Associations of fertility parameters with fatty acids and DNA methylation in Mexican women undergoing in vitro fertilization

Fatty acids (FA) likely affect human fertility at multiple levels, as deviations from physiological FA profiles are obesogenic, and FA can modify DNA methylation (DNAm). Yet, the interplay of follicular fluid (FF) and serum FA with BMI and percentage body fat (PBF) in human fertility is not completely understood. Also, associations of DNAm with fertility are largely unexplored. Reproductive parameters ranging from retrieved oocyte number to infant birth weight, were recorded in Mexican women undergoing in vitro fertilization (n = 88). Multiple regression analysis sought BMI-adjusted and age-adjusted associations. Receiver operating characteristic analysis tested for discrimination between outcomes. Associations of FF and serum FA were markedly distinct. While various FF FA (C16:1, C18:0, C20:2, C20:3, arachidonic acid) were significantly and inversely associated only with retrieved oocyte number, selected serum FA were associated with a broad range of pre-fertilization and post-fertilization parameters. Associations of BMI and FF FA were complex, as arachidonic acid was inversely associated with both BMI and retrieved oocyte number, while oleic acid (OA) was directly associated with BMI and PBF. Ultrasound-assessed clinical pregnancy outcome (CP) was directly associated with serum OA but inversely with its trans isomer elaidic acid (EA) and with BMI. Compounded BMI, serum EA and OA discriminated CP well (AUC = 0.74). Whole blood DNA methylation was significantly associated with and a moderate predictor (AUC = 0.66) of percent fertilized oocytes. Overall FF FA pool composition rather than FA identity may impact oocyte production and cellular memory of FF FA is lost as the oocyte exits the follicular environment. The contrasting associations of BMI, FF OA and arachidonic acid suggest that the control of oocyte homeostasis by FF FA is uncoupled from BMI. Further studies are warranted to assess the potential of compounding BMI with serum EA and OA to predict CP.


INTRODUCTION
Reduced female fertility can significantly impact physical and psychological health, thus underlining the importance of understanding the mechanisms behind that condition. Assisted reproductive technology (ART) successfully mitigates infertility and additionally provides an ethically acceptable quasi-experimental human model that allows to associate oocyte and embryo development stages with physiological variables. Among the latter, lipid metabolism impacts female fertility. In particular, it has been known for almost seven decades that obesity is detrimental for female reproductive physiology in the general population, an association that was recently confirmed by a meta-analysis of obesity in women undergoing in vitro fertilization (IVF) [1][2][3]. Accordingly, obesogenic Western-style diet induces a proinflammatory follicular environment and impairs embryonic development prior to any detectable metabolic alteration in rhesus macaques undergoing IVF [4]. Supporting correlative human data, experimental rodent models have shown that an obesogenic diet alters the oocyte epigenome and reduces ovulation [5,6]. Consistent with the central role of lipid metabolism, a missense mutation in the phospholipase C gene leads to defective oocytes in humans [7].
Abnormal lipid metabolism is often concomitant with altered fatty acid (FA) profiles. Accordingly, follicular fluid (FF) and serum FA profiles are associated with body mass index (BMI) [8,9]. Moreover, analysis of the oocyte FA pool revealed that BMI is inversely and directly associated with saturated FA (SFA) and selected polyunsaturated FA (PUFA), respectively, in a Spanish cohort [10]. Growing evidence in human ART settings shows that human female fertility is associated with FA, thus hinting that the latter may mediate the effects of obesity on female fertility. Indeed, associations have been detected for serum and FF FA, either free or incorporated in phospholipids or triglycerides. Consistent data have emerged across different studies, as SFA are inversely associated with oocyte and blastocyst quality, whereas PUFA are directly associated with pregnancy outcomes in two Iranian cohorts [11,12]. Accordingly, FF C18:0, a SFA, was inversely associated with blastomere score [9]. Conversely, serum oleic acid (OA), a monounsaturated FA (MUFA) was directly associated with the number of mature oocytes [12]. However, not all PUFA show consistent associations, as serum eicosaenoic acid and FF linoleic acid were associated with favorable fertility parameters, whereas FF arachidonic acid (AA) or its metabolites showed the opposite trend [11][12][13]. FF n-3 PUFA tended to be inversely associated with larger follicles, younger age and better ovarian functionality in a Spanish cohort, but serum counterparts were positively associated with ART outcomes in a US cohort [14,15]. Several functional studies in murine models support the notion that FA profiles and fertility are linked. Restoration of n-6 and n-3 PUFA levels normalizes female fertility in mice with genetically imposed low PUFA synthesis [16]. Loss of the antioxidant glutathione, essential for fertility, leads to a decrease in oocyte lipid droplets and serum unsaturated FA [17]. Also, impaired FA metabolism coincides with high fat diet-associated ovarian dysfunction [18]. Exposure of cultured early embryos to specific FA revealed that palmitic acid (PA) slowed developmental progression and induced autophagy, while OA reversed those responses [19].
