Expression Profiling in Ovarian Cancer Reveals Coordinated Regulation of BRCA1/2 and Homologous Recombination Genes

Abstract

Predictive biomarkers are crucial in clarifying the best strategy to use poly(ADP-ribose) polymerase inhibitors (PARPi) for the greatest benefit to ovarian cancer patients. PARPi are specifically lethal to cancer cells that cannot repair DNA damage by homologous recombination (HR), and HR deficiency is frequently associated with BRCA1/2 mutations. Genetic tests for BRCA1/2 mutations are currently used in the clinic, but results can be inconclusive due to the high prevalence of rare DNA sequence variants of unknown significance. Most tests also fail to detect epigenetic modifications and mutations located deep within introns that may alter the mRNA. The aim of this study was to investigate whether quantitation of BRCA1/2 mRNAs in ovarian cancer can provide information beyond the DNA tests. Using the nCounter assay from NanoString Technologies, we analyzed RNA isolated from 38 ovarian cancer specimens and 11 normal fallopian tube samples. We found that BRCA1/2 expression was highly variable among tumors. We further observed that tumors with lower levels of BRCA1/2 mRNA showed downregulated expression of 12 additional HR genes. Analysis of 299 ovarian cancer samples from The Cancer Genome Atlas (TCGA) confirmed the coordinated expression of BRCA1/2 and HR genes. To facilitate the routine analysis of BRCA1/2 mRNA in the clinical setting, we developed a targeted droplet digital PCR approach that can be used with FFPE samples. In conclusion, this study underscores the potential clinical benefit of measuring mRNA levels in tumors when BRCA1/2 DNA tests are negative or inconclusive.

1. Introduction

Ovarian cancer is the deadliest form of gynecological cancer [1,2]. The average lifetime risk of developing ovarian cancer is 1.3%, the equivalent of 1 in 78 women [1]. Although ovarian cancer accounts for 2.5% of all malignancies in women, it is responsible for 5% of female cancer deaths [2]. This high fatality rate is usually attributed to frequent late diagnosis, limited therapeutic options, and a lack of effective biomarkers for predicting treatment response [3].

The standard treatment for newly diagnosed ovarian cancer is cytoreductive surgery followed by platinum-based chemotherapy [4]. However, approximately 70% of patients have a relapse within the subsequent 3 years [3]. Recurrent ovarian cancer is typically incurable, with most patients receiving multiple additional lines of treatment.

Over the past 10 years, a new class of small molecule drugs that inhibit poly(adenosine diphosphate-ribose) polymerases (PARPs) has attracted much attention for the treatment of ovarian cancer [5]. The PARP family comprises several proteins involved in a variety of cellular processes, including the stress response, chromatin remodeling, DNA repair, and apoptosis [6]. One of the best-characterized targets of PARP inhibitors (PARPi) is PARP1, a protein that plays a major role in the detection and repair of single-strand DNA breaks [6,7]. PARP1 is also likely involved in other DNA repair pathways, such as nucleotide excision repair, non-homologous end joining, homologous recombination, and DNA mismatch repair [6,7]. Mechanistically, PARP1 is a DNA damage sensor and a signal transducer that binds to DNA breaks and then synthesizes poly(ADP-ribose) chains on target proteins (PARylation), including PARP1 itself (autoPARylation). Auto-PARylation further activates PARP1 and enables the PARylation of histones and other chromatin-associated proteins, which leads to the recruitment of additional DNA repair molecules to the site of damage [7].

In 2005, two studies revealed that tumor cells lacking BRCA1 or BRCA2 proteins are selectively vulnerable to PARPi [8,9]. Inhibition of PARP1 alone is not lethal for normal cells, but in the absence of BRCA1/2, it becomes cytotoxic for tumor cells [8,9]. The proposed mechanism of action is that PARP1 inhibition prevents PARylation and locks PARP1 on damaged DNA, stalling the progress of replication forks. In dividing cells treated with PARPi, stalled replication forks can be rescued by homologous recombination (HR), a process that enables high-fidelity, template-dependent repair of DNA damage [10]. Tumor cells that are defective in HR due to loss of BRCA1 or BRCA2 use alternative error-prone pathways that lead to fragmentation of the genome and ultimately result in cell death [5]. In addition to BRCA1 and BRCA2, several other proteins are involved in the HR process [11], and tumors with BRCA-independent HR deficiency are also sensitive to PARPi [6]. Although the first approved PARPi (olaparib) initially targeted patients with ovarian cancer that carried germline mutations in the BRCA1 or BRCA2 genes, a later study demonstrated responses with olaparib in 41% and 24% of germline BRCA mutation carriers and non-carriers, respectively [12]. Among the non-carriers, a sub-set had somatic BRCA mutations and a few others had mutations in other HR genes, such as RAD50, RAD51C/D, CDK12, or PALB2 [13]. These clinical studies were conducted in patients previously treated with platinum salts, which are DNA damaging agents that cause DNA crosslink repair in part by HR, and revealed a clear association between platinum sensitivity and response to PARPi [14]. Thus, further trials shifted towards using PARPi in patients who responded to platinum-based chemotherapy. Currently, PARPi are approved for the treatment of all platinum-sensitive patients, regardless of BRCA1/2 status. However, more benefit is seen in patients with either germline or somatic BRCA1/2 mutations [5].

