Sub-terahertz vibrational spectroscopy of ovarian cancer and normal control tissue for molecular diagnostic technology

Abstract

ABSTRACT:

We introduce here recently developed highly resolved Sub-Terahertz resonance spectroscopy of biological molecules and cells combined with molecular dynamics (MD) computational analysis as a new approach for optical visualization and quantification of the presence of microRNAs, particularly the mir-200 family, as potential biomarkers in samples from tissue of epithelial ovarian cancers for disease early detection, analysis, prognosis and treatment.

METHOD:

A set of samples for this study was prepared from anonymized archival formalin-fixed, paraffin-embedded ovarian epithelial tissue containing regions of invasive neoplastic cells from cases of high-histologic grade serous papillary ovarian carcinoma. Control samples were normal mucosa from fallopian tubes of patients with no known malignancy. Spectroscopic characterization of tissue samples in this study was performed using a continuous wave, frequency domain automated spectrometer operating at room temperature in the spectral region of 310–500 GHz. The spectral results were compared with molecular dynamics simulations and absorption coefficient calculations utilized to predict the absorption spectra.

RESULTS:

The characteristic spectroscopic features in absorption spectra, particularly the presence of absorption peaks near 13 cm ${}^{-1}$ have been identified as cancer indicators. Tissue samples heterogeneity, reflected by diverse spectral signatures, provides additional, very specific information that may be used for identification of cancer subtypes, clinical behavior or sensitivity to specific therapies. Further work is warranted to determine if this signature can be detected in bio-fluids from ovarian cancer patients. If strongly correlated with cancer burden, it may then be investigated as a potential new biomarker for disease monitoring, and also perhaps as a biomarker for cancer screening.

1. Introduction

Epithelial ovarian cancer is the most lethal female reproductive malignancy, mainly because 70% of patients are diagnosed with advanced stage (stage III–IV) disease. The main reason is lack of success in diagnosing ovarian cancer at an early stage. At the same time, ovarian cancer is highly curable if detected early (85% survival for stage I disease) [1].

Currently, epithelial ovarian cancer is detected by a physical exam of the pelvic area, looking for enlarged ovaries and/or fluid in the abdomen. Since the symptoms of ovarian cancer are very similar to gastrointestinal illness, imaging techniques such as ultrasound, CT scans, or MRI are used to detect the presence of a mass, which may not be large enough to be detected by a physical exam. Once a mass is detected, surgery is performed to remove the mass, which is then histologically examined to determine if the tissue removed is cancerous. The standard initial management of this cancer consists of aggressive surgical cytoreduction, including total abdominal hysterectomy and bilateral salpingo-oophorectomy, and platinum/taxane combination chemotherapy [2]. The cancerous (malignant/benign) determination is made by an experienced pathologist, through microscopic examination of normal and abnormal cells in the removed tissue. Yet, this analysis method is subjective, and further confirmation of the malignancy is sometimes required.

Currently, early detection interventions include serum measurement of protein biomarkers (e.g. CA-125, HE4, OVA1 ${}^{\text{TM}}$ ) and ultrasound imaging. Cancer antigen 125 (CA-125) is a high molecular weight glycoprotein that is expressed by a large proportion of epithelial ovarian cancers [2]. Since its discovery, CA-125 has become well established as a tumor marker for epithelial ovarian cancer. However, the sensitivity and specificity of CA-125 is known to be poor [2, 3]. It is only raised in approximately 50% of stage I epithelial ovarian cancers and in 75% to 90% of patients with advanced disease [4, 5, 6]. In addition, the specificity of the test is poor, and false-positive results have been noted in many medical disorders, both malignant [7] and benign [8]. Greater specificity can be achieved by a combination of CA-125 measurement and ultrasound allowing for detailed imaging of the ovaries and the detection of morphologic changes that may signify a developing malignancy. However, most ovarian masses detected by ultrasound screening are benign and the important limitation of ultrasound is the considerable variation among observers in interpreting and scoring ultrasonographic images leading to unnecessary surgery in healthy women [2].

The latest research efforts are focused toward investigating assays that might improve sensitivity for the early detection of ovarian cancer using either cancer cells or other biomarkers found in the patient’s bodily fluids (blood, urine, ascites). One of these method, proteomics, is based on the study of protein expression patterns, protein interactions, and protein pathways in the blood, individual organ systems, and tissue cells [2, 9]. Overall, proteomic studies have yielded numerous markers that unfortunately seem to perform, at best, similarly to CA-125 [10]. It seems unlikely that a single marker for epithelial ovarian cancer will be clinically useful given the biologic heterogeneity of the disease [10]. At least 30 markers have so far been combined with CA-125 increasing sensitivity by 5% to 15%, but specificity has inevitably been reduced. Greater specificity was achieved by a combination of a sequential approach with CA-125 as a primary test and pelvic ultrasonography as a secondary test, with a high specificity and positive predictive value. In some cases DNA sequencing of a tumor sample may be performed to identify mutated gene products amenable to targeted therapy, but this is not a routine part of ovarian cancer diagnostics at this time. Despite advances in the treatment of ovarian cancer, effective screening, and early detection, particularly for those women who are predisposed to epithelial ovarian cancers, has not kept pace [2]. While genetic screening can be used to detect the presence of predisposing mutations, this provides impetus for screening, but not a good screening method for early detection.

