Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Pregnancy Prediction in Single Embryo Transfer Cycles after ICSI Using QPCR: Validation in Oocytes from the Same Cohort

Pregnancy Prediction in Single Embryo Transfer Cycles after ICSI Using QPCR: Validation in Oocytes from the Same Cohort

  • Sandra Wathlet, 
  • Tom Adriaenssens, 
  • Ingrid Segers, 
  • Greta Verheyen, 
  • Lisbet Van Landuyt, 
  • Wim Coucke, 
  • Paul Devroey, 
  • Johan Smitz
PLOS
x

Abstract

Cumulus cell (CC) gene expression is being explored as an additional method to morphological scoring to choose the embryo with the highest chance to pregnancy. In 47 ICSI patients with single embryo transfer (SET), from which individual CC samples had been stored, 12 genes using QPCR were retrospectively analyzed. The CC samples were at the same occasion also used to validate a previously obtained pregnancy prediction model comprising three genes (ephrin-B2 (EFNB2), calcium/calmodulin-dependent protein kinase ID, stanniocalcin 1). Latter validation yielded a correct pregnant/non-pregnant classification in 72% of the samples. Subsequently, 9 new genes were analyzed on the same samples and new prediction models were built. Out of the 12 genes analyzed a combination of the best predictive genes was obtained by stepwise multiple regression. One model retained EFNB2 in combination with glutathione S-transferase alpha 3 and 4, progesterone receptor and glutathione peroxidase 3, resulting in 93% correct predictions when 3 patient and treatment cycle characteristics were included into the model. This large patient group allowed to do an intra-patient analysis for 7 patients, an analysis mimicking the methodology that would ultimately be used in clinical routine. CC related to a SET that did not give pregnancy and CC related to their subsequent frozen/thawed embryos which ended in pregnancy were analyzed. The models obtained in the between-patient analysis were used to rank the oocytes within-patients for their chance to pregnancy and resulted in 86% of correct predictions. In conclusion, prediction models built on selected quantified transcripts in CC might help in the decision making process which is currently only based on subjective embryo morphology scoring. The validity of our current models for routine application still need prospective assessment in a larger and more diverse patient population allowing intra-patient analysis.

Introduction

Single embryo transfer (SET) is the preferred treatment to limit multiple pregnancies after ART. In order not to compromise the carry home baby rate, the selection of the embryo for transfer in the first cycle becomes even more important. Next to the existing criterion based on morphology, other methods are currently under investigation. The use of quantitative gene expression measurements in cumulus cells (CC), which are in close contact with the oocyte during growth and maturation, seems a promising method [1]. Since the first published study on the subject where CC expression could be related to embryo development [2], several other studies have investigated this possibility and try to relate CC expression to different endpoints. Examples of endpoints investigated are: embryo development [3], [4], [5], [6], [7], [8], aneuploidy stage of the oocyte [9], oocyte nuclear maturity stage [10] and probably the most important from a patient perspective: pregnancy outcome [11], [12], [13], [14], [15]. Confirmation of results between different studies does not seem obvious in the analysis of CC gene expression. In the current literature not many genes were found in common in different studies. For example, hyaluronan synthase 2 (HAS2) was higher expressed in good quality embryos compared to low embryo morphology in two studies [2], [4], but could not be related to embryo morphology in two other studies [8], [13]. Divergences can be due to a different experimental design, with different endpoints, but gene expression can be influenced by known factors as the stimulation protocol of the patients [16], [17], [18] or not yet assessed factors such as culture media used in the different IVF laboratories.

In this study, 47 individual cumulus complexes from 47 intra-cytoplasmic sperm injection (ICSI) patients were retrospectively analyzed with quantitative real-time polymerase chain reaction (QPCR). Using the current sample set, a pregnancy prediction model from a previous study [15] was validated for its predictive power. In a next step, in an attempt to search for new genes with a stronger predictive power, new multivariable models were built considering the 3 genes (ephrin-B2 (EFNB2), calcium/calmodulin-dependent protein kinase ID (CAMK1D), stanniocalcin 1 (STC1)) described earlier and 9 novel genes (glutathione reductase (GSR), glutathione peroxidase 3 (GPX3), glutathione S-transferase alpha 3 and 4 (GSTA3 and GSTA4), transforming growth factor beta 1(TGFB1), progesterone receptor (PGR), inositol 1,4,5-trisphosphate receptor type 1 (ITPR1), solute carrier family 2 (facilitated glucose transporter) member 1 (SLC2A1) and thrombospondin 1 (THBS1)) (inter-patient analysis) (see Table 1 for an overview of all genes).

