). Structural variants that met these two criteria were confirmed by an independent sequencing tool. As illustrated in Figure 5B, a heterozygous missense mutation that was identified using whole-exome sequencing was confirmed using the Sanger sequencing method (upper panel, WT sequence; lower panel, mutant sequence). Following sequence confirmation, the parental cell line originally used for the resistance assay was genetically engineered to express the mutant cDNA to functionally confirm the role of the mutation. As illustrated in Figure 6, whereas overexpression of the WT cDNA failed to confer resistance, forced expression of the mutant cDNA significantly conferred phenotypic resistance to compound #1, confirming the functional role of this structural variation as a driver of resistance. All reagents used for this experiment are outlined in Table 1.
Figure 1. Layout of assay and control plates. Blue shade, compound treatment wells. White shade, medium only. Compounds are serially diluted 1:4 and are administered in triplicates (B-D or E-G). 2 compounds can be applied per plate.
Figure 2. Representative viability curves. Red dotted line represents the day 0 reading. x-axis indicates increasing doses of compound to the right and y-axis represents viability relative to DMSO control wells. GI100= dose used to achieve 100% growth inhibition; GI50= dose used to reduce viability to 50% of DMSO control.
Figure 3. Two approaches for picking resistant clones for expansion. Approach 1- use pipettor to pick up and transfer well-defined clones to 48-well plates. Approach 2- use trypsin-soaked cloning discs to lift clones and transfer to 48-well plates.
Figure 4. Confirmation of the resistance achieved for HCT116 clones to compound #1 in vitro. (A) Compound #1-resistant HCT116 clones emerged following continuous three-week selection. (B) Viability of control and resistant clones was tested after 72 hr treatment with various compounds. Compound #2 is a close analog of compound #1. Velcade was used as a control cytotoxic agent. Data are shown as average + standard deviation of three biological replicates.
Figure 5. Identification of a unique, recurrent mutation (single nucleotide variants, SNVs) in gene A in compound #1-resistant HCT116 clones. (A) Work flow to identify SNVs present in all resistant clones and predicted to have a high functional impact by MutationAssessor. (B) Confirmation of the mutation in gene A by Sanger Sequencing.
Figure 6. Re-expression of mutated gene A conferred resistance to compound #1 in vitro. Viability of the engineered HCT116 cell lines was tested after 72 hr treatment with compound #1. HCT116-WT or mutant cell lines are stably expressing WT or mutant cDNAs of gene A, respectively. Data are shown as average + standard deviation of three biological replicates.
Sample Name | Amino Acid | Primary Tissue | Zygosity | |
C-33-A | p.E768fs*44 | cervix | Heterozygous | |
C-33-A | p.S860* | cervix | Heterozygous | |
CML-T1 | p.? | haematopoietic_and_lymphoid_tissue | Homozygous | |
CP66-MEL | p.C822F | skin | Heterozygous | |
CTV-1 | p.0? | haematopoietic_and_lymphoid_tissue | Homozygous | |
EFO-27 | p.Q130fs*2 | ovary | Heterozygous | |
EFO-27 | p.? | ovary | Heterozygous | |
HCC2218 | p.E467K | breast | Heterozygous | |
J-RT3-T3-5 | p.R711* | haematopoietic_and_lymphoid_tissue | Homozygous | |
LNCaP | p.? | prostate | Homozygous | |
LoVo | p.? | large_intestine | Homozygous | |
MOLT-13 | p.R711* | haematopoietic_and_lymphoid_tissue | Homozygous | |
NALM-6 | p.? | haematopoietic_and_lymphoid_tissue | Homozygous | |
NCI-H630 | p.R680* | large_intestine | Heterozygous | |
SKUT-1 | p.L787fs*11 | endometrium | Homozygous | |
SKUT-1B | p.L787fs*11 | endometrium | Homozygous | |
SUP-T1 | p.? | haematopoietic_and_lymphoid_tissue | Homozygous |
Table 1. MSH2-mutated Cell Lines. Cell line (sample name), amino acid substitution, the lineage of origin and the zygosity are indicated.
