Results
Qualitative analysis
A total of 20 distinct peaks were identified and analyzed, enabling differentiation between normal tissue, ductal carcinomas, and metaplastic carcinoma (Fig. 2). The peaks at 3,280, 2,920, 1,744, 1,632, 1,543, 1,535, 1,528, 1,453, 1,446, 1,394, 1,386, 1,304, 1,274, 1,237, 1,230, 1,162, 1,155, 1,080, 1,073, and 1,043 cm−1 were characterized. Generally, most peaks corresponding to carcinomas exhibited a shift toward lower wavenumbers compared to normal breast tissue, indicating a rightward spectral shift (Table 1). This shift is consistent with specific molecular assignments presented in Table 2.11,13–17,19,24–40Table 1 reveals a distinct downward shift for peaks at 3,284, 2,922, 1,744, 1,535, 1,397, 1,308, 1,237, and 1,080 cm−1 in ductal carcinoma in situ, invasive ductal carcinoma, and metaplastic carcinoma relative to normal breast tissue, suggesting conformational changes in the biomolecular assignments corresponding to these peaks. In contrast, peaks such as 1,632 and 1,444 cm−1 shifted upward, with the former being more sensitive to protein conformation.
Table 2Peak ranges and their corresponding biochemical assignments
Peaks | Biochemical assignment | References |
---|
3,273–3,284 | Amide A | 19,28,35,36 |
2,916–2,920 | Lipids | 13,26–29,35 |
1,529–1,535 | Amide II | 13–17,24,25,38 |
1,384–1,396 | Lipids, Proteins | 28,35,36,39 |
1,274, 1,304 | Collagen | 11,32,35,36 |
1,232–1,237 | Collagen, Phosphate | 28,35,37 |
1,073–1,080 | Carbohydrate, Phosphate | 28,33–35 |
1,038–1,045 | Glycogen | 27,28,35,39,40 |
Normal breast tissue exhibited elevated absorbance in the higher wavenumber range (3,300–3,200 cm−1), followed by ductal carcinoma in situ, compared with other breast carcinoma categories. Invasive ductal carcinoma and metaplastic carcinomas showed particularly increased absorbance within the 1,400–1,000 cm−1 range compared to normal and in situ breast tissue.
Hierarchical clustering analysis (HCA)
The a-fHCA conducted on the specified peaks illustrated varying degrees of relatedness and diversity, attributed to similarities and differences in the vibrational modes of various breast tissue types (Fig. 3a–f). Between normal and DCIS, the spectral features exhibited distinct groupings based on wavenumber proximity. Peaks at 1,162, 1,154, 1,304, 1,274, 2,922, 1,080, 1,073, and 1,043 cm−1 formed a cohesive cluster, as did the sets 1,237/1,230/1,394/1,386/1,453/1,446 cm−1 and 1,535/1,528/1,543/1,632 cm−1. These groupings displayed significantly greater spectral relatedness than, for example, the relationship between peaks at 1,230 and 3,280 cm−1 or 3,280 and 2,922 cm−1. Notably, regions associated with protein vibrations were spectrally distant from clusters of lower-wavenumber peaks. In contrast, the peak at 1,744 cm−1 demonstrated the most significant dissimilarity from all other spectral clusters examined. When comparing normal and metaplastic carcinomas, vibrational spectroscopy identified key spectral signatures (1,394–1,043 cm−1) demonstrating homology within clusters. Specifically, peak intensities at 1,162–1,632 cm−1 displayed the greatest discriminatory power between normal and metaplastic carcinomatous tissues. Secondary discriminatory potential was observed at 1,237 cm−1 and 1,304 cm−1, with diminished efficacy at 1,073 cm−1, 1,155 cm−1, and 1,543 cm−1, correlating with increased Euclidean distances between the compared