In our recently published paper [11] we introduced highly resolved Sub-Terahertz resonance spectroscopy of biological molecules and cells combined with molecular dynamics (MD) computational analysis as a new approach for optical detection of known potential nucleic acid biomarkers for ovarian disease early detection, analysis, prognosis and treatment. Dramatic differences were observed between the THz absorption spectra of cancer and normal cells fixed in alcohol with much higher absorption intensity and a very strong absorption peak or a relatively narrow sub-band of peaks dominating at wavenumbers around $\sim$ 13 cm ${}^{-1}$ (frequency 390 GHz) in the spectra from cancer samples. By applying molecular dynamics simulation for analysis of experimental results we demonstrated that this specific cancer signature can originate from microRNAs, particularly the mir-200 family, which has been shown to be significantly overexpressed in ovarian cancer in numerous publications [12]. A decade ago, microRNAs (miRNAs), a novel class of evolutionarily conserved small (18–24 nucleotides) non-coding RNA molecules, were discovered as important regulators of gene expression [13]. Since miRNAs modulate the expression of critical genes and the signaling networks involved in tumorgenesis, there is sufficient evidence that microRNAs are involved in cancer development [14]. In cancer cells, microRNAs become deregulated showing increased concentration in blood, serum and tissue cells [15, 16]. According to experimental studies, miRNAs can be overexpressed by 50–600 times during cancer progression [17, 19]. Recently, it has been reported that serum and other body fluids contain sufficiently stable miRNA signatures. Thus, the profiles of circulating miRNAs have been explored in a variety of studies aiming at the identification of novel non-invasive biomarkers.

In our work [11] the signature peak at 13 cm ${}^{-1}$ was observed both in cells and in the media surrounding the cells. This is not surprising, as microRNAs are known to be excreted from cells in exosomes, and the miRNA abundance profile of bodily fluids has been shown to reflect physiological and/or pathological conditions [20, 21, 22, 23]. In addition, miRNA signatures are known to be different between ovarian carcinoma histotypes (serous, endometrioid, clear cell, and mucinous) [24, 25] potentially allowing for both, detection and identification of the cancer sub-type with a simple blood-test. However, while miRNA has been proposed as stable biomarkers in blood for cancer and other diseases, quantification of miRNA remains a time and labor intensive process. The THz spectroscopy technology introduced in [11] has the potential to address the current shortcomings of biomarker analysis for early detection, while also providing a rapid, objective analysis for confirmation of pathological examination. This technology utilizing wavelengths beyond those traditionally used for chemical and biomolecular analysis is based on interaction of electro-magnetic (EM) radiation in the Terahertz (THz) and sub-THz regions with low-frequency internal molecular motions involving hydrogen bonds and other weak bonds (see review in book chapters [26, 27]). With low absorption of water in this frequency range compare to IR, the technique is useful not only for analysis of biological molecules, but also for cellular analysis. Since the wavelengths of THz radiation in our spectrometer are large (1 mm–0.6 mm) compared with the size of biological cells (2–20 $\mu$ m) this radiation propagates through an entire cell, revealing spectroscopic information from molecular components inside the cell [28]. Thus, the resulting spectra are summations of the constitution and character of the molecular structure and composition of the cell. It means the complicated procedures of extracting biomarkers from the cells would not be necessary and could be eliminated. As this is a label free, non-destructive optical method, the sample also remains available for subsequent analysis using other analytical methods.

In this work we address the issue of potentially utilizing sub-THz spectroscopy for rapid, objective analysis of ovarian tissue biopsy samples for cancer diagnosis, using results from our previous paper [11] in the analyses of our new findings. Spectroscopic characterization of tissue samples in this study was performed using our continuous wave, frequency domain automated spectrometer operating at room temperature in the spectral region of 310–500 GHz [29, 30]. The instrument has been extensively used during more than 5 years for spectroscopic characterization of biological macromolecules and species. Combined experimental and computational modeling results confirmed in all cases that observed spectroscopic features are due to fundamental physical mechanism of interaction between THz radiation and biological macro-molecules inside bacteria or cell [11, 31, 32, 33, 34, 35]. In this study we observe biopsy samples showing differences between normal and cancer tissue as was observed with the cell line samples and in the liquid media, in which the cells were suspended [11].

While in this paper, we narrow our focus to discrimination based on the distinctive high absorption of sub-THz radiation by microRNAs, it must be emphasized that sub-THz spectroscopy is broader than simple microRNA analysis. This technology has the potential as a new method to detect cancer cells and neoplastic tissue based on the summation of many molecular characteristics within the material. MicroRNAs however have advantages over other molecules since they are small, very stable in blood circulating between all organs, their abundance and profile are related to pathological conditions, and they can carry specific information about crucial sequence modifications in DNAs and messenger RNAs that are related to cancer development and transfer this information to proteins. We expect that the results of this study will create a basis for the development of a novel, sensitive, small sample resonance spectroscopy technology in the sub-Terahertz frequency range, complimentary to existing methods, as an optical, label-free and reagent-free approach for visualization and quantification of microRNAs and other molecules to discover and study potential biomarkers for diagnosis, prognosis, early detection and treatment selection for ovarian and other cancers.

