HLA-G 3’UTR Polymorphisms Impact the Prognosis of Stage II-III CRC Patients in Fluoropyrimidine-Based Treatment

An important hallmark of CRC is the evasion of immune surveillance. HLA-G is a negative regulator of host’s immune response. Overexpression of HLA-G protein in primary tumour CRC tissues has already been associated to worse prognosis; however a definition of the role of immunogenetic host background is still lacking. Germline polymorphisms in the 3’UTR region of HLA-G influence the magnitude of the protein by modulating HLA-G mRNA stability. Soluble HLA-G has been associated to 3’UTR +2960 Ins/Ins and +3035 C/T (lower levels) and +3187 G/G (high levels) genotypes. HLA-G 3’UTR SNPs have never been explored in CRC outcome. The purpose of this study was to investigate if common HLA-G 3’UTR polymorphisms have an impact on DFS and OS of 253 stage II-III CRC patients, after primary surgery and ADJ-CT based on FL. The 3’UTR was sequenced and SNPs were analyzed for their association with survival by Kaplan-Meier and multivariate Cox models; results underwent internal validation using a resampling method (bootstrap analysis). In a multivariate analysis, we estimated an association with improved DFS in Ins allele (Ins/Del +Ins/Ins) carriers (HR 0.60, 95% CI 0.38–0.93, P = 0.023) and in patients with +3035 C/T genotype (HR 0.51, 95% CI 0.26–0.99, P = 0.045). The +3187 G/G mutated carriers (G/G vs A/A+A/G) were associated to a worst prognosis in both DFS (HR 2.46, 95% CI 1.19–5.05, P = 0.015) and OS (HR 2.71, 95% CI 1.16–6.63, P = 0.022). Our study shows a prognostic and independent role of 3 HLA-G 3’UTR SNPs, +2960 14-bp INDEL, +3035 C>T, and +3187 A>G.


Introduction
Colorectal cancer (CRC) is still a clinical burden being the third most common cancer in the United States [1] and the second leading cause of cancer death in Europe, in both women and men [2]. Recent advances in protein-and genomic-based technologies, validated predictive and prognostic biomarkers, have demonstrated that CRC should be considered as a heterogeneous disease [3][4][5]. Adjuvant chemotherapy (ADJ-CT) based on fluoropyrimidine (FL) is generally administered in stage II-III patients after surgical resection of the primary tumour. Despite optimal surgery and adjuvant therapies, the risk of recurrence for stages II or III disease is about 40% [6] and~80% of stage II CRC patients will be disease-free even without ADJ-CT [7]. Recently, gene expression profile signatures of immune-system related genes and presence of the immune infiltrates in tumour microenvironment were shown to have an independent prognostic significance in CRC compared to classical clinical factors [8][9][10][11].
Tumour immunogenicity, cancer cell capability to escape from the host's immune system surveillance, and immunogenetic background of the patient, represent a future challenge of current research [12]. Two main mechanisms contribute to the cancer immunoediting process [13] leading to poorly immunogenic tumour cell variants invisible to the immune system. The first is the minimization of the level of tumour associated antigens (TAA) presentation through the downregulation or loss of the human leukocyte antigen (HLA) class I expression by tumour cells [14]. The second is the competence of cancer cells in regulating the expression of the nonclassical HLA class I molecules such as HLA-G [15]. The HLA-G gene codifies for a tolerogenic molecule with well recognized immune-inhibitory properties on both innate and adaptive immune responses [16][17][18]. HLA-G is highly expressed in physiological conditions in trophoblast at fetal-maternal interface and has a restricted distribution in normal tissues [19]; however, an increased expression can be induced in pathological conditions such as cancer [20,21]. Several genetic variations involved in HLA-G regulation have been so far described in the 5' upstream regulatory (or promoter) region (5'URR) as well as in the 3' untranslated region (3'UTR), while in contrast to the classical HLA class I loci, a lower variability in the coding regions is observed [22][23][24]. Increased soluble HLA-G levels in biological fluids are associated with down-modulation of the immune response in the host [15]. The HLA-G 3'UTR is the most studied segment of the gene due to the presence of multiple regulatory elements implicated in the modulation of HLA-G expression.
HLA-G is overexpressed in primary CRC lesions [35][36][37], and higher levels of the soluble protein have been detected in plasma of CRC patients compared to that of patients with benign diseases or healthy donors [38,39]. Expression of HLA-G in tumour tissues has been associated with the clinical outcome of CRC as an independent and unfavourable prognostic factor of reduced OS [40,41].
To date, the characterization of HLA-G genotypes, alleles and haplotypes in CRC patients has not been explored as well as their role in the prediction of CRC prognosis [42].
Considering the relevance of the 3'UTR region in the control and regulation of the HLA-G transcripts and the lack of data in literature, our purpose was searching for associations between HLA-G 3'UTR polymorphisms detected at the germinal level, and the disease free survival (DFS) and OS of stage II-III CRC patients in adjuvant regimen. We analyzed a cohort of 253 CRC patients to investigate if functional SNPs in the 3'UTR of HLA-G gene, alone or in combination in the 3'UTR haplotypes, are associated to advantage or disadvantage for DFS and OS.

