1. Introduction
Somatic mutations in the human cytosolic isocitrate dehydrogenase 1 (
IDH1) gene are a frequent feature observed in gliomas. The
IDH1 mutation tends to occur in the early stages of gliomagenesis, hence it is most commonly found in low-grade gliomas, diffuse astrocytoma and oligodendrogliomas [
1], but is less common (10%) in primary glioblastoma (GBM) [
2,
3], except where the GBM develops from a previously diagnosed diffuse or anaplastic astrocytoma (>80%) [
4,
5]. Consequently, the
IDH1 mutation serves as a valuable diagnostic marker (
Table 1) by assisting in the differentiation of tumour entities that are often indistinguishable through histopathological analysis alone, but have different treatments and prognostic profiles [
5].
The normal function of the
IDH1 enzyme is to convert isocitrate to α-ketoglutarate (αKG). Cancer-associated mutations in
IDH1 inactivate this standard enzymatic activity, but enable a neomorphic conversion of αKG to the oncometabolite 2-hydroxyglutarate (2HG) [
7,
8]. This results in an accumulation of 2HG in the glioma cells, which is thought to drive oncogenic activity and tumorigenesis [
9]. The vast majority (~90%) of
IDH1 mutations involve transitions in codon 132, where the arginine residue is replaced by histidine (R132H-
IDH1) [
2]. Patients who have this
IDH1 mutation in their glioma have a significantly better prognosis compared to those with
IDH1-wildtype lesions of the same histologic grade [
10,
11]. For example, those with an
IDH-mutant GBM tend to have a better predicted prognosis than patients with a lower-grade
IDH-wildtype astrocytoma [
4,
12].
The presence of R132H-
IDH1 can be established through immunohistochemistry (IHC) by applying the
mIDH1R132H antibody to resected glioma tissue [
13]. IHC depends on invasive biopsy, with inherent risk for the patient. The R132H-
IDH1 expression may be present only in a fraction of tumour cells in some diffuse gliomas, therefore a negative result does not necessarily exclude a glial lesion as the concentration of immuno-positive diffuse astrocytomas ranges between 50% and 70% [
14]. Thus, several sections of tumour often need to be biopsied for a reliable result. Likewise, false positives are occasionally observed because of non-specific background staining, and the regional heterogeneity of R132H-
IDH1 expression can cause doubt in the diagnosis, which may necessitate confirmatory genetic analysis [
15].
The development of a simple, rapid and label-free diagnostic tool for
IDH1 detection would be transformative for molecular diagnosis. Analytical techniques involving vibrational spectroscopy have great potential for diagnosing disease states, namely infrared and Raman spectroscopy [
16,
17,
18]. In particular, Fourier transform infrared (FTIR) spectroscopy has been shown to be valuable for the detection of various cancers, such as breast, lung, colorectal, ovarian and prostate cancer [
19,
20,
21,
22,
23,
24,
25,
26], since it can probe the biochemical composition of normal and pathological tissue and generate the fingerprint structure of several biomolecular components, such as proteins, lipids and nucleic material [
27,
28]. Several studies have looked into diagnosing brain lesions, utilizing Raman spectroscopy [
29,
30,
31,
32]. Likewise, FTIR can detect and stratify brain malignancies through the analysis of resected tissue sections, mainly focused on transmission techniques [
33,
34,
35,
36]. On the other hand, attenuated total reflection (ATR)-FTIR is well-suited to biological fluids, such as blood serum [
37,
38]. The technique provides for the qualitative interrogation of all infrared active macromolecular constituents of blood serum, and it is well established that biomolecular imbalances in biofluids can give an indication of disease states [
28]. The plethora of spectroscopic studies highlight the capability of FTIR to become a powerful tool in the diagnostic field [
39]. Uckermann et al. recently indicated FTIR could be suitable in identifying mutated
IDH1 expression, through the analysis of 34 frozen brain tissue cryosections and 64 fresh unfixed glioma biopsies [
40]. Despite some promising results, the spectra of both the cryosections and fresh tissue biopsies demonstrated high inter-patient variability. The variance in the fresh tissue analysis may have been accentuated by the use of ATR-FTIR, which only interrogates the region of the tissue sample that is in contact with the internal reflection element (IRE), so it can be difficult to ensure that the sampling area being examined is representative of the tumour. Uckermann et al. proposed that further work would be required to fully evaluate the ability of the technique in the application of detecting the
IDH1 mutation and other potential biomarkers. Synchrotron radiation-based FTIR (SR-FTIR) microspectroscopy is a method that can be used to extract finer spatial and spectral details from biological tissue samples [
28]. In SR-FTIR, a synchrotron source emits a collimated light beam more intense than standard bench-top spectrometers [
41,
42]. Synchrotron radiation can be up to
times brighter than any other conventional broadband IR source, allowing smaller regions of tissue to be probed with superior signal-to-noise [
43]. Thus, SR-FTIR spectroscopy offers a high-resolution approach that can be valuable for proof-of-principle studies in acquiring greater biological information.
