This study is the first (to our knowledge) to generate FTIR spectra from sputum and derive chemical fingerprints for the purpose of diagnosing lung cancer.
With the knowledge that FTIR yielded excellent reproducibility for sample spectra, our primary objective in this study was to determine which wavenumbers were significantly different between sputum of cancer and normal controls. Prominent significant wavenumbers could then be used to explore the structure of patterns of similarities within and between both cohorts using MVA techniques. The data analysis strategy we employed was robust and took into consideration the data distribution at each wavenumber. We have found that many FTIR studies apply parametric tests to wavenumber data with no evidence of data distribution yet we found that data at all wavenumbers in this study did not follow a normal distribution. Interestingly, Whiteman et al
] also made the same observation in their study on FTIR spectra in sputum of COPD patients.
The six peaks described in Table all arose due to an increase in absorbance at that spectral position in cancer relative to normal controls often with a noticeable position shift. A rise in absorbance at a wavenumber in one sample relative to another can be due to different reasons including an increase in the frequency of a bond vibration mode [27
]. It should also be noted that the non-uniform distribution and degree of compaction of molecules within cells can also have a non-linear affect on absorbance level as considered for chromatin within dividing and non-dividing cells [30
]. Stronger intermolecular interactions at bonds, such as C-O in carbohydrates or COO- in proteins, result in higher absorbance levelsIt is possible that an increase in absorbance levels at 1656 cm-1
and 1577 cm-1
is due to an increase in protein levels between cancer and normal cells in sputum [27
Band shifts and significant differences in absorbance intensity at 1024 cm-1
and 1049 cm-1
may signify an increase in cancer sputum of levels in glycogen. A previous study by [18
] using FTIR suggested that a glycogen band at 1045 cm-1
was increased in lung tumours (squamous cell carcinoma and adenocarcinoma) relative to normal tissue. A subsequent study using FTIR microscopy confirmed increased levels of glycogen in lung tumour cells [19
]. FTIR showed glycogen levels were also increased in lung tumour cells of pleural fluid due to an increase in absorbance at 1030 cm-1
in the glycogen rich region [24
]. An increase or decrease in glycogen levels varies according to cancer type, for example levels are higher in colon tumour tissue relative to normal but a reduction is seen in tumours of the cervix and liver [18
]. Indeed, the significant increase in absorbance levels at 1024 cm-1
and 1049 cm-1
in lung cancer sputa observed in our study contrast the decreased levels for these wavenumbers in tumours of the cervix compared to normal tissue [27
]. Importantly, if differences in absorbance at 1024 cm-1
and 1049 cm-1
wavenumbers between lung cancer and normal sputa cells represent glycogen levels then our results confirm previous studies and suggest that increased glycogen levels, via detection by FTIR, could prove to be a powerful diagnostic factor for lung cancer.
The significant band shift at 964 cm-1
has been associated with symmetrical stretching in bonds of phosphorylated proteins but has also been associated with cancer related structural change in nucleic acids [28
]. The band increase (and shift) at 1411 cm-1
in cancer sputum spectra might be an indicator of a change in nucleic acid level [31
] alternatively, this change could also suggest COO- stretching and C-H bending due to proteins. Our results suggest that further work should explore the contribution different molecules that differentiate lung cancer from normal sputum spectra so accurate assignments of the causation of absorbance changes can be made.
An important consideration when generating FTIR spectra from biospecimens such as sputum is that cells will often be derived from mixed tissue types which can lead to spurious results [32
]. In this study, we ensured that each sputum sample was assessed by a pathologist for sufficient presence of bronchial epithelial cells. Indeed, validation of our work could involve analysis of cell pellets by FTIR microspectroscopy to generate spectra for pre-identified normal bronchial and tumour cells.