Ethics Statement
A written Informed Consent was obtained before surgery from all the participants to the use of their blood samples and clinical data for research purpose. The study was approved by the ethical committees of the participating institutes, the Centro di Riferimento Oncologico (CRO)-Aviano National Cancer Institute, Aviano, Italy, University Hospital, Florence, Italy, Istituto Oncologico Veneto, Padua, Italy, Ospedale Civile di Vittorio Veneto, Vittorio Veneto, Italy, University Hospital "S. Maria Della Misericordia", Udine, Italy, and "San Filippo Neri" Hospital, Rome, Italy.

Patients and treatment
A total of 253 CRC patients with newly diagnosed, untreated, histopathologically confirmed CRC, were included from an existing prospective collection of only blood samples stored at the Experimental and Clinical Pharmacology Unit of Centro di Riferimento Oncologico (CRO)-Aviano, based on previous multicenter pharmacogenomic studies [43,44]. Eligible criteria were: stage II-III CRC, radiologically-confirmed absence of distant metastasis, age >18 years, performance status (WHO) 0-2, normal bone marrow, renal and liver function, and Caucasian ethnicity. Overall patients after diagnosis underwent primary surgery and received ADJ-CT based on fluoropyrimidine (FL) (i.e., 5-fluorouracil/folinic acid or capecitabine) [44], or FL plus oxaliplatin (FL+OXA) [43].
ADJ-CT was continued until completion of the planned cycles, recurrence, toxicity or patient refusal. Patients follow-up was measured from the time of surgery to the last contact or disease recurrence. Biological tests, pulmonary X-ray, positron emission tomography (PET) and/or computed tomography (CT) imaging alternatively with abdominal ultra-sonography were carried out every 3 months during the first 3 years after surgery. In the next 2 years PET/ CT were performed every 6 months and then annually. Overall evaluations were conducted independently by the type of ADJ-CT. Recurrence was defined based on PET/CT scans in the case of metastasis presence, with pathologic confirmation made by the oncologist when necessary.

HLA-G 3'UTR genotyping
A peripheral blood sample was collected in acid citrate dextrose (ACD) tubes from nearly all CRC patients. Genomic DNA was extracted from whole blood or from normal colon mucosa tissue by using the EZ1 DNA Blood or Tissue kit and the BioRobot EZ1 Workstation (QIAGEN Inc., Valencia, CA, USA). The 3'UTR of the HLA-G gene was amplified by polymerase chain reaction (PCR) using the already published [23]  . The PCR cycles were as follows: 5 mins. of initial denaturation at 94°C, 30 cycles of 45 secs. at 95°C, 45 secs. at 56°C, 60 secs. at 72°C, and the final extension step at 72°C for 7 mins. Five microliters of PCR products (344 bp in presence of deletion and 358 bp for insertion) were first analyzed by electrophoresis on 4% agarose gel stained with ethidium bromide. The remaining 25 μl of PCR reactions were purified using Diffinity RapidTip 2 tips (Sigma-Aldrich, St. Louis, MO, USA). Purified reactions (1-2 μl) were sequenced (Sanger method) by the use of the Big Dye Terminator kit (Applied Biosystems, Foster City, CA, USA) and an ABI PRISM capillary sequencer with the reverse HLAG8R primer to prevent sequence overlaps in heterozygous 14-bp samples [23]. Chromatograms were visualized with Chromas software version 2.01 and all single nucleotide polymorphisms (SNPs), and single nucleotide variants (SNVs) detected were recorded for each study participant.