Our patient sample set allowed for an analysis never reported before in literature: CC from oocytes that did not result in pregnancy in the fresh transfer cycle and the CC from their sibling oocytes that resulted in pregnancy after a frozen embryo transfer (FRET) cycle were analyzed (intra-patient analysis). To our knowledge this is the first study to compare pregnant and non-pregnant CC from the same retrieval cycle in a SET setting as would be done in a final clinical application.

Materials and Methods

Patient population

This study was approved by the Ethical Committee of the UZBrussel and the patients written consent was obtained. Forty seven ICSI patients were selected based on the embryo transfer policy (single embryo transfer) and ovarian stimulation protocol prescribed: GnRH antagonist in combination with recombinant follicle stimulating hormone (FSH) (Gonal-f, Merck-Serono, Geneva, Switzerland; n = 4 or Puregon, MSD, Oss, The Netherlands; n = 43). Causes of infertility were: male factor only (n = 19), female factor only (ovulation disorder n = 3 and tubal infertility n = 2), combination male and female factor (OAT and endometriosis n = 3) and idiopathic (n = 20). Twenty patients had single embryo transfer on day 3 of culture, from these 10 became pregnant. Twenty seven patients had transfer on day 5, from these 9 became pregnant. The day of transfer was decided by the consulting doctor before oocyte retrieval took place taking into account the age of the patient or by the embryologist according to the number of good embryos available. All pregnancies resulted in live births.

Collection of human cumulus cells and embryo culture

Vaginal ultrasound was used to monitor follicular development. The endocrine profile was monitored by analysis of serum 17β-estradiol (E2), progesterone, FSH and LH by electrochemiluminescence on a COBAS 6001 immunoanalyser (Roche, Roche Diagnostics, Mannheim, Germany) using validated assays with respectively sensitivities of 5 ng/l, 0.03 µg/l, <0.1 IU/l, 0.1 IU/l and total imprecisions (%CV) of respectively <6, <7, <6 and <6. Final follicular maturation was induced with a dose of 10 000 IU hCG when at least three follicles of 17 mm in diameter were observed by transvaginal ultrasound. Oocyte retrieval was done 36 h later. CC collection was done as described in [14]. Briefly, individual oocyte denudation was performed in 40 µl droplets of HTF-SSS containing 80 IU/ml Cumulase (MediCult, Lyon, France) for not longer than 30 s and washed sequentially in droplets without enzyme. At any time, oocytes were handled individually from this point onwards in order to allow retrospective analysis of the CC per oocyte. After denudation, the CC were plunged directly in liquid nitrogen. ICSI was performed as described previously [33] and embryos were cultured in sequential media of SAGE (CooperSurgical, Leisegang Medical, Berlin). Embryos were vitrified on day 3 or day 5/6 of embryo culture as was described earlier [34] and used in a subsequent FRET cycle. The day 3 embryos were warmed on cycle day 3, cultured overnight and transferred as a day 4 embryo on cycle day 4 ( = synchronized transfer). The day 5 (and day 6) blastocysts were warmed in cycle day 5 (or day 6) in the morning and transferred on the same day.

For all 47 patients, only the CC related to those oocytes resulting in embryos selected for transfer were analyzed exception made for 7 of the 28 non-pregnant patients, 1 extra CC sample (except for 1 patient, 2 CC samples) related to a vitrified embryo giving pregnancy after a frozen single-embryo transfer cycle, was analyzed (8 extra CC in total from 7 patients). Figure 1 shows the different samples used for each analysis.

thumbnail
Figure 1. Overview of the samples used in this study.

This figure represents the distribution of the samples used for the different analyses performed in this study. The grey background fields delimit the samples that were used for the specific analyses which are marked on the left side of the field. SET: single embryo transfer; FRET: frozen embryo transfer cycle; rFSH: recombinant Follicle Stimulating Hormone.a:[15].

https://doi.org/10.1371/journal.pone.0054226.g001

Gene selection

This study is the 3rd one in a row to evaluate the predictive value of CC gene expression for oocyte quality using QPCR. Over the 3 studies, we followed a precise strategy to choose which genes to analyze regarding to oocyte quality in ICSI patients. The first study identified 4 top genes, 2 predictive for embryo morphology (inositol-trisphosphate 3-kinase A (ITPKA) and transient receptor potential cation channel, subfamily M, member 7 (TRPM7)) and 2 for pregnancy outcome (syndecan 4 (SDC4) and versican (VCAN)). It was chosen to include those 4 genes in the next study [15]. ITPKA and TRPM7 were again related to embryo development, but SDC4 and VCAN were not retained this time in the pregnancy models and were replaced by EFNB2, CAMK1D and STC1 when using stepwise multiple regression analysis. As several studies showed no relation between the genes predictive for embryo development and pregnancy outcome [13], [15], it was decided to repeat only the 3 pregnancy related genes in this third study. Our hypothesis is that by this ‘cascade’ testing strategy, the strongest pregnancy predictive genes may be filtered out through the consecutive studies (Table 2). The other 9 genes analyzed in this study were: GSTA3, GSTA4, GPX3, GSR, ITPR1, SLC2A1, THBS1, TGFB1 and PGR (Table 1). The 9 new genes were chosen from own unpublished array data comparing CCs related to pregnancy versus CCs related to no pregnancy.