Using 5 significant wavenumbers we were able to visualize the underlying structure of how all sputum samples grouped according to patterns of differences in absorbance levels. HCA and PCA in combination allowed visualization and interpretation of spectral sub-groups. The observation from the HCA dendrogram (Figure ) that no sub-groupings emerged due to histological type suggests that the 5 significant wavenumbers are not type specific. Perhaps a much larger set of samples of NSCLC subtypes and especially small cell carcinoma cases may reveal FTIR spectral differences according to tumour sub-type.
PCA gave further insight into the causal relationship between groupings of spectra and individual significant wavenumbers with 3 components explaining 96.1% of the variation. The first two components (PC1 and PC2) show that cancer spectra clearly separate from normal spectra according to the loadings on the protein, glycogen and DNA associated wavenumbers 1577 cm-1, 1024 cm-1and 1411 cm-1. These wavenumbers are thus important potential diagnostic markers for lung cancer. PC3 was highly associated with the two spectra for patients who had previously been diagnosed with invasive ductal carcinoma of the breast. This result is interesting as both cases had a final histology of NSCLC yet, PCA reveals that both spectra have a high similarity to each other but are separated from other lung cancer spectra. Although data is extremely limited one might hypothesize that FTIR has the potential to further discriminate metastatic tumours where the primary arose in the breast.
Throughout the analysis we were mindful of confounding variables that might lead to misinterpretation of differences between cancer and normal sputum spectra. It is suggested that inflammation plays a key role in the pathogenesis of lung cancer [33
]. From the patient medical histories recorded we noted conditions that could contribute to inflammation in the bronchial tubes. For example, a number of cancer cases had also been diagnosed with COPD or asthma according to standard criteria. Furthermore, the control group also included cases with COPD and asthma. However, an inspection of the grouping patterns of HCA and PCA did not reveal any similarities either within group or between groups due to the presence of these conditions. It is interesting to note however that spectra of nearly all the normal cases who had presented with a cough (prior to sputum acquisition) were more dissimilar to the large sub-cluster of normal spectra in the HCA dendrogram. We were not however able to find any association of wavenumbers with these few cases using PCA.
The spectra of cancer and COPD from sputum can be further compared in detail. Whiteman et al
] compared the FTIR spectral profiles from sputum of 15 COPD patients and 15 healthy volunteers. That study yielded reproducible spectra from sputum with no significant difference between patterns in smokers and non-smokers, factors that are mirrored in our study. The key findings of the COPD study were that major spectral changes between groups were observed as peak shifts in the regions of 1559 cm-1
, 1077 cm-1
and 1458 cm-1
. Thus, in sputum, the significant pattern of change in FTIR spectra of COPD patients is different to that seen in cancer patients. Whiteman et al
. conclude from their study that the main contributor shaping the heterogeneous FTIR spectrum in COPD patient sputa is in vivo
airway inflammation. If this is the case then airway inflammation is not a major contributor to the lung cancer sputum spectrum strengthening the argument that the molecular changes observed are cancer-specific.
It was also important to ensure that absorbance at key wavenumbers were not changed in cancer sputa simply due to differing levels of mucus despite the removal process. Absorbance levels of key mucus related peaks at 1076 cm-1 and 1120 cm-1 were either very low or non-existent. Absorbance levels of another mucus related peak at 1040 cm-1 were more difficult to establish as this wavenumber was situated in the shoulder of the glycogen related 1049 cm-1 peak. Removal of the 1049 cm-1 wavenumber during analysis ensured that differences between cancer and normal sputa were not influenced by presence of mucus.
Although the HCA dendrogram demonstrates a clear separation between the major cancer and normal clusters two normal spectra did group with cancer spectra. Thus, an important question arising from this study is: what are the potential levels of specificity and, more importantly, sensitivity when using the panel of wavenumbers to discriminate cancer from normal sputum? An exact figure should not be estimated from just 50 cases but the grouping patterns observed using MVA suggests that sensitivity and specificity would be at least greater than 80% which compares more than favourably with existing methods of lung cancer detection.