Fourier transform infrared spectroscopy of follicular fluids from large and small antral follicles

Non Thomas1, Royston Goodacre2, Éadaoin M. Timmins2, Marco Gaudoin1 and Richard Fleming1,3

1 3Assisted Conception Service, Glasgow Royal Infirmary University NHS Trust, Glasgow, G31 2ER and 2 Institute of Biological Sciences, Cledwyn Building, University of Wales, Aberystwyth, Ceredigion, Wales, SY23 3DD, UK


    Abstract
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Fourier transform infrared spectroscopy (FTIR) was used to obtain `biochemical fingerprints' for the constitution of follicular fluids from large and small antral luteinized follicles (n = 54 pairs). All samples gave reproducible characteristic biological infrared absorption spectra, with recognizable amide I protein vibrations and acyl vibrations from fatty acids. Discriminant function analysis of the first derivative FTIR spectra, together with hierarchical cluster analysis used to construct a dendrogram, showed fluid from large follicles formed a homogeneous closely related cluster, whilst that from small follicles was distinct from the large, and heterogeneous in nature. The large follicle fluids showed closer biochemical similarity to each other than to the corresponding fluid taken from small matched follicles. An artificial neural network was trained and following validation with an independent test set, successfully distinguished follicular fluids from large and small follicles. The sex steroid concentrations in the fluids from large and small follicles were significantly different. These results show that fluid from large follicles is distinct in biochemical nature from that from small follicles, but the degree of homogeneity implies size-specific changes take place. These may have consequences for the developmental potential of the oocyte.

Key words: artificial neural networks/follicle fluid/follicle size/fourier transform infrared spectroscopy/steroid analysis


    Introduction
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
In the normal menstrual cycle, follicular diameter (FD), oestradiol and inhibin are the main indicators of the maturational stage of the follicle and enclosed oocyte. FD has become the marker of choice in the treatment of assisted conception patients, to whom human chorionic gonadotrophin (HCG) is administered when a number of follicles achieve the desired size of between 17–20mm in diameter. During retrieval, oocytes are removed from follicles of a wide range of sizes but it is often presumed that follicles >16mm in diameter yield oocytes with a higher developmental capacity than those from smaller follicles. Previous studies have shown that larger follicles yield oocytes with increased developmental, fertilization and embryo developmental rates than those yielded by smaller follicles in IVF programmes (Ectors et al., 1997Go; Bergh et al., 1998Go). Conflicting results were achieved in the same studies using intracytoplasmic sperm injection (ICSI), suggesting that any developmental deficiency in small oocytes may (Bergh et al., 1998Go) or may not (Ectors et al., 1997Go) be overcome using sperm injection. However, despite the recorded differences, there are still notable rates of fertilization and pregnancy achieved with oocytes taken from smaller follicles. Indeed, although previous studies have shown fertilization rates to be lower in oocytes taken from small follicles (<16mm diameter or <1ml volume) compared with large (>16mm diameter or >1ml volume) no difference in embryo quality (Salha et al., 1998Go) or pregnancy rate could be demonstrated between the two groups (Nagai et al., 1997Go). These clinical studies indicate that there may be biochemical differences related to these later stages of follicular development.

As an ovarian follicle grows, follicular fluid, a mixture of follicular secretions and plasma exudate, accumulates between the granulosa cells. The importance of this fluid cannot be underestimated since it is closely associated with the oocyte and its composition may influence the latter part of oocyte development. Many follicular fluid factors will have been secreted by cells in close association with the oocyte and the fluid contains factors responsible for ensuring the correct maturational status of the oocyte. Constituents of the fluid include steroids, lipids, glycosaminoglycans and numerous proteins and peptides.

It has long been a goal to identify factors in follicles which may indicate improved chances of embryo development and conception. The aim of this study was to establish if a definitive difference exists between follicular fluid derived from different sized follicles. Paired samples were subjected to the analytical technique of Fourier transform infrared (FTIR) spectroscopy.