Inherited susceptibility to ovarian cancer is found in approximately 15% of patients, the vast majority of which have germline heterozygous mutations in the BRCA1 or BRCA2 genes [15]. Loss of the wild-type BRCA allele can be detected in cancers that develop in heterozygous carriers, but whether this second hit is required for oncogenesis is unclear [16]. Somatic mutations in BRCA1 or BRCA2 genes are also occasionally present in sporadic (non-hereditary) ovarian cancers [15].

Developing predictive biomarkers for PARPi is of critical importance to delivering precision treatment to patients. Currently, DNA tests to detect both germline and somatic mutations in BRCA1 and BRCA2 genes are often used in the clinic. However, the utility of such assays is limited by the high prevalence of rare DNA sequence variants of unknown significance (VUS). Indeed, roughly equal numbers of VUS and pathogenic/likely pathogenic variants in BRCA1/2 are listed in the National Institutes of Health (NIH) repository ClinVar (https://www.ncbi.nlm.nih.gov/clinvar, last accessed on 25 May 2020). Moreover, most currently used DNA assays fail to detect epigenetic modifications (such as promoter hypermethylation), as well as mutations located deep within introns that may alter the mRNA [17].Here, we hypothesized that measuring the level of BRCA1 and BRCA2 mRNA in ovarian cancer tissue may provide additional information when DNA tests are negative or inconclusive. Downregulation of BRCA mRNA was previously shown to cause dysfunctional expression of BRCA proteins, resulting in HR deficiency and sensitivity to PARPi [18]. Mechanisms leading to reduced mRNA levels include epigenetic silencing and frameshift mutations. In addition to classical frameshift mutations (i.e., insertions, deletions, or duplications of exonic nucleotides that change the reading frame of the mRNA), splicing mutations often induce a frameshift as a consequence of exon skipping, use of unnatural (cryptic) splice sites, or intron retention [17]. Changing the reading frame for translation often introduces a premature termination codon (PTC) into the mRNA, and a cellular surveillance system termed nonsense-mediated mRNA decay (NMD) degrades the vast majority of mRNAs with PTCs [19]. Because NMD leads to the downregulation of expression and loss of function of the affected genes, it contributes to many genetic diseases including cancer. An example of this is the inactivation of tumor suppressor genes harboring PTCs through NMD [20].

The extent to which BRCA1 and BRCA2 mRNA levels are regulated in ovarian cancer has remained poorly understood. In this study, we identified a sub-set of tumors lacking any detected BRCA1/2 mutation that express low levels of BRCA1/2 mRNA similar to tumors with known NMD-eliciting PTCs. Notably, we found that the expression of 12 other HR genes is also downregulated in tumors with low levels of BRCA1/2 mRNA. These results underscore the potential clinical benefit of measuring mRNA levels in tumors when BRCA1/2 DNA tests are negative or inconclusive.

2. Materials and Methods

2.1. Biological Samples

We analyzed archived formalin-fixed and paraffin-embedded (FFPE) tissue biopsies, including ovarian cancer and normal fallopian tube samples. Samples were collected from the archives of the Hospital Santa Maria/Centro Hospitalar Universitário Lisboa Norte, E.P.E., Lisbon, Portugal, and Instituto Português de Oncologia, Porto, Portugal. An experienced pathologist assessed the quality of each sample. In the case of tumor samples, the diagnosis was confirmed, and the tissue was macro dissected to ensure that only tumor regions were further analyzed. The study was approved by the ethical committee of the Lisbon Academic Medical Center, Lisbon, Portugal (approval number: 170/18).