Statistical analysis
The aim of this study was to assess associations between HLA-G polymorphisms in the 3'UTR regulatory region and DFS and OS respectively of stage II-III CRC patients treated with ADJ-CT after primary surgery. DFS was defined as the time from date of surgery to date of clinically detectable recurrence (local, regional or distant), death from any cause, or last followup evaluation. OS was defined as the time from date of surgery to date of death from any cause, or last follow-up time. Longitudinal analyses were determined by means of Kaplan-Meier method (log-rank test) and Cox models. Cox proportional hazard models were used to estimate adjusted hazard ratios (HRs) and corresponding 95% confidence intervals (CIs). Associations were firstly evaluated by means of univariate models and only those that resulted statistically significant (two-sided P0.05) were included in multivariate models. Adjustment for age (continuous variable), sex (male vs female), stadium (II vs III); first tumour location (colon vs rectum) and type of ADJ-CT (FL-alone vs FL+OXA) were computed. After performing a Cox regression using the common genomic model, the associations of SNPs with clinical outcomes were also evaluated for genomic models of transmission (dominant and recessive). In a dominant model for a SNP with a major allele "A" and a minor allele "b", the collective genotypes ("Ab"+"bb") are compared to a reference genotype "AA". For a recessive model, "bb" is compared to a collective ("AA"+"Ab") reference group. The HR of a reference genotype group is arbitrarily fixed at 1.00. Survival analyses were not computed when a genotype or a haplotype was detected in only one patient. Only haplotypes with frequency >1% were included in the survival analyses. An internal validation of the study results was carried out by a bootstrap resampling technique. We ran 1000 bootstrapped Cox models adjusted for the aforementioned variables. SAS software, version 9.2 (SAS Institute Inc., Cary, NC, 1999-2001) was adopted for the estimations. Adherences of genotypic proportions to expectations under Hardy-Weinberg (HW) equilibrium and two-locus linkage disequilibrium (LD) were evaluated by means of the Haploview program v4.2. The most probable haplotype of each sample at the unknown gametic phase, was reconstructed by the use of the PHASE method (program v2.1.1) [45].

Patient characteristics and survival analysis
The main demographic and clinical characteristics of CRC patients (N = 253) together with log-rank tests are summarized in Table 1.

Associations between HLA-G 3'UTR SNPs and Overall Survival
The +3187 A>G SNP was the only one to be associated with OS (log-rank P = 0.036), and in the corresponding recessive model (log-rank P = 0.025, Fig 1F) in univariate analysis (Table 1).  Table 3). The observed associations in both DFS and OS were unmodified in a bootstrap model confirming the internal validity of the associations observed (not shown).
The relation between types of polymorphisms and DFS or OS risk, was further examined stratifying by stage (II and III) (not shown). Although some differences in the hazard ratios were detected across strata, the observed associations were still confirmed. Moreover, these associations were compatible with the effect of random variation since heterogeneity tests were not significant.