thumbnail
Table 2. Strategy over 3 studies to obtain the strongest quality related genes on stored cumulus cells from ICSI patients.

https://doi.org/10.1371/journal.pone.0054226.t002

RNA extraction and cDNA synthesis

Total RNA was extracted as described earlier [17] using the RNeasy Micro Kit (Qiagen, Westburg, Leusden, The Netherlands) including a DNase step and addition of 5 ng/µl poly(dA) (Roche Applied Science, Mannheim, Germany) prior to extraction. Extraction was followed by a second DNase treatment (RQ1 RNase-Free DNase, Promega, Leiden, The Netherlands).

Reverse transcriptase (RT) was done as previously described [17] with the iScript cDNA Synthesis Kit (Bio-Rad Laboratories, Ghent, Belgium). Negative controls were generated by omitting the enzyme or the RNA in the RT reaction.

Real-time PCR

Primer sequences for CAMK1D, STC1 and EFNB2 are listed in Wathlet et al 2012. Primers for GPX3, GSTA3, GSTA4, PGR, THB1, ITPRA, SLC2A1, GSR and TGFb1 can be found in Table S1. Both beta-2-microglobulin (B2M) and ubiquitin C (UBC) were validated and used before as normalization factor [14]. Cycling conditions, negative controls, standard curves and normalization (with B2M and UBC) are as described earlier [14], but all PCR reactions were adapted to 10 µl reactions. All values mentioned hereafter are the normalized values to the mean of both B2M and UBC for each sample.

Statistics

Inter-patient analysis.

In a first analysis, a two-tailed t-test (GraphPad Prism version 4.01 for Windows, GraphPad Software, San Diego California USA) was used to compare cumulus complexes of oocytes resulting in pregnancy or not (19 live birth, 28 non-pregnant). All data were LOG transformed to obtain normal distribution and only P-values <0.0042 were considered significant after Bonferroni correction.

In a second analysis, a model was built using stepwise multiple regression analysis as described earlier [14]. Briefly, a linear regression model, with an equation as output ‘y = a + bx + cz + ds +et + fu + gv + hw’, was built with as response variables (x, z, s, t, u, v, w) gene expression and/or patient and cycle characteristics (all are listed in Table 3). ‘b–h’ are the respective indexes of the included variables and ‘a’ is the intercept of the equation. A variable was added to the model if the type III P-value of the variable was <0.3 and if the P-value of the model was improved. At the end of the model a backwards regression step was performed to exclude redundant variables. Four different models were built to predict pregnancy. In two models only gene expression values were allowed (first the model was restricted to 3 genes, next all genes were allowed into the models as long as they improved the P-value of the model). In two other models, the need for correction by patient and cycle characteristics was assessed by allowing all patient and cycle characteristics to the models only composed of genes, when those extra variables could improve the model. By introducing those extra factors, possible inter-patient variability on gene expression could be leveled out and increase the differences related to oocyte quality.

For all models, accuracy ((True positive + true negative)/(true positive + false positive + false negative + true negative)) and positive and negative predictive values (PPV =  True positive/(true positive + false negative) and NPV  =  True negative/(true negative + false positive) were calculated. The power of all models was represented by Receiver Operating Characteristic (ROC) and the area under curve (AUC) was calculated.

Intra-patient analysis.

Finally, the study allowed comparing 2 (or 3) oocytes originating from one oocyte retrieval cycle with known pregnancy outcome per oocyte, as all embryos were transferred individually in consecutive cycles. For this purpose, from 7 of the 28 patients that were not pregnant in the fresh cycle, the CC of the embryos that were replaced in a subsequent frozen single embryo transfer cycle resulting in pregnancy were compared to those transferred in the fresh cycle. For one patient, two consecutive frozen embryo replacement cycles were analyzed as the first embryo did not end in a pregnancy. Seven genes were chosen for this analysis based on their presence in one of the above mentioned models or their P-value of addition, when first added to a pregnancy model. A paired t-test was performed for each gene and the chance to pregnancy was calculated with the earlier defined models from the inter-patient analysis, containing only genes and built on the 47 CC samples (19 live birth), but excluding the 8 CC samples from the frozen cycles.