The association of FA with fertility may reflect various underlying mechanisms. For example, excess or reduction of specific FA reflect obesogenic lipid profiles, which in turn could mediate the adverse impact of obesity on fertility. In addition, FA exert molecular responses that are not necessarily related to obesity: the role of PUFA as ligands for specific nuclear receptors is well established [20], and growing evidence shows that FA can modulate the DNA methylome. AA and PA induced global DNA hypermethylation in human pancreatic and cultured THP-1 cells [21,22]. Conversely, OA and selected PUFA induced DNA hypomethylation in cultured human cells [22,23]. Gene-specific methylation studies identified loci involved in Ras signaling and in metabolism as targets for FA in vitro and in vivo [24][25][26][27]. As for peripheral blood cells, DNA methylation (DNAm) was linked to specific FA by both association and intervention studies in humans [28][29][30][31]. Also, intergenerational epigenetic memory of maternal or paternal FA exposure has been documented in animal models [32,33]. Moreover, the propensity to undergo betaoxidation varies between specific FA, as illustrated by the preferential beta-oxidation of PA and OA compared to AA [34]. Beta-oxidation generates acetyl-CoA, which can be used for histone acetylation or feeds into the Krebs cycle, a generator of cofactors for a panoply of chromatin modifiers (reviewed in ref. [35]). Therefore, the composition of the FA pool in the follicular and intrauterine environment can impact the transcription of critical genes. For example, DNAm and chromatin dynamics are highly regulated during embryonic development and specific FA may disrupt pregnancy outcome by imposing DNA hypermethylation of hypomethylation [36].
The above-mentioned published associations of fertility with BMI and FA may reflect ethnicity-specific phenomena. Also, to our knowledge associations with DNAm have not been documented. Therefore, we tested a cohort of Mexican women undergoing IVF, for associations of a broad range of fertility and reproductive parameters with serum and FF FA, BMI, and peripheral blood global DNAm. Additionally, we surveyed for the first time any association of percentage body fat (PBF) with the aforementioned variables. PBF is considered to be more variable across ethnicities than BMI and could therefore highlight Mexican population-specific associations [37]. We discuss the data in the context of the existing literature and previously documented associations in humans.

MATERIALS, SUBJECTS, AND METHODS Ethical considerations
Patient treatment, confidentiality and sample collection followed the internal ethical guidelines of the Institute of Sciences in Human Reproduction "Vida", a private clinic providing IVF services. All participating subjects signed a medical informed consent for the IVF procedure, and follicular fluid and venous blood donation. The CINVESTAV Irapuato Unit and University of Guanajuato facilities, only processed extracts from FF, serum, or whole peripheral blood (WB). Therefore, no permission from any institutional ethical committee was necessary.