Discussion
Emerging data demonstrates a key role of genes involved in immune response checkpoints and their associations with the CRC clinical outcomes [47]. In order to progress, malignant tumours must elude or evade the host's immune system. In the quest to develop personalized cancer therapies, researchers are increasingly examining the patient's immune response to cancer. SNPs within genes involved in immune response should be helpful to define the immunogenetic profile of the patients and to improve treatment strategies modulating anti-tumor immune response by targeting novel immune checkpoints. Improvement in immunosurveillance mechanisms may be achieved by means of immunotherapies with monoclonal antibodies and through chemotherapies and radiotherapies [48]. The purpose of personalized medicine is to identify the optimal treatment for each individual patient to maximize benefits and minimize adverse effects. To achieve this goal, novel informative biomarkers and new approaches to optimize clinical outcomes are needed in order to better stratify patients for cancer care.
The potential clinical relevance of HLA-G in cancer as a negative regulator due to its direct or indirect tolerogenic properties to avoid immune cells response, was previously highlighted in several studies [17,24]. However, HLA-G molecule may counteract or elicit the progression of cancer as a consequence of its immune-modulatory properties regulated by SNPs present in the untranslated regions [15]. Previous studies have reported significant associations between HLAG polimorphisms (in particular the +2960 14-bp INDEL) and cancer risk [30,34], but to the best of our knowledge, this is the first study indicating a role for HLA-G 3'UTR regulatory SNPs in DFS and OS after adjuvant treatment of CRC. Our results emphasize the role of the host's immunogenetic background in the CRC prognosis as well as report the molecular characterization of the 3'UTR region at the germinal level in subjects affected by colorectal cancer [42].
The common HLA-G 3'UTR polymorphisms were investigated and after multivariate survival analyses using Cox's regression models we found that +2960 14-bp INDEL, +3035 C>T and +3187 A>G SNPs had a significant and independent prognostic role with high internal validity by bootstrap modeling. HLAG 3'UTR SNPs previously reported to be associated with a reduced protein production such as +2960 14-bp INDEL and +3035 C>T were linked to a better prognosis, whereas the +3187 A>G SNP (increased HLAG production), was associated to worse DFS and OS.
This study shows that the +2960 14-bp INDEL SNP, already described as a disease riskmarker, is also a prognostic marker for DFS in CRC patients treated with standard ADJ-CT.
Intriguingly, our results on survival for +2960 14-bp INDEL SNP are in agreement with those observed in a non-solid tumour [31] and also in patients infected by the human immunodeficiency virus (HIV) [49], further highlighting the prognostic relevance for the 14-bp Ins/Del polymorphism. The +2960 14-bp INDEL (Ins/Del) SNP (rs371194629) was reported to modulate the magnitude of HLA-G production by regulating HLA-G mRNA stability [25]. In particular, the Del/Del genotype has been associated with high and stable HLA-G mRNA expression and higher levels of the soluble HLA-G, whereas the Ins/Ins genotype displays a lower production of mRNA and soluble or membrane bound molecules [29][30][31][32]. At the multivariate analysis, we estimated an association with a reduced DFS in patients who are carriers of Ins allele in the heterozygous Ins/Del patients and in (Ins/Del+Ins/Ins) agreement with the dominant model (HR 0.60, 95% 0.38-0.93, P = 0.023). CRC patients with the Del/Del genotype showed an increased relapse rate and reduced 5-year DFS %.
Concerning OS, we found a similar trend of improved prognosis for Ins allele even if not statistically significant probably due to the small sample size. In reference to SNP +3035 C>T (rs17179108), it has been reported, in a recent published study [29], that subjects presenting the +3035 C/T genotype had significantly lower levels of the soluble HLA-G compared to +3035 C/C (wild type) subjects. We observed a protective role of +3035 C/T genotype in the outcome (DFS) of CRC patients, though with a borderline (HR 0.51, 95% 0.26-0.99, P = 0.045) statistically significant association in the multivariate Cox's model. The +3035 C/T genotype was associated to prolonged 5-year DFS (77%) and a lower (21%) recurrence incidence (10/47) with respect to the wild type +3035 C/C combination that presented a decreased 5-year DFS (63%) and a 35% of relapse (72/205). CRC carriers for +3035 C/C had a decreased 5-year DFS % and an increased relapse incidence. It should be pointed out that the +3035 C/T heterozygous genotype, detected in 47 patients, was always associated to the Ins allele, in heterozygous Ins/Del (N = 27) and in homozygous Ins/Ins (N = 20) patients. Even if a protective HR in DFS resulted in the combined analysis of +3035 C/T and +2960 14-bp INDEL polymorphisms, no firm conclusion about a multiplicative or additive effect of the 2 SNPs cannot be inferred from this study. Regarding the +3187 A>G polymorphism (rs9380142), the +3187A allele has been associated to decreased HLA-G expression and the presence of +3187 G/G genotype to significantly increased soluble levels of HLA-G [29,50]. To date, no association with survival and this HLA-G SNP was reported. We found an association between +3187 G/G carriers (HR 2.61, 95% CI 1.24-5.50, P = 0.012), and according to the recessive (G/G vs A/A+A/G) model (HR 2.46, 95% CI 1.19-5.05, P = 0.015) with reduced DFS. Similarly, the +3187 G/G carriers were associated to a reduced OS (HR 2.96, 95% CI 1.22-7.15, P = 0.016), also in the recessive model (HR 2.71, 95% CI 1.16-6.63, P = 0.022). These results may highlight that the modulation of the clinical outcome in CRC patients harbouring the +3187 A>G change is due to the contribution of the G allele in double dose. CRC patients carriers for +3187 G/G mutated genotype (N = 15) had increased recurrence rate and a reduced 5-year DFS %. Carriers of +3187 G/G mutated combination were also associated to a reduced 5-year OS percentage and increased mortality rate. The opposite prognostic associations found in CRC patients for SNPs +2960 14-bp INDEL and +3187 A>G, are corroborated by the evidence that these two polymorphisms are not in LD (r 2 = 0.27). Furthermore, presence of +3187G allele is always associated to the Del allele, which are represented in the reconstructed UTR-1 haplotype. The latter, when present in double dose (UTR-1/UTR-1), has been associated to an unfavourable prognosis such as +3187 G/G genotype.