2.2. RNA isolation and Quantification

RNA was extracted from 5–8 curls of tissue (10 μm thick) sectioned from each sample. RNA was purified using the PureLink^TM FFPE RNA Isolation Kit (, Invitrogen, Carlsbad, CA, USA), as described by the manufacturer. The amount and purity of total RNA recovered were determined with the Nanodrop^TM 2000 spectrophotometer (Thermo Scientific, Waltham, MA, USA). Traces of contaminant DNA were removed by DNase I digestion. Finally, RNA concentration and size were analyzed in the Agilent 4200 TapeStation system using an Agilent RNA ScreenTape assay (Agilent, Santa Clara, CA, USA).

2.3. Gene Expression Profiling with the nCounter System

We used the nCounter^® FLEX Analysis System and the Vantage 3D™ RNA Panel for DNA Damage and Repair (NanoString Technologies, Seattle, WA, USA). The mRNA hybridization, detection, and scanning were performed at the NanoString nCounter Core Facility at the Serviço de Genética—Instituto Português de Oncologia do Porto Francisco Gentil, E.P.E., Porto, Portugal. Briefly, purified RNA (200 ng) was hybridized overnight at 67 °C. The full volume of the hybridization reactions was immediately introduced in the Prep Station, which is a liquid handling robot that performs the purification of the hybridized complexes and their immobilization onto the surface of a cartridge. The barcodes captured for each sample were then identified and counted by the Digital Analyzer. The resulting data were imported into and analyzed by the nSolver 4.0 Analysis Software System. Quality control parameters recommended by NanoString were applied using the nSolver default settings. Samples that failed the quality control metrics were excluded from further analysis. For data normalization, raw counts were the first background adjusted using the internal negative controls followed by a within-sample normalization using the internal positive controls. Finally, data were normalized across samples (i.e., corrected for input) using the mean RNA counts from reference (housekeeping) genes. This resulted in a total of 49 samples, 11 from normal fallopian tubes and 38 from ovarian cancer tissue.

2.4. BRCA1 and BRCA2 mRNA Analysis by Droplet Digital PCR

The level of BRCA1 and BRCA2 mRNA in ovarian cancer samples previously analyzed with the nCounter assay was quantified by droplet digital PCR (ddPCR) using the Bio-Rad QX200 system (Bio-Rad Laboratories, Hercules, CA, USA), according to the manufacturer’s instructions and the digital MIQE guidelines [21,22]. First, purified RNA samples were treated with DNAse I (Roche, Mannheim, Germany) 0.5 U/µL, at 37 °C for 30 min, in the presence of RiboSafe (Bioline, London, UK) 1 U/µL. After purification with an equal volume of phenol:chloroform:isoamyl alcohol (25:24:1) in the presence of 0.3M Sodium acetate pH 5.2 and 50 μg/mL GlycoBlue™ coprecipitant (ThermoFisher Scientific, Waltham, MA, USA), RNA was precipitated with 2 volumes 100% ethanol, dissolved in 0.1 mL of RNAse free water and quantified in NanoDrop 2000 spectrophotometer. Then, RNA was retrotranscribed and amplified using the One-Step RT-ddPCR Advanced Kit for Probes (Bio-Rad, Hercules, CA, USA,) and PrimeTime Std^® qPCR Probe Assays from IDT (Integrated DNA Technologies, Coralville, IA, USA). Primers and Taqman fluorescent probes were designed with IDT software tools available online and are described in Supplementary Table S1. All genes were analyzed with 6-FAM/ZEN/IBFQ labeled probes, except reference gene GUSB for which we used a HEX/ZEN/IBFQ labeled probe. Duplex experiments were performed targeting either BRCA1, BRCA2, POLR1E or NUBP1 (FAM) and GUSB (HEX) simultaneously.

The ddPCR workflow started by partitioning the RT-PCR reaction mix (total volume 20 µL) containing forward and reverse primers (900 nM final concentration), probes (250 nM final concentration each), and sample RNA into aqueous droplets in oil. Negative controls included samples with no reverse transcriptase in the RT-PCR master mix. The droplets were then subjected to a thermocycling protocol for retro transcription and amplification (50 °C for 1 h; 95 °C for 10 min; 46 cycles of 95 °C for 35 s and 60 °C for 1 min; and 98 °C for 10 min). The PCR products were kept at 4 °C for up to 1 day before reading the fluorescent content of the droplets in the QX200 Droplet Reader (Bio-Rad, Hercules, CA, USA). Data were analyzed with the QuantaSoft software (Version 1.7.4.0917, Bio-Rad, Hercules, CA, USA). The fraction of positive and negative droplets was determined based on the fluorescent signal, and the results were fitted to a Poisson distribution to estimate absolute concentrations (copies/µL) of each mRNA.