Study cohort
A profile of the study cohort is shown in Table 1. A total of 88 women were selected from patients who attended the Institute of Sciences in Human Reproduction "Vida" for assisted reproduction treatment with intracytoplasmic sperm injection, from January to June 2015. Sample size was based on convenience, and exceeded the sample size of a similar study that identified statistically significant differences or associations [9]. Inclusion criteria were age ≤38 years, no diagnosed endocrine disease, normal folliclestimulating hormone defined as <10 IU/ml and partner with no severe male factor infertility. Percentage body fat was measured with a FitScan Body Composition Monitor (Tanita).

Sample collection and fertility parameters
Serum, WB, and FF were collected from all participants. Serum was obtained by centrifugation of peripheral venous blood obtained before anesthesia, at 2700 rpm for 10 min in a sterile tube and stored at −80°C. WB was collected in EDTA-coated tubes and stored at 4°C. To obtain FF, the follicle was punctured and aspired under i.v. sedation using a 17Gx20 single lumen needle (Oocyte Pick-up Needle, Kitazato) and vacuum pressure of 50 mmHg. FF was collected in a sterile tube, centrifuged at 2700 rpm for 10 min to remove residual cells, and stored at −80°C. All subsequent manipulations were performed according to standard procedures with minimal modifications [38]. Retrieved oocytes complexes were evaluated under a stereoscope to assess maturity. On the day of injection, cumulus cells were removed using hyaluronidase (InVitroCare Inc.) with mechanical pipetting and maturity was reassessed and confirmed with the presence of the first polar body as criterion to identify metaphase II mature oocytes. Denuded oocytes were injected with spermatozoa previously capacitated by gradient centrifugation (PureSperm, Nicadon International AB combined with swim-up technique. Injected oocytes were transferred to global® total® medium (LifeGlobal), covered with paraffin oil (LifeGlobal) and incubated at 37°C in a 6.5% CO 2 , 5% O 2 , 88.5% N 2 atmosphere. After 18 h, fertilization was evaluated by the presence of two pronuclei and two polar bodies. Embryos were cultured in 20 μl droplets of global® total® medium in a 6.5% CO 2 , 5% O 2 , 88.5% N 2 atmosphere, according to standard procedures and embryo transfer was performed on day 2 or 3 of embryo culture. Three embryo transfer cycles were performed: the first included all 88 patients, the second and third included 40 and 6 patients with negative clinical pregnancy (CP) outcome in the previous transfer cycle, respectively. Two-three embryos were transferred per patient (average: 2.55; n = 134, counting all embryos transferred in the three cycles). CP was confirmed when echography revealed at least one gestational sac. Women with positive CP were stratified as positive after the first embryo transfer cycle (CP1) or as compounded CP after the second and third embryo transfer cycles (CPC). The latter group consisted of patients with negative CP after the first embryo transfer cycle who received oocytes for which FA profile was available. Infant birth weight (BW) was recorded, and average BW in twin deliveries was used for statistics. In summary, the measured fertility and relevant metabolic parameters were: IMC, percentage body fat, number of retrieved oocytes, percentage mature (metaphase II) oocytes, percentage fertilized oocytes, percentage zygotes, clinical pregnancy outcome, infant birth weight.
Fatty acid profiling FA were converted to methyl esters by resuspending lyophilized FF or serum samples in Reacti-Vial™ tubes (Thermo Scientific) with 1 ml 0.5 M NaOH in methanol, 20 μl 3 mg/ml heptadecanoic acid as internal standard and incubated for 1 h at 90°C. After cooling to room temperature, 1 ml BF 3 in methanol (Sigma-Aldrich) was added followed by incubation for 30 min at 90°C. Samples were transferred to test tubes, and 2 ml deionized water and 4 ml hexane were added and mixed. The organic phase was dried under nitrogen stream and dissolved in 400 μl isooctane. Samples were analyzed by GC-EIMS, in a 7890 Gas Chromatograph coupled to a 5973 mass spectrometer detector (Agilent Technologies, Inc.) using a Zebron ZB-1MS (60 m × 320 µm × 1 µm) capillary column. One µL sample was injected in pulsed split-less mode. Injection chamber temperature was 250°C. Helium chromatographic grade was used as carrier gas with a constant flow of 1 ml/min. The GC oven program was as follows: 150°C for 3 min, then heating by 4°C/min to 300°C and hold for 20 min. The transfer line temperature was maintained at 260°C. Temperature of the ion source and the quadrupole was 230°C and 150°C, respectively. Measurements were performed in SCAN mode in a mass range of 50-550 and operated at 2.9 scans per second. Mass spectra were obtained at 70 eV. GC-EIMS data were collected with the MassHunter Workstation version B.06.00 software (Agilent Technologies, Inc.). Retention time and mass spectrum of each component were determined with the Chemstation Data Analysis software (Agilent Technologies, Inc) and the Automated Mass Spectral Deconvolution and Identification System "AMDIS" software (http://www.amdis.net/). Compounds were identified using the NIST MS mass spectra database library and Search software version 2.0 (National Institute of Standards and Technology, USA) and/or the respective standard. Individual FA levels normalized as percentage by weight were used throughout the study.