groups. Peaks such as 1,453, 1,386, and 1,543 cm−1 exhibited marked dissimilarities when compared to clusters of peaks at 2,922, 1,744, 1,446, and 1,304 cm−1, with a linkage distance of 10. Conversely, the amide peaks at 1,528, 3,280, 1,535, and 1,632 cm−1, despite their conserved features, displayed significant diversity from other peaks, with a linkage distance of 25. Notably, peaks at 2,922 cm−1 and 1,744 cm−1, associated with lipid structures, were more closely grouped, indicating better clustering with normal rather than metaplastic carcinoma peaks. Similarly, spectral peak pairings exhibiting tight clustering and potential for discriminating between normal and IDC include 1,162/1,152, 1,304/1,274, 1,080/1,073/1,043, 1,394/1,386, 1,237/1,230, 1,453/1,446, and 1,535/1,528 (Cluster 1). Substantially more dissimilar yet related pairings include 1,162/1,304 and 1,535/1,543 (Cluster 2); 1,394/1,237, 1,237/1,453, and 1,394/1,453 (Cluster 3). Remaining clusters consist of 1,073/1,386 and 1,528/1,632 (Cluster 4); 1,274/1,043 (Cluster 5); 3,280/1,543 (Cluster 6); and 1,744/1,237 (Cluster 7). Furthermore, spectral analysis revealed distinct peak clusters (Cluster 1: 1,394, 1,386, 1,453, 1,446, 1,237, 1,230, 1,080, 1,073, 1,043; Cluster 2: 1,162, 1,155, 1,304, 1,274, 2,922, 3,280) alongside independently clustered peaks (1,535, 1,528, 1,543, 1,632). This configuration suggests potential biochemical relationships between DCIS and rarer metaplastic carcinomas. Additionally, similarities between Clusters 1 and 2 raise the hypothesis that peak 1,274 may exhibit characteristics analogous to peak 1,237, potentially due to concurrent biochemical modifications influencing its spectral signature. In differentiating DCIS and IDC, infrared spectroscopy reveals distinct peak clusters exhibiting equidistant spacing, suggesting underlying molecular regularity. Specifically, peaks at 1,162–2,922 cm−1 and 1,043–1,632 cm−1 demonstrate such patterns, predicting that a shift toward invasiveness may be accompanied by differential chemical modifications influencing peaks at 1,304 cm−1 and 1,043 cm−1, potentially altering their vibrational modes compared to previously characterized clusters. Euclidean distance analysis identifies spectral peak combinations of 1,394/1,386, 1,080/1,073, 1,073/1,043, 1,080/1,043, 1,237/1,230, 1,453/1,446, and 1,162/1,155 as highly discriminatory between IDC and metaplastic carcinoma. However, spectral overlap is observed in peak combinations such as 1,304/1,274, 1,162/1,304, 1,162/1,274, 1,535/1,543, 1,394/1,073, and 1,237/1,453, suggesting potential concurrent molecular alterations that blur distinct spectral signatures.
ROC curve and peak analysis
The selected peaks underwent ROC analysis, employing bootstrapping and Bonferroni correction to account for sample size and to evaluate parameters such as AUC, sensitivity, and specificity between normal breast tissue and breast carcinomas, as well as among the breast malignancies, as detailed in the figures and Tables 3 and 4. Peak 3,280 cm−1, likely corresponding to Amide A,19,26,28,35,36 along with peaks 1,543, 1,535, 1,528, and 1,632 cm−1 (associated with beta-pleated sheets in Amide I and II) and 1543 (assigned to α-sheet of Amide II) (Table 2),13–17,24,25,38 produced moderate to excellent AUC values (0.93) and exhibited high sensitivity (100%) with specificities ranging from 80% to 60% (Table 3 and Fig. S2).