To ensure that similar amounts of reference mRNA were loaded per sample, two assays were sequentially performed for each sample. In the first assay, 30 ng of total RNA was loaded and the number of copies per droplet (CPD) of GUSB mRNA was quantified. The amount of total RNA loaded in the second assay contained GUSB mRNA at a concentration of 0.05 CPD (50 copies/µL).

For comparison, we used the same ddPCR assays to test RNA extracted from non-fixed, freshly isolated peripheral blood mononuclear cells (PBMCs).

2.5. Statistical Analysis

2.5.1. General

Analysis and plotting were performed with R Core Team 2019 (version 3.6.1, R Foundation for Statistical Computing, Vienna, Austria). Empirical p-values were calculated as $P=\frac{n+1}{m+1}$, where n is the number of stimulants that show a value as high as or higher than the true value, and m is the total number of simulants [23], (Spearman’s correlation coefficients were converted to z’-values and back using the functions FisherZ() and FisherZInv() in the DescTools package v. 0.99.36.

2.5.2. Analysis of NanoString Data

Genes were classified into pathways based on the labels supplied with the NanoString panel. The Homologous Recombination and Fanconi Anemia pathways were grouped, and the relevant genes are referred to as Homologous Recombination (HR) genes throughout the text. Twelve of the genes in the panel (ERCC1, ERCC2, ERCC3, ERCC4, ERCC5, ERCC6, ERCC8, EXO1, FAN1, FANCA, FANCB, FANCC) were excluded from the analysis due to a problem with a particular lot of TagSets, which passed initial quality control but was subsequently found to lead to abnormally low NanoString counts. Note that a few of the samples were included in two separate NanoString runs. In these cases, the values from the two runs were averaged in all analyses. The heatmap was produced with the heatmap.2() function from the gplots package (version 3.1.1., https://CRAN.R-project.org/package=gplots, accessed on 16 November 2021), using the complete linkage clustering method with $1 – Spearman’s \rho$ as the distance function. Prior to generating the heatmap, data were scaled so that the mean of each gene was 0 and the standard deviation was 1 so as to normalize for differences in NanoString probe efficiency.

2.6. The Cancer Genome Atlas (TCGA) Data Analysis

TCGA ovarian serous cystadenocarcinoma mRNAseq RSEM values were retrieved from http://firebrowse.org/ (last accessed: 22 November 2019). The data were filtered to remove genes with very low read counts using the filterByExpr() function from the edgeR package v3.28.1 [24]. The edgeR calcNormFactors() function was used to calculate library size normalization factors, which were then used when converting the data into log 2 Counts Per Million (log2CPM) using voom(). All edgeR functions were run with default parameter settings. In total, 164 of the 168 genes that were used from the NanoString panel were available in TCGA data (after read count filtering).

3. Results

3.1. Study Cohort

Our study cohort comprised 42 FFPE tissue samples from patients diagnosed with ovarian cancer, the majority of which were classified as high-grade serous carcinomas (Supplementary Table S2). All tumor tissues had been previously screened for the presence of mutations in the BRCA1 and BRCA2 genes, and the results are indicated in Supplementary Table S2. Because most ovarian cancers are probably derived from the fallopian tube epithelium [25,26], we included 18 samples of normal fallopian tubes surgically resected from individuals with no cancer diagnosis in our study.

3.2. Quantification of BRCA1 and BRCA2 mRNAs with the nCounter Assay

We used the nCounter assay from NanoString Technologies because this is considered an ideal method for expression profiling in FFPE tissues. The assay counts individual mRNA molecules with a sensitivity similar to real-time PCR but without the bias introduced by enzymatic reactions [27]. This is particularly relevant when using RNA isolated from FFPE specimens because fixation with formaldehyde induces the formation of methylene crosslinks that limit successful reverse transcription and PCR amplification [28]. Moreover, RNA becomes progressively more fragmented in FFPE specimens, particularly when blocks are stored at room temperature [29].

In our study, RNA integrity (RIN) values ranged from 1.8 to 5.0 and concentrations from 19.7 to 353.4 ng/µL (Supplementary Figure S1A and Supplementary Table S2). One of the quality control requirements for the nCounter gene expression assay is that at least 50% of the purified RNA molecules are longer than 300 nucleotides (Supplementary Figure S1B). As shown in Supplementary Figure S1C, the RNA extracted from some samples did not meet this quality requirement and was therefore discarded from further analysis. To assess reproducibility, RNA purified from normal fallopian tubes and tumor samples was repeatedly run with different reagent and probe batches and the results showed similar RNA counts (Supplementary Figure S2).