Table 1

Pathological sample characteristic

Code	Site	Diagnosis	Grade	pT (tumor stage)	pN lymph node stage	pM metastasis stage
TG-1	Fallopian tube	Normal mucosa
TG-2	Fallopian tube	Normal mucosa
TG-3	Fallopian tube	Normal mucosa
TG-4	Ovary	Serous papillary	High grade	3B ( $<$ 2 cm peritoneal	X (lymph nodes not sampled)	1 (known metastases)
		carcinoma		implants)
TG-5	Ovary	Serous papillary	High grade	3C ( $>$ 2 cm peritoneal	0 (no lymph node metastases	0 (no known metastases)
		carcinoma		implants)	detected)
TG-6	Ovary	Serous papillary	High grade	3C ( $>$ 2 cm peritoneal	X (lymph nodes not sampled)	1 (known metastases)
		carcinoma		implants)

For pT, the tumor stage, 3B means macroscopic peritoneal implants beyond the pelvis that are 2 cm or less in greatest dimension were found; 3C means macroscopic peritoneal implants beyond the pelvis that are more than 2 cm in greatest dimension were found, pN indicates the lymph node stage, with X meaning no lymph nodes were sampled for histologic examination and 0 meaning lymph nodes were examined but no metastases were detected, and pM is the metastasis stage with 1 meaning known metastases, and 0 means no known metastases.

2. Materials and methods

2.1 Pathological and control samples

A set of samples for this study has been prepared from anonymized archival formalin-fixed, paraffin-embedded ovarian epithelial tissue obtained at the University of Virginia Health System, with IRB approval (Protocol # 13310). Tissue selection for this set of samples was performed by a Board-certified surgical pathologist (C.A.M.) to identify regions of invasive neoplastic cells from cases of high-histologic grade serous papillary ovarian carcinoma, and for controls, normal mucosa from fallopian tubes of patients with no known malignancy undergoing steril- ization procedures. Tissue samples were obtained by histology-guided manual sampling of formalin fixed tissue blocks using 1.5 mm core biopsy instruments. The core samples were minced with a scalpel blade, extracted twice with xylene to remove the paraffin, and then extracted with 100% ethanol. Following drying of any residual ethanol, 200 $\mu$ l of RIPA buffer with phenylmethylsulfonyl fluoride (PMSF) (1 mM) and sodium orthovanadate (2 mM), was added along with a few mg of stainless steel beads (0.9–2 mm). The sample was mixed at 1200 RPM for 15 minutes at room temperature in a vortex mixer (Maxi-Mix III, Thermolyne), then agitated in a Bullet Blender (Next Advance) for 4 minutes on setting 8. These steps were repeated once, then the sample spun briefly to pellet the steel beads, with the supernatant (containing a suspension of tissue fragments and extraction fluid) removed and stored frozen until analysis. Table 1 lists the samples that were prepared for this study, including the stage of each cancer sample based on the TNM staging system.

2.2 Spectrometer and measurement procedure

The spectrometer used for measurements, shown in Fig. 1, was a second generation prototype built by Vibratess. It is a continuous wave, frequency domain instrument operating at room temperature in the spectral region of 310–500 GHz. The system satisfies simultaneously the requirements of sufficiently high sensitivity, required spectral resolution [31], and spatial resolution necessary to interrogate nanogram amount of sample materials. This novel spectrometer is based on a very strong local enhancement of the electromagnetic field (EM) in the metal channels of a sample holder, thus allowing increased sensitivity due to better coupling of the THz radiation with the sample biomaterials [32, 33]. The disposable sample holders contain a microchannel array fabricated by depositing copper (5 $\mu$ m thick) on 25 $\mu$ m thick polyimide, using photolithography to create a pattern containing an array of 21 microchannels, which were 10 $\mu$ m wide on 150 $\mu$ m centers

Figure 1.

Sub-THz Spectrometer with attached syringe (front, left) for loading sample material on a sample chip that is positioned on a movable table under a micro-detector (illuminated). Objective of the visualization system (on the back, right) is focused on the chip from the side. Micro-detector is in the up position to make loading possible.

The experimental measurement procedure is described in details in our previous papers [11, 34, 35]. For each sample, an empty sample holder was first installed into the THz spectrometer and the detection probe was positioned directly above the array ( $\sim$ 6 $\mu$ m) to measure the reference background spectrum. Frequency scans were performed by varying the voltage to the THz radiation source (Virginia Diode, LLC). The sample holder can be moved in two directions with the step of 0.75 $\mu$ m, such that the probe could be accurately positioned along one channel or moved to adjacent channels to allow background scans to be performed at multiple locations close to the center of the array. In the next step, the detection probe was raised, and a dilute solution (suspension) of a sample was added using micropipette or micro-syringe onto a small spot ( $\sim$ 1 mm ${}^2$ ) on the array, where background scans had been performed, and allowed to dry. A typical size of one drop from a micropipette was $\sim$ 0.3 $\mu$ L, and from a micro-syringe from 0.02 to 0.1 $\mu$ L. The detection probe was then repositioned to the same height above the array, at a position where a background measurement had been performed. The frequency was again scanned over the range of the instrument to measure the intensity of radiation passing through the sample material inside the channel. The sample holder stage was them moved to scan at another location on the sample spot, where a background measurement was earlier performed, and the frequency scan repeated. For each sample, multiple measurements were performed, and multiple arrays were utilized with each sample to show measurement repeatability. This protocol permitted also to detect and register the heterogeneity of malignant cells in a tissue suspension.

Sample transmission (T) was calculated as the ratio of the signal spectrum with material to the background spectrum. This practice is justified by a very low refractive index and reflection from sample and substrate surfaces, which might be neglected. Transmission was then recalculated for absorbance A, which is proportional to absorption coefficient, using $A=-\text{log}(T)$ . Scaling absorbance depending on the amount of sample material in the channel under the detector probe, can be used for results quantification and for further comparison with the computational modeling results.