Peripheral blood global DNA methylation profiling Whole blood genomic DNA was extracted by standard protocols. 5-methylcytosine was determined with the MethylFlash Global DNA Methylation (5-mC) ELISA Easy Kit (Colorimetric) (Epigentek) according to the manufacturer's instructions.

Statistics
All percentage values were logit-transformed. Based on sample size and the normalizing effect of the logit transformation, we used linear multiple regression (regression between continuous variables) or t-test (comparisons between groups) [39]. In multiple regression, homoscedasticity and normal distribution of residuals were tested with the Breusch-Pagan-Godfrey test and the Lillefors test, respectively. Associations of FA with reproductive parameters were adjusted for age and BMI. Associations with BMI or PBF were corrected for age. Chisquare test was used to compare paired percent data. The area under the receiver operating characteristic curve (ROC-AUC) was used to assess predictive powers. The test employed in each comparison is indicated in the respective result table or figure.
All tests were performed with the StatPlus (AnalystSoft) package. All tests were two-sided.

RESULTS
As a general view of the study, 88 patients participated after complying with inclusion criteria. Pregnancy after fresh embryo transfer cycle was achieved in 47 patients with 39 live births, including 7 twin pairs. Cumulative pregnancy rate including the second and third embryo transfer cycles was 59 out of 88 patients, or 67.0% rate. The range of infant birth weight was 1750-3700 g. Overall, the obtained parameters are considered successful in IVF interventions compared with reported assisted reproduction surveys (see for example the 2019 statistics data by the Society for Assisted Reproductive Technology in the US: https://www.sartcorsonline.com/rptCSR_PublicMultYear.aspx? reportingYear=2019). All detected significant associations are shown in Fig. 1 and detailed in the following paragraphs.
Associations of FF FA with fertility parameters Among FF FA, several SFA (C18:0), MUFA (C16:1) and PUFA (C20:2, C20:3, AA) were inversely and significantly associated only with total number of retrieved oocytes, after adjusting for BMI and age ( Table 2). No significant correlation was observed for any other FF FA or fertility parameter. Non-significant associations are listed in Supplementary Table 1.
Associations of serum FA with fertility parameters Several serum FA were associated with pre-fertilization and postfertilization parameters. Among individual serum FA, the SFA PA (C16:0) and the MUFA transC18:1 (elaidic acid or EA) showed significant inverse associations with percentage mature oocytes and percentage zygotes, respectively (Table 3). Consistently, SFA showed the same direction of association as C16:0 with percentage mature oocytes. PUFA were also inversely associated with total number of retrieved oocytes. In turn, OA and its trans isomer EA were directly and inversely associated, respectively, with both CP1 and CPC (Table 3). No significant association was observed for any other serum FA or fertility parameter. Nonsignificant associations are listed in Supplementary Table 2.
Distribution of significantly associated FA in serum and FF The differential associations of FF and serum FA might reflect differences in distribution-i.e., associations would be detected preferentially in broadly distributed samples and would not reach significance in tightly clustered ones. Yet, the standard deviation normalized as percentage of the average did not reflect the differential associations of the two FA pools (Supplementary   Fig. 2). Non-significant associations of DNAm with fertility parameters are listed in Supplementary Table 3.