The implementation of FTIR spectroscopy during surgical biopsy could present a fast, label-free method for the molecular genetic classification of gliomas.
IDH1 mutation is associated with a better prognosis, as is maximal surgical resection [
44]. Attempted maximum safe surgical resection may be more justified in patients with
IDH1-mutant gliomas, whilst a more limited resection may be more appropriate for
IDH1-wildtype gliomas. Intra-operative rapid determination of tumour
IDH1 status could therefore inform neurosurgical decision-making [
45]. In this study, SR-FTIR has been used to examine human brain glioma tissue, where single-point spectra have been collected from tissue microarray (TMA) sections comprising
IDH1-mutated and
IDH1-wildtype glioma tissue cores. Additionally, we examine the potential for earlier molecular subclassification of tumours by identifying the biomolecular alterations caused by the genetic
IDH1 mutation in glioma patient serum. The combination of centrifugal filtration and ATR-FTIR serum spectroscopy could be implemented prior to biopsy or resection to determine
IDH1 status even before surgery.
4. Discussion
In the synchrotron data analysis, a spectral cut of 1800–1200 cm
−1 was included in the initial grid search as there appeared to be a drop in signal below <1200 cm
−1. This cut-off effect for
in the low wavenumber region is common in scanning microscopy and can be caused by the change in refractive index (RI), e.g., the RI of
decreases from ~1.4 at 5 μm to ~1.3 at 10 μm. Additionally, when using a synchrotron source in the scanning microscopy mode for high spatial resolution, the diffraction limit is achieved when the microscope’s aperture defines a spot size scaled with the longest wavelength of the spectral region of interest [
63]. Here, we used 10 μm slits, and therefore the diffraction limit could be affecting the signal towards 1000 cm
−1 (full width half maximum (fwhm) ~l/2*NA = l = 10 μm, i.e., same size as the slit size used for scanning microspectroscopy) [
64]. This did not appear to be a significant problem as the optimal pre-processing method involved the removal of this spectral region.
In
Figure 7, arguably the largest difference between the
IDH1-mutated and
IDH1-wildtype groups arises within the Amide I band, associated with the stretching of double-bonded carbonyl groups (C = O) and C-N bonds, as well as N-H bending vibrations in proteinaceous biomolecules [
27]. The lower-wavenumber side of the Amide I band (1620–1600 cm
−1) was more intense in the
IDH1-mutated spectra (positive regions in the difference spectrum), whereas the band intensities between 1700 and 1650 cm
−1 were lower compared to the
IDH1-wildtype tissue spectra. This is consistent with a study wherein
IDH1-mutated cell lines exhibited an elevated absorbance at 1610 cm
−1, but a lower intensity around 1690 cm
−1 [
40]. These findings are not directly comparable to the results presented here, as cell lines may not adequately represent primary cells in clinical specimens [
65]. The observed differences are likely to result from alterations in overlapping bands existing within the broad Amide I envelope, accounting for various protein secondary structures that can only be revealed with deconvolution techniques [
66]. It is thought that the large negative peak in the difference spectrum at ~1660 cm
−1 may represent a deviation in the levels of α-helical structures, and the smaller positive peak ~1615 cm
−1 may be tentatively assigned to β-sheet components [
67]. The band intensities at approximately ~1750 cm
−1 and ~1560 cm
−1 were lower in the mean
IDH1-mutated spectrum, while those at ~1495 cm
−1 and between 1450 and 1200 cm
−1 displayed a higher absorbance than the
IDH1-wildtype spectra (
Figure 7).
Several spectral differences are similar to previous findings, namely, the variances in Amide III of proteins (mainly N-H in plane bending and C-N stretch, ~1300 cm
−1), nucleic material such as DNA and RNA (
asymmetric stretch, ~1230 cm
−1), and lipidic contributions (C = O stretch, ~1750 cm
−1; CH
3 bending, ~1450 cm
−1) [
40]. The disparities in the IR spectra could potentially be attributed to the increase in 2HG in the
IDH1-mutated glioma tissue, which is known to be elevated in tumour cells with the
IDH1 mutation [
8]. With reference to an IR spectrum of pure 2HG [
40], the bands around 1589, 1450, 1416, 1344, 1311, 1267, 1236 and 1203 cm
−1 could explain some of the differences observed between
IDH1-mutated and
IDH1-wildtype tissue in this study, as the band intensities at these wavenumbers are all elevated in
IDH1-mutated patients, as described by the difference spectrum (
Figure 7). That being said, it may only indicate a global change in biomolecular content, reflected by the systemic response of the genetic mutation within the glial tumour cells.