FTIR is a physico-chemical method which measures the vibrations of bonds within functional groups (Griffiths and de Haseth, 1986Go; Stuart, 1997Go). In FTIR analysis a particular bond absorbs light electromagnetic (EM) radiation at a specific wavelength; for example, the infrared (IR) spectra of proteins exhibit strong amide I absorption bands at 1653 cm–1 associated with the characteristic stretching of C=O and C-N and the bending of the N-H bonds (Stuart, 1997Go). Therefore by interrogating a biological sample with EM radiation of many wavelengths in the mid-IR range (usually defined as 4000–600 cm–1) one can construct an IR `fingerprint' of the original biological sample under investigation. Since different bonds absorb different wavelengths of EM radiation, these `fingerprints' are made up of the vibrational features of all chemical components in the sample analysed. Thus this method gives quantitative information about the total biochemical composition of a follicular fluid, without its destruction, and produces `fingerprints' which are reproducible and distinct for different biological materials. Within medicine there are precedents for using IR measurement techniques for the analysis of human biofluids such as urine, blood and synovial fluid (Wang et al., 1996Go; Jackson et al., 1997Go; Diem et al., 1999Go).


    Materials and methods
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Patients and treatments
Follicular fluid was collected at oocyte retrieval from patients undergoing routine IVF and ICSI at the Glasgow Royal Infirmary. Paired follicular fluid samples were obtained from 54 patients (mean age 33, range 26–42 years). Diagnoses were as follows:polycystic ovarian syndrome (PCOS) (n = 1); endometriosis (n = 6); endometriosis plus PCOS (n = 1); male factor (n = 9); tubal (n = 25); tubal plus endometriosis (n = 3); tubal plus male (n = 1) and unexplained (n = 8). All patients were initially given a long course of GnRH analogue (buserelin) as a method of down-regulation. On day 3 of the cycle, gonadotrophin stimulation was commenced and the follicular phase was monitored by transvaginal ultrasonography and serum oestradiol measurements. When the presence of three or more follicles, >17 mm diameter was detected, the patients received HCG administration and the follicular fluids and oocytes were collected ~36 h later.

Local ethical committee approval was granted for this project by the Glasgow Royal Infirmary University NHS Trust.

Follicular fluid collection
Two follicular fluid samples were collected from each patient. One from a follicle with a diameter >17mm (large) and one from a follicle <15mm (small). Follicle size was determined by using the mean of two perpendicular measurements. In order to minimize contamination of fluids, samples were collected following a needle flush so that there would be no dilution in the catheter with fluids from follicles not included in the study. In practice the small follicle was the first to be aspirated at retrieval. The large follicle used in the study was the first approach to the contralateral ovary. Only mid-stream samples, free from blood contamination, were collected and aliquots were stored at –20°C. All the samples used in this study were paired.

Fourier transformed IR spectroscopy
Aliquots (20µl) of the 108 follicular fluids were evenly applied onto a sand-blasted aluminium plate. Prior to analysis the samples were oven-dried at 50°C for 30 min. Samples were run in triplicate. The FTIR instrument used was the Bruker IFS28 FTIR spectrometer (Bruker Spectrospin Ltd., Banner Lane, Coventry, UK) equipped with an MCT (mercury-cadmium-telluride) detector cooled with liquid N2. The aluminium plate was then loaded onto the motorized stage of an adapted reflectance thin-layer chromatography accessory (Timmins et al., 1998Go). The personal computer used to control the IFS28 was also programmed (using OPUS version 2.1 software running under IBM O/S2 Warp provided by the manufacturers) to collect spectra over the wavenumber range 4000 cm–1 to 600 cm–1. Spectra were acquired at a rate of 20 s–1. The spectral resolution used was 4 cm–1. To improve the signal-to-noise ratio, 256 spectra were co-added and averaged. Each sample was thus represented by a spectrum containing 882 points and spectra were displayed in terms of absorbance as calculated from the reflectance-absorbance spectra using the Opus software [which is based on the Kubelka-Munk theory (Griffiths and deHaseth, 1986)].

