A 7 gene expression score predicts for radiation response in cancer cervix

Cervical cancer is a preventable but the most common cancer among Indian women and second most common cancer among women worldwide [1, 2]. More than 500,000 women are expected to develop the disease every year worldwide. The tragedy is worsened by most of the women presenting in locally advanced disease as well. Human Papilloma virus has been identified as an important factor in the development of cervical cancer, but it alone is not sufficient and requires additional events [3]. Cervical cancer progresses through initial dysplastic or pre-malignant stages before becoming invasive. The evolution to an invasive cancer may take up to 15 years or more [4]. PAP smear testing has been found to aid in early detection of pre-malignant lesions, thereby preventing the morbidity and mortality of cervical cancer. However, organized screening is primarily available in the Western countries and most of the developing countries including India do not have an organized screening program.

The standard treatment currently recommended for stage IIB, IIIA, IIIB and IVA has been concurrent chemo-radiotherapy, usually with weekly cisplatin based chemotherapy [5]. This is based on the data from 5 randomized trials which had shown a 30 -50% benefit in reducing the risk of death compared to the radiotherapy only arm. While the addition of chemotherapy is clearly beneficial, not all patients may need it. Currently, there are no reliable ways of identifying the individuals who may respond to the radical radiotherapy from those who will not respond.

The use of molecular approaches, such as gene expression studies, has helped tailor treatment based on the molecular characteristics. The Oncotype Dx testing for breast cancer is an example of using the newer techniques to help tailor the treatment to individual patient [6]. Several potential markers for tumour response have been identified in the past, such as c-myc, Bcl2, Bax, HIF2α, Ki67 index, c-erbB2 [7‐11]. However, most of these studies were on smaller number of patients and have not been used routinely to predict radiation response. More recently a few studies have used microarray based technology to identify gene signatures predictive of radiation response [12‐14] however, none of them had validated their genes using further techniques such as Real time Polymerase Chain Reaction (PCR).

Avoiding morbidity and the cost of chemo-radiation in cervical cancer would be relevant particularly in a developing country. This study aims to develop a test which can help identify patients who may not need chemotherapy and who can be cured only with radical radiotherapy.

Archival total RNA extracted from punch biopsy samples collected in RNA later (Ambion, Austin, Tx; Cat no: AM7021) and stored in the tumour bank after an informed consent were used, after obtaining the Institutional Ethical committee's approval for the study. The RNA had been extracted from the biopsy samples using the RNeasy RNA extraction kit (Qiagen, Gmbh, Hilden; Cat no: 74106) as per the manufacturer's instructions.

Twenty five patients' samples were included in the study. The criteria for inclusion in the study were as follows: 1. good quality RNA as assessed by Bio-analyser (RIN 6 or above); 2. paired paraffin block having at least 70% tumor cells; 3. sufficient quantity of RNA be available; 4. patient should have completed prescribed radiotherapy and follow-up information till death/last disease free status be available.

All the patients had received radical radiotherapy, with 6 MV X-ray from Linear accelerator and brachytherapy with HDR-Ir¹⁹². The total dose given was between 64 - 66 Gy, over 56 - 60 days. The median and mean follow-up period among the NED group was 43 months and 42 months, respectively, with a minimum and maximum period of follow up of 38 and 45 months, respectively. All the Failed patients had been followed up till progression/relapse/death. Complete remission refers to complete resolution of measurable disease by clinical examination, hematologic, biochemical or radiological examination; Partial remission refers to 50% or more disease reduction lasting longer than one month; Progression refers to worsening or advancing disease.

The patients' clinico-pathological status and their response to treatment are given in Table 1. Twenty four of the tumours were Squamous cell carcinomas (16 Large cell non-keratinizing, 4 large cell keratinizing and 4 unspecified) and one was a poorly differentiated carcinoma. Fourteen were HPV16 positive, 7 were HPV18 positive and 4 were HPV16 and 18 subtype negative (but HPV positive).

Using different methods, as described above, genes that were found to be differentially expressed between the two classes (NED and Failed) were obtained. The List of genes significant by different methods of analysis are given as a Additional file 2.

Twenty genes were selected for further validation by RQ-PCR using the Taqman Low Density Array card (TLDA) format (Table 2). These 20 genes formed part of the 95 genes selected for analysis using the TLDA format. Most of the additional genes were those which had been found to be differentially expressed between cervical cancer and normal cervix; cervical cancer and CIN1; CIN3 and CIN1; CIN3 and Normal cervix. Apart from the mandatory endogenous 18S rRNA included in the TLDA cards, based on the microarray data, UBC and β2 microglobulin, were included as additional endogenous reference genes.