The molecular dynamics simulations protocol and absorption coefficient calculations utilized to predict the absorption spectra for microRNAs duplexes are those described in our previous works [36, 31, 11].

3. Results-absorbance spectra in the sub-THz range

3.1 Sample measurements procedure

Table S1 (see Supplemental Information) lists all measurements of all samples in this study. The samples TG-1, TG-2 and TG-3 were prepared normal controls (NC), while the samples TG-4, TG-5 and TG-6 were found by histology as carcinoma samples.

In this paper, we present the results of the very first attempts to characterize tissue samples using their absorption spectra in the sub-THz spectral range. This is a much more complicated task compared to the characterization of cultured cancer cells in alcohol samples described in our previous paper [11]. The most significant problem that we faced in the current work was due to samples heterogeneity. We were trying to identify signatures of cancer vs normal control tissue, but small pieces of the very neighboring tissue samples can include cells with different type and stage of diseases as well as some normal cells. It is obvious that our results and conclusions will depend on experiment design, especially on differences in sample preparation procedures, amount of material in the sample, averaging procedures, sample thickness and many other details, of which we were not aware in advance. To accumulate this initial experience, many tests were performed to analyze measurements sensitivity, repeatability and reproducibility, as well as to eliminate all possible artifacts. One important criterion validating the procedure and the results is that transmission, calculated as the ratio of radiation intensity passed through sample material to the reference radiation passed without sample can not be greater than 1 at all tested frequencies. Transmission values above 1 would indicate the presence of artifacts, usually due to multiple reflections of radiation in the system that can be affected when sample material is added to the system. In absorbance spectra these artifacts are manifested as having negative values, which does not have physical meaning except the special case when there is emission of radiation from the sample. The artifacts may also happen when too much sample material is applied for characterization, thus modifying the dielectric constant in part of the radiation path, and the 3D standing waves radiation pattern in the system. On the other side, too small an amount of sample material does not provide sufficient sensitivity.

Figure 2a and b give examples of the sensitivity, reproducibility and variability of measurements for the typical drop size of 0.3 $\mu$ L. The results are reproducible within the limit of several % for the amplitude (transmission or absorbance) and less than 0.05 cm ${}^{-1}$ on the frequency position for measurements in the same spot (the same channel of the chip and the same coordinate along the channel) for the same sample.

Figure 2.

Absorption spectra demonstrated measurements procedure: accuracy and reproducibility. a. Absorbance spectra taken over 3 days (left axis) and the error standard deviation, S, from 16 scans of the normal mucosa sample TG-1 in a single sample holder location (brown curve, right axes). b. Reproducibility of multiple runs of cancer sample TG-5 (suspension – no solid pieces).

The Fig. 2a presents repeatability and sensitivity results of absorbance spectra measurements and the error standard deviation S. The amount of dry material in a 0.3 $\mu$ L drop of normal control sample TG-1 in Fig. 2 is $\sim$ 40 ng normalized to proteins. This amount covers an area about 1 mm in diameter on the sample array. The amount of dry sample material actually interrogated in each measurement normalized to proteins is 0.8 ng as calculated for a channel of 10 $\mu$ m width, a 200 $\mu$ m detector opening, and a channel depth of 3 $\mu$ m. The error standard deviation is less than 4% almost at all points, except the spectral sub-ranges between 11.4 and 11.8 cm ${}^{-1}$ and at the frequency $\sim$ 16.4 cm ${}^{-1}$ . Figure 2b shows reproducibility of independent measurements of absorption spectra from cancer sample TG-5 interrogated at different spots of the same channel utilizing the background measurement at each position. Although the intensities may fluctuate, the frequencies of all absorption peaks are reproduced quite well with the accuracy not worse than $\sim$ 0.05 cm ${}^{-1}$ on the frequency scale.

A big difference in absorption from normal control and cancer tissue samples is observed between Fig. 2a and b. In normal control (NC) tissue sample (TG-1), the main peak is at 13.1 cm ${}^{-1}$ (Fig. 2a), compared to the frequency 12.95 cm ${}^{-1}$ for the cancer sample TG-5 (Fig. 2b). Not only the pattern of entire spectrum is different, but the level of absorption for the same amount of material is usually higher for cancer samples compared to normal control. Note, that there was $\sim$ 3–5 times more dry material in the normal control sample (Fig. 2a) compared to the cancer sample (Fig. 2b).

3.2 Cancer tissue characterization

We continue describing the results for cancer sample TG-5. This cancer, classified according to Table 1 as a high grade material (3C), with more than 2 cm peritoneal implants, represents nevertheless the simplest case among cancer samples tested, since it has no lymph node metastases and contrary to TG-4 and TG-6 it has no known metastases. Figure 3a compares the spectrum from the TG-5 tissue sample displayed on Fig. 2b in the previous section with the signature of the ES-2 ovarian cancer cell line sample (E-line) studied in [11] (dashed curve). Not only the structure around 13 cm ${}^{-1}$ is reproduced, but an entire pattern with many other features is reproduced across the frequency scale as well (see peaks at 11.4; 12; 15.45, 15.8 and 16 cm ${}^{-1}$ ). It is an important result that both these epithelial cancer samples, although prepared by completely different technologies, reveal very similar sub-THz signatures with a strong absorption peak positioned at $\sim$ 12.95 cm ${}^{-1}$ . The same specific peak was present in the signature of another sub-type of epithelium ovarian cancer line (SKOV3) studied in [11].