To minimize problems arising from baseline shifts the following procedure was implemented: (i) the spectra were first normalized so that the smallest absorbance was set to 0 and the highest to +1 for each spectrum, (ii) next the first derivatives of the original FTIR spectra were smoothed using the Savitzky-Golay algorithm (Savitzky and Golay, 1964Go) using 5-point smoothing. Typical spectra, pre- and post-processing, are shown in Figure 1Go.




View larger version (49K):
[in this window]
[in a new window]
 
Figure 1. FTIR spectra of follicular fluids from large and small antral follicles from patient 1. Spectra shown are either raw (A) or the smoothed first derivatives (B). In the derivative spectrum (B) the original spectrum (A) is differentiated. The resolution is enhanced in the first derivative because it is a measure of the change in gradient, therefore the y axis is a measure of the difference in relative absorbance divided by the difference in wavenumber.FA = fatty acids; PH = phosphates; PP = peptides; PS = polysaccharides; U = unassigned; M = mixed region of proteins, fatty acids, phosphate rich.

 
Cluster analysis
The initial stage involved the reduction of the multidimensional FTIR data by principal components analysis (PCA; Jolliffe, 1986). PCA is a well known technique for reducing the dimensionality of multivariate data whilst preserving most of the variance; PCA was performed according to the NIPALS algorithm (Wold, 1966Go). Discriminant function analysis [DFA; also known as canonical variates analysis (CVA)] then discriminated between groups on the basis of the retained principal components and the knowledge a priori of which spectra were replicates, and thus this process did not bias the analysis in any way (Manly, 1994Go). Finally, the Euclidean distance between a priori group centres in DFA space was used to construct a similarity measure, with the Gower general similarity coefficient SG (Gower, 1966Go), and these distance measures were then processed by an agglomerative clustering algorithm to construct a dendrogram (Manly, 1994Go). These cluster analysis methods were implemented using Matlab version 5 (The Math Works, Inc., Natick, MA, USA), which runs under Microsoft Windows NT on an IBM-compatible PC.

Artificial neural network analysis
When the desired responses (targets) associated with each of the inputs (spectra) are known then the system may be `supervized'. The goal of supervized learning is to find a model that will correctly associate the inputs with the targets; this is usually achieved by minimizing the error between the target and the model's response (output) (Massart et al., 1997Go).

The artificial neural network (ANN) method of standard back-propagation multi-layer perceptrons (MLP) (Rumelhart et al., 1986Go; Bishop, 1995Go) was used, and was carried out with a neural network simulation program, NeuFrame version 3,0,0,0 (Neural Computer Sciences, Southampton, Hants, UK), which runs under Microsoft Windows NT on an IBM-compatible PC. In-depth descriptions of the modus operandi of this type of ANN analysis are given elsewhere (Goodacre et al., 1998Go). The architecture of the ANNs was 878 input nodes, 10 nodes in the hidden layer, and a single output node (this topology can be represented as 878–10–1). The ANNs were trained with the first 27 samples from the fluids from both small and large follicles [therefore there were 162 spectra (27*3*2) in the training set]. These ANN were trained for 1.5x103 epochs (calculations), when the RMS (root mean squared) error between the observed and desired outputs was typically 0.046 ± 0.004; on a Pentium 133, with 128 MB RAM, this typically took ~90 min. The ANN was then interrogated with both the training and test sets and a correct identity for a fluid from a small antral follicle was taken as <0.5 and for a fluid from a large follicle as >=0.5; note that after training, the interrogation of these ANN takes only a few milliseconds.

Steroid hormone assays
Oestradiol and progesterone were analysed using a solid-phase chemiluminescent enzyme immunoassay (DPC, Los Angeles, CA, USA) on an Immulite automated immunoassay system. Matching paired samples were assayed in two batches with intra-assay coefficients of variations of 4.5% (oestradiol) and 6% (progesterone).