Associations with BMI
We tested for associations of BMI with FF or serum FA, and fertility parameters. BMI was included in the multiple regression model as independent variable, based on the following rationale. First, the causal directionality of FA and BMI is complex, as obesity is caused by excess lipids but can also modify lipid metabolism, for example by promoting mitochondrial dysfunction [40]. Additionally, biological considerations suggest that the BMI is the cause rather than the consequence if any functional association with FF FA profiles exists. Therefore, considering the BMI as the independent variable against both serum and FF FA would make the outcomes of the two association tests readily comparable. The BMI showed significant associations with selected FF FA and fertility parameters. FF AA and OA were inversely and directly associated with BMI, respectively, as was CP1 (Table 5). CPC was not tested in this case, as BMI at the second and third embryo transfer cycle was not recorded. No other significant association was detected. Non-significant associations are listed in Supplementary Table 4.
We also tested for any association between the BMI and DNAm. Since at least two studies have indicated that the BMI causes DNAm rather than vice versa, we employed a linear regression model with BMI as independent variable, adjusted for age [41,42]. No significant association was detected (Supplementary Table 4).
Predictive value of BMI and FA for CP outcome Individually, the three variables associated with CP outcomei.e., BMI at oocyte retrieval, and serum EA and OA -were weak predictors of CP outcome (AUC range: 0.61-0.62). Since our data suggested that the effects of BMI and FA were independent and potentially additive, we reasoned that a score built by composing those three variables could better predict CP outcome. Given the directions of the respective associations with unfavorable CP outcome-i.e., positive for BMI and EA, and negative in the case of OA-we calculated a score by subtracting OA from the sum of BMI and EA (BMI + EA-OA). The score will be referred to as BEO. BEO was a good discriminator of CP1 outcomes (AUC = 0.74, 95% CI: 0.63-0.84, 63.4% sensitivity, 83.0% specificity) ( Fig. 2A). BEO was a good discriminator also for CPC (AUC = 0.73, 95% CI: 0.61-0.85, 63.3% sensitivity, 81.0% specificity), notwithstanding the caveat that BMI at oocyte retrieval may be unrelated to BMI at the time of second and third embryo transfer cycle (Fig. 2B). As expected, a version of BEO calculated with FF OA and EA was a poor discriminator of either CP1 or CPC outcome (AUC = 0.63 and 0.61, respectively).

Associations with PBF
Only FF OA (a MUFA) was directly associated with PBF and so were MUFA (Table 6). Among serum FA, only C18:2 was associated (directly) with PBF. Non-significant associations are listed in Supplementary Table 5.
Association of infant BW with FF and serum FA No FF or serum FA was significantly associated with infant BW, except a marginally significant direct association of FF EA (p < 0.1) (Supplementary Table 6).