Results

Patient population

No statistical differences were found between the patient and cycle characteristics of the pregnant and the non-pregnant groups (Table 3). Mean age and BMI for both groups of patients was low but comparable. Progesterone levels were low in both groups. Mean cycle number attempt was not different in both groups: respectively in the pregnant and non-pregnant group 12 and 17 patients underwent their first cycle, 6 and 10 their second cycle and 1 for both groups their third cycle. Percentages of oocyte maturity and fertilization were normal in both groups, and the percentage of good quality embryos on day 3 was more than 50%. Embryo quality score on moment of transfer was not different for both groups. Nineteen fresh SET cycles resulted in live birth.

Pregnancy prediction

Validation of predictive genes (CAMK1D, EFNB2 and STC1) using a model built on a previous sample set.

A model predicting pregnancy composed of 3 genes, considered in a previous study [15], was tested in the current, independent patient series. The gene expression values for CAMK1D, EFNB2 and STC1 of the CC of the 47 patients were introduced in the equation obtained before and gave a value predicting the chance to pregnancy for each of the 47 oocytes. The obtained PPV and NPV calculated with the 47 samples of this study were 62% and 86%, with an accuracy of 72%.

Inter-patient analysis: t-test of all 12 genes in the current sample set.

The cumulus complexes of 28 oocytes not resulting in pregnancy were compared to 19 CC of oocytes that resulted in a live birth. Only EFNB2 was statistically higher in the pregnant group. CAMK1D, GSTA4 and GSR only showed a trend of higher expression in the pregnant group (respective P-values: 0.0068, 0.0123 and 0.0507). Graphs for all genes can be found in Figure 2.

thumbnail
Figure 2. t-test of normalized gene expression values of non-pregnant versus live birth related cumulus cell samples.

The graphs represent the differences in gene expression between the cumulus cell samples associated to an oocyte that after in vitro fertilization treatment resulted in a live birth (n = 19) or not (n = 28). Normalization was done to the mean of B2M and UBC. Only P-values <0.0042 were considered significant after Bonferroni correction. The total range of expressions found is depicted by the boxes and whiskers respectively representing the two inner and the two outer quartiles with centrally the median.

https://doi.org/10.1371/journal.pone.0054226.g002

Inter-patient analysis: Stepwise multiple regression analysis.

In a first step to build a pregnancy model, the P-value of addition when added as first variable was calculated for all genes and can be found in Table 4.

First, the model was restricted to the inclusion of 3 of the 12 genes (Model 1). EFNB2, GSTA4 and PGR were retained and gave a model with a P-value of 0.0015, a PPV of 68%, a NPV of 79% an accuracy of 73% and an AUC of 0.82. When trying to improve this model by also allowing patient and cycle characteristics (from Table 3), no improvement on the previous model was found (Model 1 bis).

In a next step, more than 3 genes were allowed into the model if improving the P-value (Model 2). Five of the 12 genes were retained in this model (i.e. EFNB2, GSTA4, PGR, GPX3 and GSTA3) which yielded a P-value <0.0001 with a PPV of 78%, a NPV of 83%, an accuracy of 81% and an AUC of 0.93. Adding patient and cycle characteristics improved the model (Model 3). The retained parameters were: days of stimulation, relative E2 and age. The PPV, NPV and accuracy of the extended model all increased to 93% and the AUC to 0.95 (Table 5). Full mathematical models can be found in Table 6. ROC curves are shown in Figure 3.

thumbnail
Figure 3. Receiver operating characteristic (ROC) curve of the 3 pregnancy models.

Multivariable models were built to predict the chance to pregnancy including the gene expression levels measured in cumulus cell samples associated to an oocyte that after in vitro fertilization treatment resulted in a live birth or not. For Model 3 patient and cycle characteristics were also included (from Table 3). Model 1 was limited to 3 genes and is composed of EFNB2, PGR and GSTA4. In Model 2 all genes were allowed into the model as long as they could improve the model. Five genes were retained for Model 2: EFNB2, PGR, GSTA4, GSTA3, GPX3. To try to improve Model 2, in Model 3 patient and cycle characteristics were allowed into the model if they could improve the P-value of the model: EFNB2, PGR, GSTA4, GSTA3, GPX3, age, Relative E2, and number of days of ovarian stimulation. The respective areas under the curve are 0.82, 0.93 and 0.95. Relative E2: E2 value measured on day of hCG over the number of cumulus oophorus complexes.

https://doi.org/10.1371/journal.pone.0054226.g003

thumbnail
Table 5. Schematic overview of the multivariable models for live birth prediction.

https://doi.org/10.1371/journal.pone.0054226.t005

Intra-patient pregnancy prediction.