Sample Name | Amino Acid | Primary Tissue | Zygosity | |
CCRF-CEM | p.R100* | haematopoietic_and_lymphoid_tissue | Heterozygous | |
CCRF-CEM | p.? | haematopoietic_and_lymphoid_tissue | Heterozygous | |
CW-2 | p.Y130fs*6 | large_intestine | Homozygous | |
DU-145 | p.? | prostate | Homozygous | |
GR-ST | p.? | haematopoietic_and_lymphoid_tissue | Homozygous | |
HCT-116 | p.S252* | large_intestine | Homozygous | |
IGROV-1 | p.S505fs*3 | ovary | Homozygous | |
MN-60 | p.? | haematopoietic_and_lymphoid_tissue | Homozygous | |
NCI-SNU-1 | p.R226* | stomach | Homozygous | |
P30-OHK | p.? | haematopoietic_and_lymphoid_tissue | Homozygous | |
PR-Mel | p.? | skin | Homozygous | |
REH | p.? | haematopoietic_and_lymphoid_tissue | Homozygous | |
SK-OV-3 | p.0? | ovary | Homozygous | |
SNU-1544 | p.S2L | large_intestine | Heterozygous | |
SNU-1746 | p.E523K | large_intestine | Homozygous | |
SNU-324 | p.C233R | pancreas | Heterozygous | |
SNU-324 | p.V384D | pancreas | Heterozygous | |
SNU-478 | p.V384D | pancreas | Heterozygous |
Table 2. MLH1-mutated Cell Lines. Cell line (sample name), amino acid substitution, the lineage of origin and the zygosity are indicated.
Selection of cell line(s): Characterization of genetic state and genomic instability
Undoubtedly, the single most critical factor in successfully uncovering clinically relevant resistance mechanisms is the initial cell line selection. Two factors should be considered. First, aim to select cell line(s) of the same lineage/subtype harboring the defining genetic traits of the disease (e.g., BRAFV600E in melanoma). Interrogation of publically available transcriptomic and mutation data for both cell lines30-32 and primary/metastasis tumors for a variety of indications33,34 will facilitate the selection process. Although identification of cell lines with clinically relevant genetic alterations is ideal, in some instances this may not be possible due to lack of available cell lines or feasible due to factors such as difficulty and length of the screening process.
With regards to the aforementioned point, a second factor to consider then during cell line selection is the ease or feasibility of resistance screening using the desired line. For example, factors such as proliferation and intrinsic mutation rates can greatly impact on the speed of discovery. To this end, cell lines can be utilized with faster growth kinetics and deficient in DNA MMR mechanisms in the hopes that emergence of spontaneous resistance can be accelerated35. Based on the COSMIC database, several candidate cell lines exist with deficiency in one of two frequently mutated MMR genes, MSH2 or MLH1 (Tables 2 and 3). Alternatively, if cell lines of interest do not exist bearing defects in MMR, acute treatment with physical or DNA reactive chemical mutagens such as the alkylating agent N-ethyl-N-nitrosourea (ENU) can be used to enhance genomic instability. Although both approaches may significantly shorten the time to attain resistant clones and follow-up sequencing, stringent functional testing should be performed on candidate genes as a greater number of non-functional, passenger mutations are likely to emerge. SNVs can be rank ordered to maximize chances for identifying functionally relevant mutations. Firstly, choosing those mutations that are recurrent in independent clones will increase the likelihood these mutations are drivers of resistance. In the event recurrent mutations are not identified, focusing on SNVs that fit the mechanism of action of the drug (for example, the drug target or a known downstream effector of the drug target) may be meaningful. Ultimately, the gold standard is always experimental evaluation of resistance-conferring activity of the candidate SNVs by ectopic cDNA expression in drug-sensitive parental cells.
DNA vs RNA sequencing
Once resistant clones have been generated, DNA and/or RNA can be sequenced depending on the need. DNA sequencing, either exome or whole-genome sequencing, will allow identification of germline and somatic variants, such as SNPs, indels and copy number variants. Whereas the more cost-friendly exome sequencing focuses on generating reads from known coding regions, whole-genome sequencing will generate sequencing data for the entire genome which may facilitate identification of mutations in non-coding elements such as enhancers or miRNAs36. However, since gene expression data is not gauged during DNA sequencing, it is difficult to predict which mutation(s) is likely to be a functional driver. In this regard, sequencing of RNA, although more costly, offers this advantage. The fact that mutation calling is only performed on expressed RNA species enhances the likelihood that the mutation of interest may be a functional driver. In addition to allowing a more focused survey of mutations for follow-up functional testing, RNA sequencing also offers the added advantage of being able to identify gene expression alterations, alternative splicing and novel chimeric RNA species, including gene fusions that may also serve as potent drivers of resistance.