DISCUSSION
To our knowledge, we present the first comprehensive survey of associations among FF or serum FA, and a broad range of fertility parameters from number of retrieved oocytes to infant birth weight. We show that the associations of FA with fertility parameters are clearly distinct in FF and serum. Notably, in FF, associations were in all cases inverse and only with number of oocytes. Our data suggest that a FF FA pool composition in favor of a diverse set of FA irrespective of the individual FA identity, negatively impacts retrieved oocyte number/quality. Further work is needed to test whether this phenomenon is causal and reflects overloading of oocyte mitochondria or any other cellular  mechanism. Also, we replicate the previously observed negative impact of FF AA on oocyte production [11][12][13]. AA is a potent bioactive FA that might interfere with oocyte homeostasis by different mechanisms such as promoting the synthesis of proinflammatory factors or altering DNA methylation profiles in crucial stages of development [22,43]. In contrast with previous reports we failed to observe any association of FF SFA, nor the reported associations of AA and OA with postfertilization parameters [11,44]. In turn, serum FA associations were distributed across prefertilization and post-fertilization parameters. The SFA C16:0 and SFA as a whole were inversely associated with percentage mature oocytes, although we failed to detect any association reported for FF counterparts [44]. The data are consistent with the adverse impact of SFA and specifically C16:0 on inflammation and broad health outcomes [45]. It is noteworthy that the great majority of cellular FA in oocytes that fail fertilization are SFA [46]. Notably, serum OA and EA at the time of oocyte retrieval are associated not only with CP1 but also with CPC. As the second and third embryo transfer cycles were carried out up to four weeks after oocyte retrieval, the results imply that memory of exposure to specific FA at that stage could persist in the early embryo, although other factors such as that maternal FA favor a receptive uterine milieu, or additional unidentified players, acting individually or in combination, cannot be ruled out. Also, the respective directions of association of OA and EA with CP are in line with the documented differential biological effects of cis and trans FA. Indeed, serum EA is inversely associated with various post-fertilization parameters in a US cohort [47]. Our data point to a degree of specialization of FF and serum FA, in which the former are linked-whether causally or not-to retrieved oocyte number, while the latter are related to a range of prefertilization and post-fertilization parameters ranging from retrieved oocyte number/maturation to CP outcome. Overall, the data suggest that the cellular memory of FF FA is lost as the oocyte exits the follicular environment. As caveat, it is possible that ethnicity is a significant factor in our study, as serum OA was associated with oocyte maturation in an Iranian cohort but not in our study [12]. In general, the effects of FA were relatively small in our multiple regression models, casting doubts on their suitability as markers.
The associations with pre-pregnancy (i.e., at oocyte retrieval) BMI were complex. The inverse association of BMI with CP, together with the direct and inverse associations of BMI with FF OA and AA, respectively, and the inverse association of FF AA with number of retrieved oocytes, show that the BMI is uncoupled from oocyte quality control and acts mainly postfertilization. The inverse association with CP was expected as it mirrors the well-documented adverse impact of BMI on fertility. Yet, the predictive value of BMI for CP outcome was modest, in line with published data [48]. Interestingly, the BEO score was a good predictor of CP1, suggesting additive effects of BMI, EA, and OA. BEO predicted well both CP1 and CPC outcome. The corresponding index calculated with FF OA and EA was a poor predictor, thus supporting BEO as biologically plausible. Whether embryos retain a memory of the milieu at oocyte retrieval that overrides the exposure to recipient uterus' conditions, warrants further investigations. At any rate, our data reinforce the body of evidence pointing to the adverse effects of obesity and selected industrial food on human fertility.
As for PBF, the direct association of FF OA with both BMI and PBF, points to that FA as a potential reliable sensor of obesity. In turn, the direct association of the serum PUFA C18:2 with PBF to our knowledge has not been reported before and might reflect an ethnicity-specific phenomenon [37]. PBF was not associated with any fertility parameter, in accordance with time-to-pregnancy data obtained in a non-IVF cohort [49].
In the case of peripheral blood global DNAm, we show for the first time a direct association with percentage fertilized oocytes. The latter trait was not associated with any of the other physiological or biochemical parameters analyzed here, therefore its association with DNAm could be causal with no confounders. DNAm was only a modest predictor of percentage fertilized oocytes. Nonetheless, the observed high specificity warrants larger studies across different populations. In apparent accordance with our data, low DNA methyltransferase activity due to either mutations or biochemical inhibition induces female infertility in lower eukaryotes, although the observed phenotype is likely the consequence of widespread global transcription disruption [50][51][52]. At any rate, a detailed survey of peripheral blood DNA methylome at individual CpG dinucleotide level is needed to identify potential DNAm markers of impaired female fertility.  The main limitations of this study are lack of patient lifestyle and dietary habit information, and a design that does not allow to assess causality.
In conclusion, we show that FF and serum FA are associated with clearly distinct stages of embryonic and fetal development. These associations are uncoupled from BMI. Also, we identify a score based on BMI and selected FA, which is a candidate predictor of pregnancy outcome in IVF.