For seven patients, cryostored CC samples related to cryopreserved embryos which had led to a clinical pregnancy after transfer in a subsequent transfer cycle were analyzed. This material was used to analyze the genes present in the above obtained multivariable models (Table 5) and/or the 5 genes with the smallest P-value of addition (Table 4) i.e. CAMK1D, EFNB2, GPX3, GSR, GSAT4, GSTA3 and PGR. As a first explorative method a paired t-test for those seven genes was performed and can be found in Figure S1. Five genes had an upwards trend in the CC of an oocyte resulting in pregnancy (P-value <0.05; only P-values <0.07 are significant after Bonferroni correction for multiple comparisons): EFNB2 (only significant difference between pregnant and non-pregnant), CAMK1D, GSR and PGR. GSTA4, GPX3 and GSTA3 had all P-values >0.05. Fold changes between the CC from the same patient were calculated and an average per gene was made (from 1.1 to 2.5) (Table 7). For the five genes with a P-value <0.1 in the paired t-test, the percentage of correctly estimated CC was 71% to 88%, based on the level expected (e.g. higher expression is expected in the pregnant compared to the non-pregnant related CC) from the paired t-test. In a next step the predictive power of the multivariable models (the 3 gene model ( = Model 1) and the 5 gene model ( = Model 2)) obtained in the inter-patient analysis (47 CC samples) was used to rank the CC of each patient (Table 5) for their chance to pregnancy. As all patient and cycle characteristics are identical for oocytes within a single retrieval cycle it makes obviously no sense to use models containing those variables. All CC, except for patient 4, were correctly ranked (Table 7) for their chance to pregnancy.

thumbnail
Table 7. Comparison of gene expression levels of fresh cycles not resulting in pregnancy to frozen transfer cycles resulting in pregnancy (intra-patient analysis).

https://doi.org/10.1371/journal.pone.0054226.t007

Discussion

The expression of 12 genes in the CC and their capacity to predict the pregnancy potential of the oocyte they surrounded were analyzed in the present study. Three genes (CAMK1D, EFNB2 and STC1) were included from a previous study [15], where they were coming out as the most predictive ones for pregnancy prediction. EFNB2 was also significantly up-regulated in the current study in the CC from the oocytes that gave pregnancy and CAMK1D showed the same trend (P = 0.0068). The model composed of genes only from the previous patient dataset, using CAMK1D, EFNB2 and STC1, was validated in this independent patient group and yielded a similar overall performance of the model, with accuracies of 72% in the current study and 79% in the previous study.

Besides the EFNB2 and CAMK1D coming from our previous study, two newly studied genes, GSTA4 and GSR, also showed an up-regulated trend in the CC of the pregnant group.

A multivariable approach tested whether it would be possible to predict live birth by using only 3 of the 12 tested genes in the current sample set (inter-patient analysis). The 3 most predictive genes were GSTA4, PGR and EFNB2 and resulted in a similar accuracy as the previous model with EFNB2, CAMK1D and STC1 (73% versus 72%). Using three genes, none of the patient or cycle characteristics could improve the model, suggesting that the expression of the three genes was minimally influenced by patient and cycle factors. The live birth model could be improved by including two more genes (GPX3 and GSTA3), which increased the accuracy up to 81% and resulted in an AUC of 0.93. This 5 gene model was improved with three patient and cycle characteristics (age, relative E2 and days of stimulation) and resulted in an optimized model with a PPV, NPV and accuracy of 93%, but similar AUC as the model only containing genes. Ideally, pregnancy prediction based on gene expression should be possible with a limited set of genes to reduce analysis time and cost. Of the 12 genes tested in this patient population, the 2 most recurrent genes are GSTA4 and EFNB2 (see Table 5). These 2 genes are present in all 3 pregnancy models and have respectively in 3 out of 3 and in 2 out of 3 predictive models a type-III P-value <0.01. The models also showed that the gene expression values are always more important than the patient and cycle characteristics, as the type-III P-values in the models are only significant for the genes and not for the patient and cycle characteristics. Furthermore, the AUC results are comparable between the model containing only genes and the model combining genes and patient and cycle characteristics.