The features centered at 12.95 cm ${}^{-1}$ are very strong and reproducible for absorbance of the same cancer sample TG-5 in two different channels of a single array, as shown in Fig. 3b. We have already demonstrated [11] that the fine structure of absorption in cancer samples around 13 cm ${}^{-1}$ is directly attributed to the presence of double-stranded microRNA molecules having defects after maturation, specifically miR-141 and possibly miR-200c and miR-200a. This conclusion was based on direct comparison of experimental spectra with MD simulations of spectroscopic signatures of microRNAs known to be overexpressed in ovarian cancer. The feature between 11.4–11.6 cm ${}^{-1}$ was also earlier observed in the spectrum of a cancer sample signature [11]. The peaks in absorptions spectra from the tissue sample in Fig. 3a are not as well resolved as in the signature we observed in the E-line samples [11]. This could be expected because of the much more complicated tissue composition, with the presence of other molecules in cancer tissue that would not be found in cell line samples. The main difference between the

Figure 3.

Absorption spectra of cancer samples TG-4, TG-5 and TG-6 are shown in Fig. 3a–h. a. Comparison of absorption spectra from the cancer sample TG-5 (only liquid), red line, with a cell-free (c.f.) ovarian cancer cell sample (E-line) solution from [11]. b. The similarity of spectra from the cancer sample TG-5 (liquid) in two channels with small drops. c. Absorption spectrum of the cancer sample TG-5 from a sample location containing a piece of tissue in the channel as shown in Fig. S1 (supplemental information). d. Spectra of cancer samples TG-5 and TG-6, both high grade tumor and peritoneal implants larger than 2 cm, which are potentially revealed as some similarity of signatures between 13.2 cm ${}^{-1}$ and 13.7 cm ${}^{-1}$ . e. Absorption spectra from cancer tissue sample TG-4 in two channels (one drop $\sim$ 0.2–0.3 $\mu$ l each). f. Absorption spectra from cancer tissue sample TG-4, measured at 2 different times, showing scaling of results. g. Good scalability of Absorption from cancer sample TG-6. The revealed cancer peak is centered at 12.95 cm ${}^{-1}$ as in TG-5 (Figs 2a, 3a and b). h. Variations in absorption of cancer sample TG-6. Center peak is located at 13.1 cm ${}^{-1}$ as in TG-5 on Fig. 3c (green line) and at 13.4 cm ${}^{-1}$ (blue) as in TG-4 on Fig. 3e and f.

two spectra is observed in the band between 13.5 and 15.1 cm ${}^{-1}$ .

At the same time, the central absorbance peak in the spectrum of sample TG-5 with a piece of tissue as shown in the image in Fig. S1 is shifted from 12.95 cm ${}^{-1}$ to the higher frequency close to 13.1 cm ${}^{-1}$ (Fig. 3c). This completely different spectrum reproduced very well in repeat measurements demonstrates the heterogeneity of the same sample material. It might indicate some amount of molecular components other then microRNAs, particularly proteins.

Figure 3d allows us to compare absorption spectra from two sample materials, TG5 and TG-6. These samples both have the same high grade and the same tumor stage 3C with peritoneal implants $>$ 2 cm. Detailed comparison of their spectra identify the regions with some similarities occurring between 13.2 cm ${}^{-1}$ and 13.7 cm ${}^{-1}$ that will be shown later as absent in sample material TG-4 and might be considered as a sign of peritoneal implants. Additional complication for sample TG-6 due to expanding to lymph node and metastases stages compared to TG-5 is probably revealed as specific absorption peaks at frequencies 10.7–10.8 cm ${}^{-1}$ as well as 14.4 cm ${}^{-1}$ and 14.7 cm ${}^{-1}$ .

The results of measurements of cancer tissue sample TG-4 are shown in Fig. 3e. This material is classified as a high grade serous papillary carcinoma with tumor metastasis beyond the pelvis, which means that the tumor had extended beyond the ovary and spread intraperitoneally. The spectra taken in two channels on the same chip indicate that this is also a heterogeneous material. In channel 0 (center channel of array), the two specific absorption peaks occur at $\sim$ 13.2cm ${}^{-1}$ and 13.4 cm ${}^{-1}$ , similarly to absorption of TG-6 in Fig. 3d. These high intensity peaks are of interest for further clinical correlation as they exist in samples TG-4 and TG-6, but are absent in samples TG-5, although all 3 tumor samples were similar histologically. The absorption peak from the sample material in the adjacent channel is not so intense and the spectra in the two channels reveal different fine structure.

Spectra shown in Fig. 3f taken also from cancer material TG-4 confirm the presence of intense absorption peaks at frequencies 13.1 and 13.4 cm ${}^{-1}$ that again are above frequency 12.95 cm ${}^{-1}$ as found in S and E-lines cancer cells in [11]. Possible assignment of all 4 microRNAs highly overexpressed in ovarian cancer to specific absorbance peaks that will be discussed in Section IV is shown in Fig. 3f by arrows. Again, the very different pattern of absorbance spectra of TG-4 samples in Fig. 3e and f compared to the pattern of TG-5 sample in Fig. 3a might indicate that this analysis is detecting molecular differences in tumor samples that are similar in histologic morphology.