The disparities in the classification results between the initial grid search and the LDA model resampled 51 times highlight the importance of utilising a reasonable number of iterations, in order to minimise the variance whilst maintaining a respectable analysis time. However, the sensitivity and specificity remained well-balanced and above 80%, which are highly promising results. As shown in
Table 3, the standard deviation is much higher for the sensitivity than the specificity, which is not entirely surprising because of the lower number of
IDH1-mutated samples within the dataset. A 70:30 split between training and testing data meant that there were only seven randomly selected
IDH1-mutated samples in each of the 51 resampled test sets. Therefore, when a known mutated tissue core is misdiagnosed as
IDH1-wildtype, it has a substantial effect on the sensitivity. As described in the confusion matrices in
Figure S3, there is a drop of ~15% in sensitivity when an
IDH1-mutated sample is predicted wrongly. Conversely, there is only a ~4% difference in specificity with a misdiagnosed
IDH1-wildtype, as there were 26
IDH1-wildtype samples in every test set. Thus, the addition of more glioma samples with the
IDH1 mutation would be beneficial for this analysis, in order to minimise the associated error. Nevertheless, these values demonstrate significant potential, and a mean balanced accuracy of 82.9% indicates synchrotron-based transmission FTIR is capable of identifying mutated and wildtype
IDH1 tumours.
Regarding the ATR-FTIR results, the whole serum classifiers seemed to be more effective at predicting the
IDH1-mutated serum samples from the test sets, as the sensitivities were much higher than the specificities in each case. It is not clear why this may be, as there were an equal number of samples in each class, and therefore there should be no bias present in the models. That being said, the results did not appear to be reliable, and given the poor balanced accuracies (~50%), it could be assumed that the correct predictions were ultimately made by chance. Likewise, the low AUC values from the ROC curves (
Figure 10) suggest they had no diagnostic ability. The LMW fraction of the serum is believed to contain disease-specific information, making the spectroscopic signature of this fraction useful for diagnostics [
68]. Thus, after the poor classification performance for the whole serum data, it was thought that discrete molecular differences could potentially be emphasised through the use of centrifugal filtration. The balanced accuracies were enhanced to between 60 and 70% for all tested filtrate models. The centrifugal filtration step produced a significant improvement in the model’s performance, by delivering more balanced sensitivities and specificities. Similar to the tissue-based results, these findings are based on a relatively small cohort with only 36 patients in each class, thus misdiagnosed patients have a profound effect on the sensitivity or specificity values. Additional analysis with a larger patient cohort would be beneficial in identifying the true potential of the technique for this particular clinical application.
5. Limitations
Despite the promising results reported in this preliminary study, it is important to highlight some of the limitations. Since we have utilised the UK’s synchrotron facility here, the current methodology is not directly translatable to the hospital setting. Synchrotron instruments are admirable for high spatial resolution; thus, it was chosen in this project to attain the greatest level of biological and diagnostic detail from the glioma samples. Synchrotron measurements can be subject to lengthy analysis times, but standard bench-top FTIR spectrometers can acquire similar data quality with more efficient analysis. The ability to discriminate the IDH1 mutation in glioma TMA sections with 80% accuracy would likely be clinically acceptable, although future studies should also consider probing fresh tissue biopsies rather than FFPE tissue microarrays, which would be better suited to the determination of a patient’s IDH1 status mid-surgery.
The whole serum results reported an accuracy of ~50%, which would not be deemed acceptable in the clinic. Blood serum comprises thousands of different proteins, ranging from the more abundant HMW serum albumin (50 g/L) to the LMW proteins like troponin (1 ng/L) [
69]. Due to the wealth of various biomolecules that exist in a normal serum sample, it was expected to be a significant challenge to identify the subtle alterations in blood composition that may have been associated with the
IDH1 mutation. The filtration step did improve the classification performance, increasing the accuracy up to almost 70%. Although, it has been suggested elsewhere that the large absorbance band observed in the filtered serum spectrum (
Figure 9, ~1030 cm
−1) is due to glycerine interference, introduced into the sample from the centrifugal filters [
70]. This could potentially be obscuring crucial information that may help improve the test performance. Future research could implement a washing step prior to centrifugation. There are also many filter sizes to choose from, hence filtration with a different cut-off point may also further improve classification performance. Many cytokines and chemokines exist at molecular weights greater than 3kDa, which may be indicative of disease.
As already stated, both the tissue and serum analysis could benefit through the addition of more patients, preferably from prospective trials, which would likely reduce the standard error within the classification models. It is vital that more efficient methods are developed for this application before clinical translation can be realised.