For 7 patients, 2 (or 3) CC from oocytes giving embryos with a good morphology and consecutively transferred, were analyzed (intra-patient analysis). For all oocytes, pregnancy outcome was known as all embryo transfers were done in subsequent SET cycles. Having assessed the expression of 7 genes (EFNB2, CAMK1D, GSR, PGR, GSTA4, GSTA3 and GPX3), it was easy to estimate, based on single genes or on a combination of them in a model, which oocyte of the two considered ones would have the highest chance to pregnancy. The best predictions, when looking at each gene individually, were found for CAMK1D, GSR, EFNB2, PGR and GSTA4 with 71% to 86% correct estimations. Combining 3 or 5 genes, according to the earlier models, 6 of the 7 patients had the cumulus complexes correctly predicting the chance to pregnancy. This prediction was the same when using the 3 or 5 gene model. Surprisingly, the multivariable model failed on a different patient than the single gene analysis. The multivariable approach seems stronger than the individual genes as all genes wrongly predicted the pregnancy outcome for the second comparison of patient 6, but the multivariable model predicted successfully, especially as the CC of the oocytes that resulted in pregnancy were not considered for building the predictive model. The failed prediction of the multivariable approach for patient 4, where the individual genes were correct in 4 out of 5, could be due to other factors (e.g. the endometrium status at the moment of transfer). Although this analysis is still limited in patient numbers, these results are encouraging. This is the first study that confirms that some genes predicting pregnancy between patients might also be capable of ranking the quality of oocytes within patients using a multivariable approach and providing a chance to pregnancy for each oocyte. CC gene expression analysis might become a valuable tool in the ART lab but does obviously not take into account the eventual influence of poor sperm quality and out-of-phase endometrium.

This study could confirm the use of earlier found predictive genes in a new patient population, with the same stimulation protocol, but with different culture media. This data suggests that EFNB2 and CAMK1D, the only genes that we analyzed in two studies using two different culture media, were not affected by culture media. To test the validity of the current models, a future analyze should involve patients with different stimulation protocols and different culture media. Reasons why the specific genes of this study are important for pregnancy prediction remains speculative. As mentioned previously [15], CAMK1D may, among other pathways, be related to steroidogenesis by its strong correlation with steroid related genes (CYP11A1, STC2 and HSD3B1). In this data set CAMK1D also strongly correlated with a steroid related gene i.e. PGR next to EFNB2 and GSTA4 (all P<0.0001 with Pearson correlation analysis) still leaving the possibilities open for more than one pathway to which CAMK1D could be associated with. The presence of PGR in the pregnancy models and its predictive power within patients might indicate that steroid-related genes could be helpful in pregnancy prediction. PGR has been described before, but no difference between pregnant and non-pregnant could be confirmed with quantitative PCR [28]. GSTA3 that was present in some of the pregnancy models also has a link with progesterone production [24] reiterating the importance of the steroidogenesis pathway. The function of EFNB2 in the ovary is not yet known, but B-class ephrins were proposed to be related to the luteinization process [19] and the ephrin B2 receptor was found differently expressed between CC from normal oocytes compared to aneuploid oocytes [9]. Out of the other genes tested, only members of the glutathione family were retained in the models or were significant in the t-test. Glutathione enzymes are important for detoxification actions (of free radicals) in the cells through the use of glutathione. Hypoxia leads to the formation of reactive oxygen species (ROS) which can cause lipid peroxidation, enzyme inactivation and cell damage, resulting in apoptosis [35] not only in CC, but also in the oocyte [36]. Oxidative stress has already been reported by other groups as a possible target to assess oocyte quality [7], [37]. One reason that some transcripts of those pathways are higher in the CC from oocytes giving pregnancy might be that those oocytes are better protected against a stressful environment if needed. Before knowing what are exact role and importance of the considered genes in this study, more experiments need to be performed using animal models to be able to access the antral growth and periovulatory period since in human only 1 time point (i.e. 36h after hCG) is available for analysis.

Conclusion

By testing presumably important genes batch wise in three consecutive studies on different patient groups for which cumulus complexes were frozen individually per oocyte, we retained the most predictive genes for pregnancy and opposed these every time to new candidate genes. This ‘cascade’ strategy attempted to increase the power of pregnancy prediction using CC gene expression as quality marker for oocytes in ART. The strategy proved effective as the model with EFNB2, CAMK1D and STC1 from the second study [15] could be confirmed on an independent patient sample set. In an attempt to further improve prediction models for live birth, models were built (inter-patient analysis), still retaining EFNB2 together with PGR and genes related to the glutathione metabolism. The new models proved to be able to rank oocyte for their potential for pregnancy (intra-patient analysis). The validity of our current models, for routine application, still need prospective assessment in a larger and more diverse patient population allowing intra-patient analysis.

Supporting Information

Figure S1.

Paired t-test for the intra-patient analysis pregnant versus non-pregnant. The graphs compare for each patient the gene expression of a cumulus complex that corresponded to an oocyte that did not result in pregnancy to one that resulted in pregnancy in a subsequent single embryo frozen transfer cycle. Per patient the oocytes originate from one retrieval cycle. One patient had 2 consecutive frozen cycles, the first one not resulting in pregnancy. One color represents one patient. The dashed lines show (only in the graphs with a major trend: up or down from non-pregnant to pregnant with P<0.1) the pairs not following the major trend. Those pairs are also marked with ‘a’ in Table 7.

https://doi.org/10.1371/journal.pone.0054226.s001

(PDF)

Acknowledgments

The authors thank their colleagues from the Centre for Reproductive Medicine of the UZ Brussel.