To profile the expression of BRCA1 and BRCA2 genes, we used the nCounter Vantage™ RNA Panel for DNA Damage and Repair that hybridizes to mRNA from a total of 192 genes, including 12 housekeeping genes and 180 unique genes related to DNA damage and repair pathways (Supplementary Table S3). In normal fallopian tubes, we estimated means of BRCA1 and BRCA2 mRNA levels corresponding to 185 and 86 NanoString counts, respectively, with standard deviations of 45 and 29 (Figure 1A). In tumors, the mean level of BRCA1 and BRCA2 mRNA corresponded to 280 and 338 NanoString counts, respectively, with standard deviations of 159 and 192. Thus, in contrast to the little variation in mRNA levels of normal fallopian tubes, BRCA1 and BRCA2 expression was highly variable among tumors. Some tumors had values comparable to normal tissue while others were up to 10-fold higher than the median of normal samples (Figure 1A–C). The median expression was significantly higher in tumors than in normal tissue for both BRCA1 (fallopian tube median: 195.23; tumor median: 256.515; p~0.010; two-tailed Welch’s t-test on ranks) and BRCA2 (fallopian tube median: 74.13; tumor median: 305.87; p~5.59 × 10⁻¹²; two-tailed Welch’s t-test on ranks).

Next, we focused on tumors harboring mutations in the BRCA1/2 genes (Figure 1B,C). Ten tumors had frameshift mutations located in either BRCA1 exon 11 or BRCA2 exons 10 and 11 (Supplementary Figure S3). These mutations generate PTCs and are predicted to elicit NMD according to recently validated rules [30]. Indeed, the PTCs were more than 50–54 bp upstream of the last exon-exon junction, both BRCA1 and BRCA2 genes comprised more than two exons, and the PTCs were more than 200 bp downstream of the start codon (Supplementary Figure S3). Despite inter-sample variability, tumors with frameshift mutations were among those with lower expression levels (Figure 1B,C). In these samples, the median BRCA1 expression was 235.63 (median of all tumors: 256.52), and the median BRCA2 expression was 225.71 (median of all tumors: 305.87). Three tumors had the c.156–157 insAlu BRCA2 mutation, which has only been reported in families of Portuguese ancestry. This Portuguese BRCA2 founder mutation is present in ~30% of all Portuguese families with hereditary breast and ovarian cancer, representing 55% of all BRCA2 germinal mutation carriers in Portugal [31]. This mutation leads to full in-frame skipping of exon 3 [31,32], which encodes amino acids 23 to 105. Possibly, this protein deletion reduces the interaction between BRCA2 and PALB2, and impairs homologous recombination, thus conferring hypersensitivity to DNA damage [33]. The levels of BRCA2 expression in these tumors ranged from 157.44 to 686.05, with a median of 438.28 (Figure 1C). Four tumors had missense mutations. In tumors with BRCA1 missense mutations, the levels of BRCA1 expression ranged from 195.1 to 557.5, whereas the range of BRCA2 expression in tumors with frameshifts in BRCA2 ranged from 273.93 to 477.36 (Figure 1B,C).

3.3. Expression Profiling of DNA Damage and Repair Genes

Having shown that the levels of BRCA1/2 mRNAs were highly variable between tumor samples, we next analyzed the expression of the additional genes probed with the nCounter Vantage™ RNA Panel for DNA Damage and Repair (Supplementary Table S3). We observed strikingly different expression patterns between normal tissue and many of the tumor samples, although there was high variation between tumors (Figure 2A, see Supplementary Table S4 for the NanoString counts of all the genes in all the samples). We found 72 genes to be differentially expressed between tumor and normal samples (two-tailed Welch’s t-test on ranks with Bonferroni correction; genes were considered to be significantly differentially expressed if the corrected p-value was below 0.05; Figure 2B,

Table 1. Genes significantly differentially expressed in tumors.