    Results
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Fourier transformed IR spectroscopy
All samples gave reproducible characteristic biological IR absorption spectra with recognizable amide I protein vibrations and acyl vibrations from fatty acids (see Figure 1AGo). However, these FTIR spectra (and all the others collected) showed broad and complex contours, highlighting the fact that there was very little qualitative difference between the spectra, although at least some complex quantitative differences between them were observed. Such spectra, essentially uninterpretable by the naked eye, readily illustrated the need to employ multivariate statistical techniques for their analyses.

DFA were used to observe the relationships between the follicular fluids as judged from their derived FTIR spectra, and DFA was performed as detailed above. The resulting DFA plot is shown in Figure 2Go where it can be seen that the follicular fluids from large follicles formed a homogeneous, closely related cluster whilst those fluids from small follicles were very heterogeneous and distinct from the fluids from large antral follicles.



View larger version (16K):
[in this window]
[in a new window]
 
Figure 2. Discriminant function analysis on 54 pairs of follicular fluids. Data points plotted are the discriminant function score averages and the identifiers are either S (small antral follicles) or L (large).

 
An alternative method of viewing the relationship between these follicular fluids is by using hierarchical cluster analyses and the resulting dendrogram is shown in Figure 3Go. This plot also showed clearly that the fluids from large follicles were more closely related to each other than to their corresponding paired follicular fluids from small antral follicles.



View larger version (17K):
[in this window]
[in a new window]
 
Figure 3. Hierarchical cluster analysis on 54 pairs of follicular fluids.

 
To distinguish between spectra from fluids from large and small antral follicles, the supervized learning method of ANNs was implemented. ANN were trained on the first 27 fluids from both the small and large antral follicles, with the output node encoded as either `0' for fluids from small follicles or `1' for fluids from large follicles. Once calibrated, the ANN was then interrogated with derivative spectra from both the training and test sets, and its predictions are shown in Table IGo. It can be seen that both the training set and independent test set were correctly identified for whether the follicular fluids had been taken from large or small antral follicles. It is therefore evident that there was reproducible (bio)chemical information in the FTIR spectra of these follicular fluids that could distinguish unequivocally between large and small antral follicles.


View this table:
[in this window]
[in a new window]
 
Table I. Identity of the size of the small and large antral follicles from the FTIR spectra of follicular fluids in the training and test sets as judged by artificial neural networks (ANN).
 
Steroid analysis
The sex steroid concentrations of the follicular fluids also showed that fluid from large follicles was significantly different from small (Table IIGo). Furthermore, the standard deviations of the steroid hormone concentration values were significantly greater in the small follicles (progesterone, 67%; oestradiol, 73%) than in the large follicles (42% and 52% respectively), again demonstrating greater heterogeneity in the smaller follicles.


View this table:
[in this window]
[in a new window]
 
Table II. Steroid concentrations of follicular fluids
 

    Discussion
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
DFA showed the FTIR spectra of fluids collected from large antral follicles to be tightly clustered, indicating a similar biochemical profile, whereas fluids from small follicles were heterogeneous. This heterogeneity in the FTIR `biochemical fingerprint' of fluids from small follicles may reflect differences in the maturational stage of the antral follicles. Dendrogram analysis of the same fluids also showed similarities in the biochemistry of the fluids from large follicles which were more closely related to each other than to fluids from their corresponding small antral follicles.

FTIR spectroscopy coupled with ANN analysis distinguished successfully follicular fluid from large and small antral ovarian follicles indicating that biochemical differences do indeed exist between the fluids in different size follicles.

The differences in distribution of steroid concentrations do not reflect those of the FTIR, suggesting that other aspects of biochemical analysis are not under the same control mechanism as the steroids. These differences between factors in the fluids from large and follicles, which may be quantitative or qualitative, may be related to differences in oocyte quality. These results highlight a difference in the biochemical nature of fluids from large and small follicles which we suggest reflect the developmental stage of the follicle. Further work will investigate this premise and relate the FTIR spectra of a follicular fluid to the developmental capacity of the oocyte that the follicle contained.

In conclusion, this is the first time that FTIR spectroscopy has been used to analyse follicular fluids and it has shown that gross biochemical differences do exist between the follicular fluids from large and small antral ovarian follicles.