Author Contributions

Selection of the patients, scored the embryos and revised the manuscript: LVL GV. Supervised the work at the clinical centre: PD. Conceived and designed the experiments: SW TA JS. Performed the experiments: SW. Analyzed the data: SW TA IS WC. Wrote the paper: SW TA JS.

References

  1. 1. Huang Z, Wells D (2010) The human oocyte and cumulus cells relationship: new insights from the cumulus cell transcriptome. Mol Hum Reprod 16: 715–725.
  2. 2. McKenzie LJ, Pangas SA, Carson SA, Kovanci E, Cisneros P, et al. (2004) Human cumulus granulosa cell gene expression: a predictor of fertilization and embryo selection in women undergoing IVF. Hum Reprod 19: 2869–2874.
  3. 3. Hasegawa J, Yanaihara A, Iwasaki S, Otsuka Y, Negishi M, et al. (2005) Reduction of progesterone receptor expression in human cumulus cells at the time of oocyte collection during IVF is associated with good embryo quality. Hum Reprod 20: 2194–2200.
  4. 4. Cillo F, Brevini TA, Antonini S, Paffoni A, Ragni G, et al. (2007) Association between human oocyte developmental competence and expression levels of some cumulus genes. Reproduction 134: 645–650.
  5. 5. Zhang X, Jafari N, Barnes RB, Confino E, Milad M, et al. (2005) Studies of gene expression in human cumulus cells indicate pentraxin 3 as a possible marker for oocyte quality. Fertil Steril 83 Suppl 11169–1179.
  6. 6. Feuerstein P, Cadoret V, Dalbies-Tran R, Guerif F, Bidault R, et al. (2007) Gene expression in human cumulus cells: one approach to oocyte competence. Hum Reprod 22: 3069–3077.
  7. 7. van Montfoort AP, Geraedts JP, Dumoulin JC, Stassen AP, Evers JL, et al. (2008) Differential gene expression in cumulus cells as a prognostic indicator of embryo viability: a microarray analysis. Mol Hum Reprod 14: 157–168.
  8. 8. Anderson RA, Sciorio R, Kinnell H, Bayne RA, Thong KJ, et al. (2009) Cumulus gene expression as a predictor of human oocyte fertilisation, embryo development and competence to establish a pregnancy. Reproduction 138: 629–637.
  9. 9. Fragouli E, Wells D, Iager AE, Kayisli UA, Patrizio P (2012) Alteration of gene expression in human cumulus cells as a potential indicator of oocyte aneuploidy. Hum Reprod.
  10. 10. Ouandaogo ZG, Haouzi D, Assou S, Dechaud H, Kadoch IJ, et al. (2011) Human cumulus cells molecular signature in relation to oocyte nuclear maturity stage. PLoS One 6: e27179.
  11. 11. Assidi M, Montag M, Van der Ven K, Sirard MA (2011) Biomarkers of human oocyte developmental competence expressed in cumulus cells before ICSI: a preliminary study. J Assist Reprod Genet 28: 173–188.
  12. 12. Assou S, Haouzi D, Mahmoud K, Aouacheria A, Guillemin Y, et al. (2008) A non-invasive test for assessing embryo potential by gene expression profiles of human cumulus cells: a proof of concept study. Mol Hum Reprod 14: 711–719.
  13. 13. Gebhardt KM, Feil DK, Dunning KR, Lane M, Russell DL (2011) Human cumulus cell gene expression as a biomarker of pregnancy outcome after single embryo transfer. Fertil Steril 96: 47–52 e42.
  14. 14. Wathlet S, Adriaenssens T, Segers I, Verheyen G, Van de Velde H, et al. (2011) Cumulus cell gene expression predicts better cleavage-stage embryo or blastocyst development and pregnancy for ICSI patients. Hum Reprod 26: 1035–1051.
  15. 15. Wathlet S, Adriaenssens T, Segers I, Verheyen G, Janssens R, et al.. (2012) New candidate genes to predict pregnancy outcome in single embryo transfer cycles when using cumulus cell gene expression. Fertil Steril.
  16. 16. Adriaenssens T, Mazoyer C, Segers I, Wathlet S, Smitz J (2009) Differences in collagen expression in cumulus cells after exposure to highly purified menotropin or recombinant follicle-stimulating hormone in a mouse follicle culture model. Biol Reprod 80: 1015–1025.
  17. 17. Adriaenssens T, Wathlet S, Segers I, Verheyen G, De Vos A, et al. (2010) Cumulus cell gene expression is associated with oocyte developmental quality and influenced by patient and treatment characteristics. Hum Reprod 25: 1259–1270.