(A) Genes Upregulated in Tumors.
Pathway	Genes
Apoptosis	BCL2L1, CASP8, MYD88
Base excision repair	APEX2, FEN1, NEIL3, PARP1, SMUG1, UNG
Base excision repair—Translesion Synthesis—Cell Cycle and Signaling	PCNA
Cell Cycle and Signaling	AURKA, BUB1B, CDKN2A, KRAS, RAD21, RM12, SUMO3
Checkpoint Activation	H2AFX
Checkpoint Activation—Cell Cycle and Signaling	CHEK1/2
Homologous Recombination and Fanconi Anemia	BRIP1, BLM, BRCA2, FANCD2, FANCG, FANCI, GEN1, RAD51, RAD54L, UBE2T, USP1, XRCC2, XRCC3
Independent Repair Enzymes/Polymerases	DNA2, POLQ
Independent Repair Enzymes/Polymerases–Base excision repair	POLD1, POLD4, POLE2
Independent Repair Enzymes/Polymerases—Cell Cycle and Signaling	MAD2L2
Independent Repair Enzymes/Polymerases–Nucleotide Excision Repair	POLR2D, POLR2H
Independent Repair Enzymes/Polymerases–Translesion Synthesis	HLTF
Mismatch Repair	MSH2
Mismatch Repair–Translesion Synthesis	RFC4
Non-homologous End Joining—Cell Cycle and Signaling	PRKDC
Translesion Synthesis	RAD18
Housekeeping	SF3FA3
(B) Genes downregulated in tumors.
Pathway	Genes
Apoptosis	AKT3, BCL2, NFKB2, PIK3R1
Base Excision Repair	NEIL1, NEIL2, OGG1
Cell Cycle and Signaling	ABL1, CCND2, CDKN1A, EGFR
Checkpoint Activation	TIPIN
Homologous Recombination and Fanconi Anemia	RAD51B, WRN
Independent Repair Enzymes/Polymerases	ALKBH3, CRY1, POLK, REV1
Independent Repair Enzymes/Polymerases–Translesion Synthesis	POLI
Mismatch Repair	MSH3
Non-homologous End Joining	LIG4
Nucleotide Excision Repair	XPA
Nucleotide Excision Repair—Apoptosis	PTEN
Housekeeping	COG7, NUBP1

). Among the 47 genes showing significantly higher expression in tumors, 13 (~28%) belonged to the Homologous Recombination (HR) pathway ((A) in Table 1).

We next asked whether the expression of BRCA1 and BRCA2 correlates with the expression of the other HR genes that we found upregulated in tumors. We divided the cohort into low and high groups based on their BRCA1/2 expression. The low group included tumors with BRCA1/2 mRNA levels equal to or below the median value, while the high group included tumors with mRNA levels above the median. As shown in Figure 3A,B, the expression of significantly differentially expressed HR genes was systematically downregulated in the BRCA1/2 low groups and upregulated in the BRCA1/2 high groups (Supplementary Table S5 contains the Spearman coefficients for the correlations between the NanoStrings’ expression levels of BRCA1/BRCA2 and the genes shown in Figure 3A,B).

HR genes were not the only genes to be upregulated in tumors ((A) in Table 1). It is possible that BRCA expression is not correlated specifically with the expression of other HR genes but rather that all genes that we detected to be upregulated in tumors show similarly correlated expression with each other, independently of the DNA damage pathway that they belong to. To distinguish between these possibilities, we calculated the Spearman correlation coefficient between BRCA1/2 and each of the non-BRCA HR genes that were upregulated in tumors. We took the mean of these coefficients as our statistic for the extent of co-regulation between BRCA1/2 and the other upregulated HR genes (BRCA1: ~0.59; BRCA2: ~0.69; note that a Fisher’s Z-transformation was applied to the correlation coefficients prior to calculating the mean, with the mean of z’ values then converted back to a Spearman’s correlation coefficient). We then performed a simulation where, over 10,000 iterations, we randomly picked 12 of the upregulated genes (12 is the number of upregulated HR genes, excluding BRCA2) and calculated the mean correlation coefficient with BRCA1/2 for this random sample. The mean of the values obtained with the different random samples is the random expectation for how strongly BRCA1/2 expression is expected to correlate with a random set of upregulated genes (BRCA1: ~0.50; BRCA2: ~0.53). For both BRCA1 and BRCA2, this random expectation was lower than the value actually observed (BRCA1: p~0.012; BRCA2: ~4.000 × 10⁻⁴), suggesting that BRCA1/2 are specifically coregulated with a subset of other HR genes (see Methods for further details on the calculation of empirical p-values).