    Acknowledgments
 
R.G. and E.M.T. are indebted to the Wellcome Trust for financial support (grant number 042615/Z/94/Z).


    Notes
 
3 To whom correspondence should be addressed at: Assisted Conception Service, Glasgow Royal Infirmary University NHS Trust, Glasgow G31 2ER, UK. E-mail: gqta13{at}udcf.gla.ac.uk Back


    References
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 
Bergh, C., Broden, H., Lundin, K. et al. (1998) Comparison of fertilization, cleavage and pregnancy rates of oocytes from large and small follicles. Hum. Reprod., 13, 1912–1915.[Abstract]

Bishop, C.M. (1995) Neural networks for pattern recognition. Clarendon Press, Oxford.

Diem, M., Boydston White, S. and Chiriboga, L. (1999) Infrared spectroscopy of cells and tissues: Shining light onto a novel subject. Appl. Spectrosc. 53, A148-A161.

Ectors, E., Vanderzwalmen, P., Van Hoeck, J. et al. (1997) Relationship of human follicular diameter with oocyte fertilization and development after in vitro fertilization or intracytoplasmic sperm injection. Hum. Reprod., 12, 2002–2005.[Abstract]

Goodacre, R., Timmins, É.M., Burton, R. et al. (1998) Rapid identification of urinary tract infection bacteria using hyperspectral, whole organism fingerprinting and artificial neural networks. Microbiology, 144, 1157–1170.[Abstract]

Gower, J.C. (1966) Some Distance Properties of Latent Root and Vector Methods used in Multivariate Analysis. Biometrika, 53, 325–338.[ISI]

Griffiths, P.R. and de Haseth, J.A. (1986) Fourier transformed infrared spectrometry. John Wiley, New York.

Jackson, M., Sowa, M.G. and Mantsch, H.H. (1997) Infrared spectroscopy: a new frontier in medicine. Biophysical Chemistry, 68, 109–125.[ISI][Medline]

Jolliffe, I.T. (1986) Principal Component Analysis. Springer-Verlag: New York.

Manly, B.F.J. (ed.) (1994) Multivariate Statistical Methods: A Primer. Chapman & Hall, London.

Massart, D.L., Vandeginste, B.G.M., Buydens, L.M.C. et al. (1997) Handbook of Chemometrics and Qualimetrics: Part A. Elsevier, Amsterdam.

Nagai, S., Yasumizu, T., Kasai, T. et al. (1997) Effect of oocyte retrieval from a small leading follicle in fixed-schedule in vitro fertilization program. J. Obs.Gynaecol. Res., 23, 165–169.

Rumelhart, D.E., McClelland, J.L. and The PDP Research Group (1986) Parallel Distributed Processing, Experiments in the Microstructure of Cognition, Vol. I and II, MIT Press, Cambridge, MA.

Salha, O., Nugent, D., Dada, T. et al. (1998) The relationship between follicular fluid aspirate volume and oocyte maturity in in vitro fertilization cycles. Hum. Reprod., 13, 1901–1906.[Abstract]

Savitzky, A. and Golay, M.J.E. (1964) Smoothing and differentiation of data by simplified least squares procedures. Anal. Chem., 36, 1627–1623.[ISI]

Stuart, B. (1997) Biological Applications of Infrared Spectrocopy. John Wiley & Sons, Chichester.

Timmins, É. M., Howell, S. A., Alsberg, B. K. et al. (1998) Rapid differentiation of closely related candida species and strains by pyrolysis-mass spectrometry and Fourier transform-infrared spectroscopy. J. Clin. Microbiol., 36,367–374.[Abstract/Free Full Text]

Wang, J., Sowa, M., Mantsch, H.H. et al. (1996) Comparison of different IR measurement techniques in the clinical analysis of biofluids. TRAC – Trends in Analytical Chemistry 15, 286–296.

Wold, H. (1966) Estimation of principal components and related models by iterative least squares. In Krishnaiah, K.R. (Ed) Multivariate Analysis, Academic Press, New York pp. 391–420.

Submitted on November 5, 1999; accepted on April 25, 2000.