The spectra shown in Fig. 3g (cancer sample TG-6) represent an advance stage of cancer (high grade 3C with peritoneal implants and with one known metastases). This Figure demonstrates high consistency of spectra scaling with the amount of sample material. The left side of the spectrum includes the absorption feature at $\sim$ 12.95 cm ${}^{-1}$ observed in these tissue samples that is very similar to the peak in the spectra of ovarian cancer cell line samples [11] and in sample TG-5 shown in Fig. 3a and b. At the same time, the spectrum from tissue of TG-6 in Fig. 3h (green line) is similar to that of cancer TG-5 in Figs 2b, 3a and b with the strongest peak at the frequency 13.1 cm ${}^{-1}$ , thus indicating the different cancer subtypes according to the classification in the Table 1. Although the spectrum from a sample containing only liquid phase (Fig. 3h in blue line) also shows a small peak at 13.1 cm ${}^{-1}$ , it looks very different with low absorption below frequency 13 cm ${}^{-1}$ . Thus, spectra from samples TG-6 shown in Fig. 3g and h demonstrate variability that might confirm the most advance cancer that includes features from ovarian cancer itself as in samples TG-5 with the central peak at $\sim$ 12.95 cm ${}^{-1}$ as well as a central peak shift to the frequency of 13.1 cm ${}^{-1}$ or higher as in TG-4 that possibly belongs to cancer extension.

All these facts might suggest that the peak centered at 13.1 cm ${}^{-1}$ is the sign of cancer expansion to the lymph node or metastases – a conclusion confirmed by the presence of a similar central peak in the spectrum of cancer TG-4 in channel 0 (green line) in Fig. 3e. Both sample materials, TG-4 and TG-6 show a strong absorption peak centered at $\sim$ 13.1 cm ${}^{-1}$ , which indicates the cancer presence, and additional pronounced features in the 11–13 cm ${}^{-1}$ frequency range. Thus, the shift of the central peak position on the frequency scale from 12.95 to 13.1 cm ${}^{-1}$ might again indicate the difference in the cancer lines and micro RNAs involved.

Figure 4.

Absorption spectra of normal control samples TG-1, TG-2 and TG-3. a. Variability of absorbance spectra from a large drop of NC TG-1 in two channels demonstrates the noticeable difference between 12.8 and 14 cm ${}^{-1}$ . b. Absorbance spectra from a piece of tissue (brown) and surrounding liquid (blue) in the sample drop of NC TG-2. c. Absorbance spectra from two different drops of the sample NC TG2. d. Absorbance spectra from NC TG-3 sample containing solid fragments, in blue, (E179), and only liquid material, in rose, (E169).

Comparison of signatures presented in this section showed that not only the exact frequency of the largest absorption peak around 13 cm ${}^{-1}$ can be different, but the entire pattern of the absorption spectrum of cancer tissue in the spectral range available in our instrument can be different. This is true even in the same sample of material, leading to an obvious conclusion of heterogeneous character of samples taken from a patient. This result, that cells with different stage of disease coexist in a very close proximity in the same tumor, is natural and could be expected. We can see that depending on the stage, not only the intensity of the absorption peak around 13 cm ${}^{-1}$ will be changed, but the contribution from absorption bands of other components will be modified.

3.3 Normal control tissue characterization

Normal Control (NC) tissue samples are even more difficult to characterize since absorption of these materials is usually much lower compared to cancer tissue for samples with the same amount of material. There are possibly a larger number of molecular components that are present in normal control tissue compared to cancer samples and individual features can originate from different sources. Tissue includes cells with all their molecular components (proteins, DNAs, RNAs and others) as well as capsules, lipids, basement membranes and other components that are transparent for THz radiation. However, if noticeable absorption peaks in the sub-range of 12.9–13.4 cm ${}^{-1}$ are present in the spectra of samples in the NC category, this potentially can be an indicator of at least an initial cancer stage.

Repeatability and sensitivity results of absorbance spectra from NC sample TG-1 have been already demonstrated in Fig. 2a, Section 3.1. Several big drops of material were applied to obtain the absorption spectra from TG-1 and in 2 different channels of a sample holder shown in Fig. 4a. Although the spectra are not identical, the entire pattern is rather reproducible. As expected, the spectra reveal many similar features, but other features specifically between 12.7 and 14 cm ${}^{-1}$ are different. We do not expect complete reproducibility for different spots and different portions of the same sample material because of possible variation of sample internal characteristics. These differences might reflect the contributions from different molecules present in the tissue samples.

A low intensity shoulder in absorption at 12.95 cm ${}^{-1}$ is revealed in Fig. 2a and in CH:0 in Fig. 4a (blue curve) for sample TG-1, and a small peak is present in the spectrum of this sample in CH:-1, which might be a first sign of cancer, but more normal samples will have to be examined to fully evaluate this possibility. The other features of the NC pattern in Figs 2a and 4a include a broad and relatively intense absorption band between 13 and 14 cm ${}^{-1}$ that might originate from proteins or DNA/RNA molecules, but also from specific microRNA. The pattern between 13 and 14 cm ${}^{-1}$ for this normal control tissue sample TG-1 is somewhat similar to that of a cancer tissue sample TG-4, file E194 in Fig. 3f. A strong peak or several peaks just above the frequency 13 cm ${}^{-1}$ look like characteristic features of a cancer as demonstrated in [11], but here they are shifted to higher frequencies. These features and their differences might be important. Further discussion will be presented in Section 4.

One reason for a big difference between signatures from solid pieces of tissue and just the homogenized liquid for control sample TG-2 shown in Fig. 4b is due to the amount of material. In concentrated tissue at the edge of a sample drop, with multiple solid pieces present, there is again a low intensity shoulder around 12.95 cm ${}^{-1}$ (brown curve), which might be a sign of initial cancer stage, but the total absorbance is still low.