3.4. Coordinated Expression of BRCA1/2 and a Subset of HR Genes in TCGA Data

In order to determine whether the coordinated expression of BRCA1/2 and the 12 other HR genes identified above could be confirmed in an independent data set, we obtained mRNA sequencing data for genes related to DNA damage and repair pathways in 299 ovarian cancer samples from The Cancer Genome Atlas (TCGA). Similar to our analysis of NanoString data, we divided the samples into a high and low BRCA1/2 expression group along the median. We asked whether the HR genes found to be upregulated in tumors in NanoString data showed evidence of co-regulation with BRCA1/2 also in TCGA data. As can be observed in Figure 4A,B, when BRCA1/2 was low, the expression of HR genes was also low compared to their expression when BRCA1/2 was high (see Supplementary Table S6 for the Spearman coefficients for the correlations between the expression levels of BRCA1/BRCA2 and the genes shown in Figure 4A,B). TCGA data thus support co-regulation of BRCA1/2 with the other HR genes. To know if this effect was significantly stronger for HR genes than for upregulated genes in general, we also repeated the simulation reported in the previous section with TCGA data (using the list of significantly upregulated genes that was determined with NanoString above). We obtained a true mean Spearman correlation coefficient of ~0.30 (random expectation: ~0.22; empirical p~0.028) for BRCA1, and ~0.41 for BRCA2 (random expectation: ~0.28; empirical p~0.001).

3.5. Targeted Quantitation of BRCA1/2 mRNA by ddPCR

Having established that BRCA1 and BRCA2 expression was correlated with that of other HR genes in ovarian cancers, we sought to develop a targeted and less costly approach to measure BRCA1/2 mRNA levels in FFPE clinical samples. We decided to explore droplet digital PCR (ddPCR) because this approach was previously shown to provide reliable quantification of HER2 mRNA in FFPE breast cancer specimens [34,35]. We further opted for a one-step method in which the cDNA synthesis is initiated with the reverse primer within each droplet, thus minimizing bias introduced by retro transcription efficiency. We designed primers and probes that amplified and detected different regions of the BRCA1 and BRCA2 transcripts (Figure 5A,B). As reference genes, we analyzed GUSB, POLR1E, and NUBP1. Duplex experiments were performed targeting simultaneously either BRCA1, BRCA2, POLR1E or NUBP1 and GUSB (Supplementary Figure S4). The number of read droplets varied from 5549 to 19,264. Values obtained in control experiments with no retro transcriptase were null for 75% of the cases, with a maximum of 0.16 copies/µL. The concentration of BRCA1/2 target sequences varied from 1.8 to 57.2 copies/µL. This yield was approximately 15 times less than the CPD values obtained from non-fixed, freshly isolated PBMCs.

We tested 24 tumor samples for BRCA1 and BRCA2 mRNA levels, normalized to the reference genes GUSB, POLR1E, and NUBP1. For each tumor, we performed three different assays for BRCA1, and another three distinct assays for BRCA2 (Figure 5A,B). Similar values were observed with the assays targeting the different regions of the same mRNA molecule in PBMCs from 10 individuals, as well as in a sub-set of the tumor samples; however, in other tumor samples, the levels differed among the three regions analyzed (Figure 5A,B). For further analysis (see below), we selected the assay that consistently revealed higher mRNA levels across all tumors (i.e., for BRCA1 the assay that targets exons 11/12, and for BRCA2 the assay that targets exons 7/8).

A comparison between the relative expression values for BRCA1 and BRCA2 estimated by the NanoString nCounter system and ddPCR showed a significant correlation (Figure 5C); only one tumor sample was clearly discordant, with a high BRCA2 mRNA value estimated by the NanoString nCounter system but a low value quantified by ddPCR (Figure 5C, arrow). These results indicate that ddPCR is a reliable method to quantitate levels of BRCA1 and BRCA2 mRNA in clinical ovarian cancer samples.

4. Discussion

In this study, we measured the levels of BRCA1 and BRCA2 mRNA in ovarian cancers and normal fallopian tubes, the organ from which most ovarian cancers originate [25,26]. We used the nCounter assay from NanoString Technologies to circumvent the limitations caused by varying extents of RNA degradation in FFPE samples. We found that BRCA1 and BRCA2 expression was more variable among tumors than across normal tissues, with many tumors showing higher mRNA values compared to normal fallopian tubes. Previous studies showed that the expression of BRCA1 and BRCA2 genes is upregulated in rapidly proliferating cells [36,37]. Thus, the higher level of BRCA1 and BRCA2 mRNA observed in tumors may result from the proliferation status of cancer cells. Consistent with the view that the cellular proliferation rate is higher in ovarian cancers than in normal fallopian tubes, we identified several cell-cycle-related genes that were significantly upregulated in the tumors ((A) in Table 1). These included genes encoding proteins that stimulate cell proliferation, such as KRAS and AURKA, the mitotic checkpoint proteins BUB1B and MAD2L2, as well as proteins involved in DNA replication and repair of replication-associated lesions, such as Replication factor C subunit 4 (RFC4), DNA replication helicase/nuclease 2 (DNA2), DNA polymerase delta 1 (POLD1), DNA polymerase delta 4, accessory subunit (POLD4), DNA polymerase epsilon subunit 2 (POLE2), Flap structure-specific endonuclease 1 (FEN1), and Proliferating cell nuclear antigen (PCNA).