The variation sites described in the 3'UTR are mainly arranged in haplotypes (known as UTR-1 to UTR-44) with the UTR-1 and UTR-2 as the most frequent in the worldwide population [46]. Therefore, we performed a haplotype analysis on HLA-G 3'UTR variants to test whether haplotypes are more predictive than single variants. UTR-1 (DelCTGCGCCGCGTCGCG) haplotype carrying the 14-bp Del, +3003T, +3010G, +3027C, +3035C, +3142C, +3187G and +3196C alleles, has been considered as a high expressing haplotype. In particular, individuals with the +3187 G/G genotype and thus the UTR-1/ UTR-1 combination in double dose, exhibit significant higher levels of the soluble HLA-G [29,51]. UTR-1/UTR-1 haplotype combination shares characteristics of high HLA-G producer presenting the 14-bp Del/Del, the 3142 C/C and the 3187 G/G genotypes. UTR-1 was the only HLA-G 3'UTR haplotype associated with prognosis in CRC patients in multivariate regression Cox's analysis. We found an association of reduced DFS (HR 2.53, 95% CI 1.20-5.32, P = 0.014) and OS (HR 2.82, 95% CI 1.17-6.77, P = 0.021) in CRC patients carrying the UTR-1 haplotype in double (Hom) dose. Estimations found are not surprising since patients homozygous for UTR-1 haplotype (N = 15) are the same carriers for +3187 G/G change and share the same survival pattern.
Moreover, we observed a strong LD (S1 Fig, available online) between the +3010 C>G (rs1710) and +3142 G>C (rs1063320) SNPs, consistently with the data reported for the worldwide population [46]. Both polymorphisms were not associated to DFS or OS in multivariate analysis ( Table 2). In summary, our results demonstrates an independent potential prognostic value after multivariate analysis for three HLA-G 3'UTR polymorphisms, the +2960 14-bp INDEL, the +3035 C>T, and +3187 A>G. CRC patients carrying the Ins allele (lower HLA-G producer) were associated with a favourable DFS with a reduced risk of relapse (protective prognostic role). CRC patients carriers for the +3187 G/G genotype and UTR-1/UTR-1 haplotype (higher HLA-G producer) were associated with an unfavourable prognosis in both DFS and OS. Furthermore, patients carrying the +3035 C/T genotype (lower HLA-G producer) and therefore the Ins allele (in heterozygous or homozygous state) were associated with an improved prognosis though with a borderline significant association.
Finally, this study supports associations between the non-metastatic colorectal cancer outcome after CT treatment and polymorphisms of a gene involved in immune tumour escape. Our preliminary findings share a functional rationale considering that patients with higher levels of HLA-G would be more immunosuppressed and have a worse clinical prognosis as postulated by Rizzo et al [31]. The effect of immune surveillance on the outcome of patients after chemotherapeutic treatments is one of the currently attracting issues in cancer therapy. It is well recognized that the individual variability of drug response depends also on the genetic variations in human genome, thus enforcing the concept of personalized medicine [52]. Due to the functional impact of the HLA-G protein in cancer immune contexture, and the known correlations of functional-regulatory SNPs in the 3'UTR with the HLA-G protein level, the concept of germline genomic variation is very attractive. Only a small sample of blood is required for the genetic test, and the genotyping process is a standard and common method used in clinical practice. In addition, it should be especially considered for inoperable patients. Future perspectives will be analyze HLA-G 3'UTR polymorphisms in the outcome of metastatic CRC patients. These results could provide new insights to better stratify patients and also for combination therapy between CT and monoclonal antibodies (i.e. cetuximab and bevacizumab) currently adopted in clinical practice.
Some limitations of this work should be highlighted. Despite an internal validation (bootstrap analysis) confirming the results, HLA-G 3'UTR SNPs should be analyzed for their prognostic role in an independent CRC cohort treated with surgical resection and without any chemotherapy. Moreover, our data should be explored considering also the contribution of validated prognostic and predictive biomarkers for CRC. Another limit is the lack of biological samples such as tumour or plasma samples to correlate genetic data with tissue and/or soluble levels of HLA-G to sustain functional hypothesis about the regulatory role of these SNPs, even if previous published studies support consistent data [29][30][31][32]51]. Nonetheless, these novel findings presented here for the role of HLA-G 3'UTR region in prognosis of colorectal cancer provide the basis for implementation of personalized cancer treatments. Identification of high risk patients with well-recognized prognostic, predictive and novel immune-related genomic biomarkers may represent a new frontier in the management of CRC patients. In conclusion, +3035 C>T and in particular, +2960 14-bp INDEL and +3187 A>G polymorphisms in the regulatory 3'UTR of the HLA-G gene, have emerged as novel prognostic biomarkers in determining survival outcome in colorectal cancer. Therefore, our exploratory findings should be verified in independent larger CRC cohorts as well as further relevant functional studies are required.