The absorption spectrum in blue for a small dried liquid portion of the same sample, further from the edge of the drop, is better resolved with more narrow and less overlapping of neighboring features, although the sensitivity is not enough to see the entire signature because of not enough material. Many features present in a tissue sample containing solid fragments (in brown) are completely absent in spectra from a liquid phase sample, however, note the broad absorbance band around 16 cm ${}^{-1}$ that is present in both spectra of Fig. 4b. The pattern of both absorption spectra in the sub-range between 12.9 cm ${}^{-1}$ and 13.8 cm ${}^{-1}$ is similar to the pattern in sample NC TG-1 in Fig. 4a. There are some similarities in the pattern of spectra in the sub-range 12–12.7 cm ${}^{-1}$ shown in Fig. 4a and b, but otherwise the NC signatures on these two figures are very different. Note a broad absorbance band around 16 cm ${}^{-1}$ that is present in both spectra of Fig. 4b.

Very different absorption spectra from the same control tissue material of TG-2 are shown in Fig. 4c demonstrating again variability of spectra. In File E180 (in blue), a large amount of material was applied (0.5 uL), which resulted in poor resolution of many features below 12.5 cm ${}^{-1}$ . Using a smaller amount of the same material TG-2 (0.3 $\mu$ L drop, in red) resulted in better resolved spectral absorbance features. In both cases, a split absorption peak around 12.95 cm ${}^{-1}$ is observed, which was a characteristic feature of cancer presence [11]. This peak however does not dominate the spectra as was observed in the cancer cell lines. This result is important for the analysis of spectra variation depending on concentration. It can also be used for a better choice of material amount to place more dry material in the channel to improve the sensitivity.

Low intensity absorbance spectra from control sample TG-3 in two files, E169 and E179, shown in Fig. 4d were obtained from relatively large amount of solid free material in two separate measurements. There remains a pronounced absorbance peak centered at $\sim$ 13.1 cm ${}^{-1}$ with the signature most resolved when the sample was deposited using a micro-syringe resulting in the smallest drop with mainly liquid material. Variability of spectra from NC TG-3 sample is also observed in Fig. 4d: solid tissue in the File E179 and only liquid in E169. We expect more contribution from tissue proteins in the sample E179, in particular, as in liquid material mostly DNA and RNA are present. The origin of observed peaks will be discussed in the Section 4.

4. Analysis of experimental results

Figure 5a demonstrates the absorption spectrum from a very small amount (0.02 $\mu$ L) of the normal control tissue material TG-3 deposited using a micro-syringe. Very sharp absorption sub-bands with resolved fine structure are observed at a frequency around 13.1 cm ${}^{-1}$ similar to the features of normal control tissue sample TG-2 shown on Fig. 4b (in brown), Fig. 5b (in blue), and to cancer tissue sample TG-5 (Fig. 3c). More work is needed along with follow up testing to determine if liquid phase spectra from Fallopian tube mucosa might be able to indicate early signs of cancer.

Table 2

Studied microRNAs found in http://www.mirbase.org using entries in the database as indicated

Name	Sequence	Entry in the database
miR-200a (22 bp)	CAUCUUACCGGACAGUGCUGGA	MI0000737
miR-200b (22 bp)	UAAUACUGCCUGGUAAUGAUGA	MI0000342
miR-200c (22 bp)	CGUCUUACCCAGCAGUGUUUGG	MI0000650
miR-141 (22 bp)	UAACACUGUCUGGUAAAGAUGG	MI0000457

Figure 5.

Absorption spectra of normal control samples: experiment and simulation. a. Absorption spectrum from very small amount of NC TG-3 (0.02 $\mu$ l liquid). Peaks at 13.1 cm ${}^{-1}$ and 16.05 cm ${}^{-1}$ are very strong and sharp. b. Comparison of molecular absorption signatures from NC TG-2 and cancer TG-4 samples. c. Absorption spectra of four micro RNAs: miR-200a, -200b, 200c, and miR-141. Modeling molecules with mismatches. d. Comparison spectra from control tissue sample NC TG-2, having multiple cells under the probe, with model prediction.

Our analysis showed major and repetitive spectral differences between tumor samples and normal samples that may form the basis of future diagnostic testing. Additionally differences in spectra between the tumor samples were observed, not explained by comparison of histologic morphology or major differences in clinical stage, indicating that this technique identifies molecular differences between samples that may correlate with other biological attributes of the tumors that are not captured by the limited clinical data in this pilot study.

Figure 5b compares spectra from NC sample and cancer TG-4 to demonstrate that many similar signatures can be present, specifically at frequencies below 12.9 cm ${}^{-1}$ and above 14.3 cm ${}^{-1}$ that reflect the presence of the same molecular components. The major differences, which can be considered as cancer identifiers, are revealed between 12.9 and 14.3 cm ${}^{-1}$ .

The latest results from the medical literature suggested crucial role for the microRNA-200 family in Ovarian Cancer with high overexpression levels of these molecules (see, for example, recent review [37]. Published results of our research in [11] confirm that the contribution from miR-141 and miR-200c determines the fine structure of the dominating absorption sub-band from epithelial ovarian cancer cell lines ES-2 and SKOV3, with the central peak at frequencies $\sim$ 12.95 cm ${}^{-1}$ .

In this study we continued the analysis and have performed MD simulations of absorption spectra from all four micro RNAs of this family shown on Table 2 to compare their signatures with experimental results.