All tumors expressed BRCA2 mRNA at levels similar to or higher than the median of normal tissues (Figure 1). In contrast, a subset of tumors expressed BRCA1 mRNA at lower levels than the median of normal tissues (Figure 1). A possible cause for such a reduction could be epigenetic silencing via promoter hypermethylation, which has been detected in the BRCA1 gene in 10–20% of ovarian cancers [13,38,39]. Additional mechanisms that can lead to reduced BRCA1/2 expression include microRNA-mediated regulation [40] and frameshift-inducing mutations that trigger mRNA degradation by NMD. Notably, the majority of known germline and somatic BRCA1/2 mutations are frameshift insertions or deletions [13], and in our study, among tumors with lower levels of expression, we identified eight cases harboring frameshift mutations that are predicted to activate NMD. The frameshift mutations that we identified in the BRCA1/2 genes were heterozygous. Thus, only expression of the mutant allele is expected to be reduced. However, high variability of mRNA levels was observed between different tumors with frameshift mutations. One possibility is that transcription may be differentially upregulated depending on the proliferation rate of tumor cells. Alternatively, the normal allele may be differentially inactivated by loss of heterozygosity, which has been reported in a large fraction of ovarian cancers [13].

Several lines of evidence suggest that reduced BRCA1/2 gene expression in tumors leads to defective DNA repair via HR. Indeed, downregulation of BRCA1 protein induced by overexpression of miR-182 was sufficient to impair HR-mediated repair and hypersensitize breast tumor cells to PARPi [40]. PI3K inhibition was also shown to downregulate BRCA1/2 expression leading to DNA damage and sensitization to PARPi [41]. Phosphatidyl-inositol-3-kinases (PI3Ks) constitute a family of lipid kinases that when activated trigger a signaling cascade. One of the key proteins involved in this pathway is AKT/protein kinase B [42]. BRCA1 is one of many targets phosphorylated by AKT [43]. Genomic alterations leading to activation of the PI3K pathway are present in ovarian cancer [13], and both preclinical and clinical studies suggest that PI3K pathway inhibitors improve the cancer response to PARP inhibition [44]. Moreover, knockdown of BRCA1 in MCF-10A cells, which are immortal human mammary epithelial cells of nonmalignant origin, caused HR deficiency without affecting cell cycle progression and resulted in downregulation of mRNAs encoding several other HR factors and enzymes required for double-strand break repair [18]. This study further revealed that knockdown of distinct HR genes resulted in similar transcriptomic changes, suggesting that the expression of HR genes is coordinately regulated in HR deficient cells [18]. Our work extends these observations by showing that the expression of BRCA1 and BRCA2 genes significantly correlates with that of an additional 12 HR genes in clinical ovarian cancer samples. This implies that reduced levels of BRCA1/2 mRNA in tumors may result from a variety of genetic or epigenetic alterations in either the BRCA1/2 genes or other HR genes.

We further found an excellent correlation between BRCA1/2 gene expression levels in tumors analyzed by the NanoString nCounter system and ddPCR. Thus, our study demonstrates, for the first time, that ddPCR can be used for the accurate quantitation of BRCA1 and BRCA2 mRNA in FFPE ovarian cancer specimens. Compared to traditional RT-PCR approaches applied to clinical samples, ddPCR offers greater precision, improved reproducibility, and superior diagnostic performance [45]. In the particular case of RNA extracted from FFPE samples, ddPCR has the advantage of allowing extremely low target quantitation [46].

Previous studies described an association between reduced BRCA1 mRNA or protein levels and responsiveness to cisplatin therapy in different types of cancer [47,48,49,50,51,52,53]. Furthermore, platinum sensitivity correlated with lower BRCA1/2 mRNA expression in ovarian cancers [54]. Thus, measuring the levels of BRCA1 and BRCA2 mRNAs in ovarian cancers using either the NanoString nCounter system or ddPCR may help predict which patients will benefit the most from platinum-based chemotherapy and PARPi when DNA tests are negative or inconclusive. Future validation studies in large cohorts are needed to define cut-off values of BRCA1/2 mRNA levels capable of identifying which tumors are more sensitive to PARPi.