  18. 18. Grondahl ML, Borup R, Lee YB, Myrhoj V, Meinertz H, et al. (2009) Differences in gene expression of granulosa cells from women undergoing controlled ovarian hyperstimulation with either recombinant follicle-stimulating hormone or highly purified human menopausal gonadotropin. Fertil Steril 91: 1820–1830.
  19. 19. Egawa M, Yoshioka S, Higuchi T, Sato Y, Tatsumi K, et al. (2003) Ephrin B1 is expressed on human luteinizing granulosa cells in corpora lutea of the early luteal phase: the possible involvement of the B class Eph-ephrin system during corpus luteum formation. J Clin Endocrinol Metab 88: 4384–4392.
  20. 20. Verploegen S, Lammers JW, Koenderman L, Coffer PJ (2000) Identification and characterization of CKLiK, a novel granulocyte Ca(++)/calmodulin-dependent kinase. Blood 96: 3215–3223.
  21. 21. Luo CW, Kawamura K, Klein C, Hsueh AJ (2004) Paracrine regulation of ovarian granulosa cell differentiation by stanniocalcin (STC) 1: mediation through specific STC1 receptors. Mol Endocrinol 18: 2085–2096.
  22. 22. Luo CW, Pisarska MD, Hsueh AJ (2005) Identification of a stanniocalcin paralog, stanniocalcin-2, in fish and the paracrine actions of stanniocalcin-2 in the mammalian ovary. Endocrinology 146: 469–476.
  23. 23. Pastore A, Federici G, Bertini E, Piemonte F (2003) Analysis of glutathione: implication in redox and detoxification. Clin Chim Acta 333: 19–39.
  24. 24. Raffalli-Mathieu F, Orre C, Stridsberg M, Hansson Edalat M, Mannervik B (2008) Targeting human glutathione transferase A3-3 attenuates progesterone production in human steroidogenic cells. Biochem J 414: 103–109.
  25. 25. Govinden R, Bhoola KD (2003) Genealogy, expression, and cellular function of transforming growth factor-beta. Pharmacol Ther 98: 257–265.
  26. 26. Liu X, Andoh K, Abe Y, Kobayashi J, Yamada K, et al. (1999) A comparative study on transforming growth factor-beta and activin A for preantral follicles from adult, immature, and diethylstilbestrol-primed immature mice. Endocrinology 140: 2480–2485.
  27. 27. Makrigiannakis A, Coukos G, Christofidou-Solomidou M, Montas S, Coutifaris C (2000) Progesterone is an autocrine/paracrine regulator of human granulosa cell survival in vitro. Ann N Y Acad Sci 900: 16–25.
  28. 28. Hamel M, Dufort I, Robert C, Leveille MC, Leader A, et al. (2010) Identification of follicular marker genes as pregnancy predictors for human IVF: new evidence for the involvement of luteinization process. Mol Hum Reprod 16: 548–556.
  29. 29. Berridge MJ (2009) Inositol trisphosphate and calcium signalling mechanisms. Biochim Biophys Acta 1793: 933–940.
  30. 30. Olson AL, Pessin JE (1996) Structure, function, and regulation of the mammalian facilitative glucose transporter gene family. Annu Rev Nutr 16: 235–256.
  31. 31. Adams JC (1997) Thrombospondin-1. Int J Biochem Cell Biol 29: 861–865.
  32. 32. Hayashi H, Sakai K, Baba H, Sakai T (2012) Thrombospondin-1 is a novel negative regulator of liver regeneration after partial hepatectomy through transforming growth factor-beta1 activation in mice. Hepatology 55: 1562–1573.
  33. 33. Van Landuyt L, De Vos A, Joris H, Verheyen G, Devroey P, et al. (2005) Blastocyst formation in in vitro fertilization versus intracytoplasmic sperm injection cycles: influence of the fertilization procedure. Fertil Steril 83: 1397–1403.
  34. 34. Van Landuyt L, Verpoest W, Verheyen G, De Vos A, Van de Velde H, et al. (2011) Closed blastocyst vitrification of biopsied embryos: evaluation of 100 consecutive warming cycles. Hum Reprod 26: 316–322.
  35. 35. Buttke TM, Sandstrom PA (1994) Oxidative stress as a mediator of apoptosis. Immunol Today 15: 7–10.
  36. 36. Tatemoto H, Sakurai N, Muto N (2000) Protection of porcine oocytes against apoptotic cell death caused by oxidative stress during In vitro maturation: role of cumulus cells. Biol Reprod 63: 805–810.
  37. 37. Lee MS, Liu CH, Lee TH, Wu HM, Huang CC, et al. (2010) Association of creatin kinase B and peroxiredoxin 2 expression with age and embryo quality in cumulus cells. J Assist Reprod Genet 27: 629–639.