Bioinformatics pipeline
Example commands are provided for illustration purposes, but more detailed documentation and tutorials are available from the Broad Institute16 and should be read thoroughly before beginning NGS analysis. The following commands are designed for a UNIX shell environment on a system on which all tools and reference data have been pre-installed. These commands also assume FASTQ files containing paired-end sequence reads from two samples, named "parental" and "resistant," have been received from the vendor and placed in the "data" directory. In most cases these commands should be adapted or optimized for a specific application using additional command-line arguments (e.g., adding "-t 8" to the bwa command allows multithreaded operation across 8 CPU cores). Read groups (which assign alignments to biological samples), must often be added to BAM files even if there is only one sample per bam file, in order to comply with file format requirements for certain tools. Read group parameters RGID, RGSM, RGPL, RGPU, and RGLB can be arbitrary strings describing the sample name, sequencing platform, and library strategy.
In vitro vs in vivo assays
Although several resistance mechanisms identified by in vitro selection have been verified to be clinically relevant, there exists a possibility that the mechanisms may not serve as relevant or predominant mechanisms of clinical resistance. One reason for this may include an essential role for the micro-environment in driving resistance to therapy, a component that is devoid in the experimental protocol/setup discussed thus far. Indeed, several studies have shown that anti-cancer agents that are capable of killing tumor cells are rendered ineffective when the tumor cells are cultured in the presence of stromal cells implying innate mechanisms of resistance conferred by the stroma37,38. To identify such stroma-induced acquired resistance mechanisms, one may consider performing in vitro co-culture or in vivo tumor resistance assays. Since the former assay is quite complex, many have resorted to generating drug-resistant tumor xenografts to address the potential role of the stroma in driving resistance. Such studies have uncovered both identical5 and unique39 mechanisms of resistance relative to in vitro selection, implying that the stroma may indeed play a role in the latter. However, one must be mindful of the length of time it may take to generate such resistant tumors and the complexity of the follow-up genomic analysis-complexities due to the intra-tumoral molecular and cellular heterogeneity.
Target identification
In addition to uncovering drug resistance mechanisms, this NGS-based genomic profiling approach can also be applied to identify cellular targets of chemical probes. Historically, multiple unbiased methods have been used to identify the cellular mechanisms of action and targets of low-molecular weight chemicals with biological activities, including affinity purification coupled with quantitative proteomics, yeast genomic methods, RNAi screening, and computational inference approaches40. As an extension to elucidation of drug-resistance mechanisms using NGS-based genomic or transcriptomic profiling of phenotypically resistant cell populations, identification of unique recurrent single nucleotide variations (SNVs) or expression alterations that enable resistance can offer insights into functional cellular targets of compounds. This is based on the idea that a subset of resistance mechanisms observed may involve recurrent mutations in genes that encode the direct protein targets of the small molecule. Recently, several reports validated the utility of the approach, particularly by combining with other approaches including large-scale cancer cell line sensitivity profiling, to revealing the cellular targets of small-molecule probes9,10.
The authors have nothing to disclose.
The authors would like to acknowledge our colleagues at H3 Biomedicine for their feedback during the manuscript preparation.
48-well plates | Fisher Scientific | 07-200-86 | Expanding clones |
96-well plates | Fisher Scientific | 07-200-588 | Generating GI50 curves |
CellTiter-Glo | Promega | G7572 | Viability testing |
Cloning discs (3 mm) | Sigma | Z374431 | Picking clones |
Sterile forceps | Unomedical | DF8088S | Picking clones |
RNeasy Plus RNA extraction kit | Qiagen | 74134 | Isolating RNA |
Blood and Tissue DNeasy extraction kit | Qiagen | 69581 | Isolating gDNA |
GATK | The Broad Institute | Indel realigner | |
MuTect | The Broad Institute | Paired variant calling tool | |
Oncotator | The Broad Institute | Variant annotation tool | |
MutationAccessor | The Broad Institute | Functional impact prediction tool |