Figure 5c shows the simulated absorption spectra of these 4 molecules in the frequency range of our spectrometer, as average of all orientations. Two of these molecules, miR-200a and miR-200c show significant absorption peaks centered at frequency 13.05–13.1 cm ${}^{-1}$ that matches with cancer samples. In addition miR-141 shows a peak centered at frequency 12.95 cm ${}^{-1}$ followed by consecutive peaks up to 13.5 cm ${}^{-1}$ .

In Fig. 5d we compare the experimental spectrum taken from the normal tissue sample NC TG-2 (File E162 with multiple cells present in the channel inside the material spot under the probe) with the results of computational prediction for 4 micro RNAs mixed in a fixed ratio. Reasonably good correlation is observed for weight of components; miR 200a-25%, miR 200b-10%, miR 200c-40% and miR-141-25%. The central peak at 13.1 cm ${}^{-1}$ is well defined, although its intensity is comparable with other peaks. The absorption peaks on both sides from the central peak at 13.1 cm ${}^{-1}$ are very well resolved. The fact that it is possible to represent a large number of peaks in the spectrum from tissue as a contribution of just 4 molecules is surprising on first glance and requires an explanation. We think that although tissue samples have many molecular components, most of them including lipids, hydrocarbons, and cell membranes are transparent in the sub-THz range (or have very low capability to absorb sub-THz radiation). However we can not completely eliminate possible contribution of proteins to observed signatures from NC tissue samples because some proteins give the absorbance pattern between 13 and 16.5 cm ${}^{-1}$ similar to that observed for NC samples. Much more experimental work is required to come to definite conclusion regarding what molecular components contribute most to the NC samples signatures.

We were not able, however, to get a good correlation between measured cancer tissue samples and simulated miRNA spectra because, as it was indicated in Section 3.2, the individual peaks are not well resolved. Very intense absorption peaks in the simulated spectrum at frequencies 11–12 cm ${}^{-1}$ , 14 cm ${}^{-1}$ , 14.9 cm ${}^{-1}$ and 16.2 cm ${}^{-1}$ are damped in the tissue samples probably due to contributions from other molecular components present in tissue, particularly proteins, which do not have absorption features around 13 cm ${}^{-1}$ . The right side of the measured absorption spectra in cancer tissue samples at frequencies above the major peak usually has intensity that can not be explained by the same micro RNA molecules. One possible explanation is further changing of microRNA spectra in advance cancer stages due to continuing mutations.

It is too early now to make more detailed analysis of specific molecules contributing signatures of normal samples. The main differences between cancer and normal samples from tissue is however identified: 1) For the same amount of material, cancer samples have significantly higher absorption intensity over the entire spectrum, and 2) There is a specific absorption pattern at frequencies 12.8–13.5 cm ${}^{-1}$ – cancer identifier, that is absent in normal samples. In addition, the details of this pattern might permit to discriminate cancer subtypes including extension beyond the ovarian themselves as well as expansion to lymph nodes and metastases.

5. Conclusions

In this work we applied sub-THz vibrational spectroscopy and Vibratess’ spectrometer having high spectral and spatial resolution to characterize absorption spectra from tissue samples and to discriminate 3 high grade serous papillary carcinoma samples taken from ovaries and 3 normal mucosa tissue from fallopian tubes. Requirements for the amount of material in spectroscopic samples and other details of sample preparation as well as experimental procedure have been identified to receive meaningful results. Reproducible spectra have been demonstrated using the set of tissue samples with known pathological status. We demonstrated the heterogeneity of spectroscopic samples prepared from the same tissue material on the scale of hundreds of microns. The characteristic spectroscopic features in absorption spectra have been identified as cancer indicators. We demonstrated that using sub-THz spectroscopy in a sub-band of 10.4–16.9 cm ${}^{-1}$ ( $\sim$ 310–490 GHz) we can easily discriminate between advanced cancer and normal tissue. These new findings open potentially new applications of vibrational spectroscopy as a complementary technique for early ovarian cancer detection and to guide this cancer therapy.

It is clear from all presented experimental results and simulations that micro RNA molecules contribute to spectroscopic signatures from tissue samples in our spectral range. Based on the preliminary data generated here, we conclude that the presence of absorption peaks near 13 cm ${}^{-1}$ is a sign of cancer presence. As such, it will be worth exploring if this signature can be detected in biofluids from ovarian cancer patients. If strongly correlated with cancer burden, it may then be investigated as a potential new biomarker for disease monitoring, and also perhaps as a biomarker for cancer screening.

The most significant problem that we faced in the current work was due to samples heterogeneity, which was reflected by diverse spectral signatures. At the same time this attribute provides additional, very specific information that may be used for identification of cancer subtypes, clinical behavior or sensitivity to specific therapies. This will need to be developed by the application of this technique to an expanded number of cancer samples with annotated clinical data. Because sub-THz spectroscopy provides information not reflected in tissue morphology, there is promise for this methodology as a supplemental analysis to traditional tissue diagnostic approaches.

In a subsequent paper, we plan to use fitting programs to compare all simulation results with experimental data and receive more detail information about contributions from specific micro RNA to the spectroscopic signature in each patient sample. By targeting molecular detection, our sub-THz spectroscopic instrumentation promises to simplify and facilitate diagnostic procedures and help in discovering and deciphering the basic mechanisms underlying cancer initiation and its progression. This can provide results which might then be used as guidance for personalized medicine.