One of the sample CXL19-hov160 which had worked in microarray did not amplify in the RQ-TLDA assay and had to be removed from further analysis. In addition, OPTC and C11orf9 genes did not amplify in any of the samples.

The RQ values after calibrating with the NED samples (Mean) for all the 95 genes showed two additional genes, UBE2C and IGF2BP2, to be overexpressed in the Failed samples compared to the NED samples. These two genes had initially been chosen for validation of the differentially expressed genes between cancer and normal cervix. After excluding the genes which did not amplify, we now had 20 genes for further analysis (see Additional file 3).

There were 10 genes (including UBE2C and IGF2BP2) with more than 2 fold difference between the two classes. In addition, 4 genes had a fold change between 1 and 2. There was only 44% (8/18) concordance between the microarray data and the Relative quantitation RT-PCR data, when a 2 fold difference was used as the criteria. The best concordant rates were with the SAM survival analysis, wherein of the 11 genes taken up for validation, 8 were concordant (73%).

We then tried to analyse the RQ data using each of the NED sample once as a calibrator. The focus was on genes which had an over-expression of at least two fold between the two classes. A scoring system was then developed as described in the Methods section. Using this we found that the seven gene score (UBE2C, MMP3, DCUN1D5, SDCCAG8, IGF2BP2, CCL18 and FST) can help distinguish between the Failed and the NED samples. Table 3 gives the mean score of the 24 experiments, the standard deviation, the standard error (SE), and score ± 2SE (see Additional file 4 for the details). Using a cut-off value of > 5.60, the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) are 0.64, 1.0, 1.0, 0.67, respectively, for the Mean ± 2SE. At this cut-off point, 9 of the 14 (64) NED samples were classified as low risk, while all the failed samples were classified as high risk.

The Kaplan Meier survival curves are given in Figures 1 and 2, the former based on the stage of the disease and the latter based on the gene expression signature.

Cervical cancer although a preventable disease, still kills a significant number of women. Most of the tertiary centres in developing countries still see patients in advanced stages. Radiotherapy has been the main stay of treatment and the addition of concurrent cisplatin based chemotherapy has been found to improve cure rates. However, not all patients will need concurrent chemotherapy, which has its own cost implications and morbidity, particularly in a developing country. With the emphasis shifting to tailoring therapy to individuals based on the patient and tumour characteristics, it is essential to develop newer tests which can help identify individuals who can be cured only with radiotherapy. This paper provides one such approach, wherein a seven gene score was found to help distinguish those who are likely to be cured with radiotherapy alone from those who may require additional forms of treatment.

The concordance between the microarray and the RQ-PCR data on genes differentially expressed with more than 2 fold was 44% (8/18). SAM Survival analysis had identified 13 genes (of the 14 genes, one was a duplicate -YAP1) as being differentially expressed and 11 of these genes were included in our validation. One gene did not have a suitable assay for the TLDA format. Of the 11 genes tested, 8 were found to be concordant in the RQ-RT-PCR validation (73% concordance rate). The SAM Survival analysis had correctly identified 5 of the 7 genes which we had used for the prediction.

One group had addressed the issue of concordance, when they compared Affymetrix data with Real time quantitative data for 48 genes. Their concordance rate was around 69% and they have reported that this may be usually due to differences in the target transcripts being identified by the two platforms (microarray versus Real time PCR) [20].

Klopp et al [14] had used microarray analysis on samples obtained prior to treatment and 48 hours after start of chemoradiation in 12 patients and found that a 58 gene signature can help predict recurrence. Others have developed either a Lymph node prediction model using 156 gene signatures, which had a prediction accuracy of 77% in predicting lymph node metastasis [21] or thermo-radio-response prediction model using a 35 gene signature [22]. Wong et al [13] used supervised clustering analysis to classify radiosensitive and radioresistant tumours in 13 patients.

Six of the seven genes, identified in our study, are known to be involved in cancers, with some of them having a role in inducing therapeutic resistance.

Locally advanced cancer of the cervix is now being treated with chemo-radiotherapy. However, nearly 60% of stage IIB and 40% of stage IIIB can be cured with radiotherapy only. Currently there are no reliable ways of identifying those patients who can be cured with radiotherapy alone.

Using microarray technique followed by Relative Quantitation Real Time PCR (RQ RT-PCR), we have identified a 7 gene signature with differential expression between cervical cancers with and without recurrence/progression after radiation treatment. This finding needs to be validated in an independent larger patient population.

The authors declare that they have no competing interests.