Table 3AUC, sensitivity, and specificity of different peaks for differentiation between normal breast tissue and malignant variants
Peaks | Normal | METC | AUC | p-value | Sensitivity% | Specificity% | Normal | DCIS | AUC | p-value | Sensitivity% | Specificity% | Normal | IDC | AUC | p-value | Sensitivity% | Specificity% |
---|
1,045 | 0.0433 | 0.0611 | 0.267 | 0.24 | 100 | 10 | 0.0433 | 0.051 | 0.36 | 0.84 | 100 | 15 | 0.0433 | 0.064 | 0.16 | 0.06 | 66 | 7 |
1,073 | 0.0464 | 0.0597 | 0.267 | 0.24 | 100 | 10 | 0.0464 | 0.052 | 0.46 | 0.84 | 100 | 23 | 0.0464 | 0.063 | 0.24 | 0.14 | 66 | 7 |
1,080 | 0.0642 | 0.0607 | 0.333 | 0.17 | 100 | 10 | 0.0642 | 0.052 | 0.46 | 0.84 | 100 | 23 | 0.0642 | 0.063 | 0.24 | 0.14 | 66 | 7 |
1,155 | 0.04 | 0.047 | 0.367 | 0.5 | 100 | 10 | 0.04 | 0.044 | 0.39 | 0.54 | 67 | 15 | 0.04 | 0.049 | 0.29 | 0.24 | 66 | 7 |
1,162 | 0.041 | 0.047 | 0.433 | 0.74 | 100 | 20 | 0.041 | 0.045 | 0.41 | 0.64 | 100 | 15 | 0.041 | 0.048 | 0.31 | 0.26 | 67 | 10 |
1,230 | 0.0641 | 0.0605 | 0.567 | 0.89 | 100 | 40 | 0.0641 | 0.059 | 0.62 | 0.55 | 100 | 40 | 0.0641 | 0.062 | 0.52 | 0.93 | 100 | 29 |
1,237 | 0.065 | 0.06 | 0.567 | 0.74 | 100 | 40 | 0.065 | 0.059 | 0.64 | 0.46 | 100 | 46 | 0.065 | 0.062 | 0.57 | 0.69 | 100 | 35 |
1,274 | 0.051 | 0.05 | 0.5 | 1 | 100 | 30 | 0.051 | 0.049 | 0.54 | 0.84 | 100 | 30 | 0.051 | 0.051 | 0.47 | 0.88 | 100 | 26 |
1,304 | 0.048 | 0.047 | 0.467 | 0.87 | 100 | 30 | 0.048 | 0.047 | 0.44 | 0.74 | 100 | 15 | 0.048 | 0.053 | 0.34 | 0.38 | 67 | 16 |
1,386 | 0.059 | 0.06 | 0.467 | 0.87 | 100 | 30 | 0.059 | 0.057 | 0.51 | 0.95 | 100 | 30 | 0.059 | 0.063 | 0.39 | 0.52 | 67 | 20 |
1,394 | 0.06 | 0.06 | 0.5 | 1 | 100 | 30 | 0.06 | 0.057 | 0.54 | 0.84 | 100 | 30 | 0.06 | 0.062 | 0.44 | 0.74 | 100 | 23 |
1,446 | 0.0652 | 0.0585 | 0.6 | 0.61 | 100 | 40 | 0.0652 | 0.059 | 0.62 | 0.55 | 100 | 46 | 0.0652 | 0.06 | 0.6 | 0.61 | 100 | 45 |
1,453 | 0.583 | 0.596 | 0.633 | 0.5 | 100 | 50 | 0.583 | 0.058 | 0.62 | 0.55 | 100 | 46 | 0.583 | 0.059 | 0.61 | 0.56 | 100 | 45 |
1,528 | 0.0947 | 0.0802 | 0.633 | 0.5 | 100 | 50 | 0.0947 | 0.082 | 0.62 | 0.55 | 100 | 38 | 0.0947 | 0.082 | 0.65 | 0.41 | 100 | 48 |
1,535 | 0.0955 | 0.0789 | 0.733 | 0.24 | 100 | 60 | 0.0955 | 0.082 | 0.62 | 0.55 | 100 | 46 | 0.0955 | 0.082 | 0.65 | 0.41 | 100 | 45 |
1,543 | 0.0951 | 0.078 | 0.767 | 0.18 | 100 | 70 | 0.0951 | 0.08 | 0.67 | 0.38 | 100 | 46 | 0.0951 | 0.079 | 0.7 | 0.26 | 100 | 55 |
1,632 | 0.105 | 0.0879 | 0.733 | 0.24 | 100 | 60 | 0.105 | 0.09 | 0.64 | 0.46 | 100 | 46 | 0.105 | 0.088 | 0.68 | 0.35 | 100 | 52 |
1,744 | 0.0029 | 0.0105 | 0.133 | 0.06 | 0 | 60 | 0.0029 | 0.006 | 0.26 | 0.2 | 33 | 70 | 0.0029 | 0.011 | 0.16 | 0.06 | 33 | 42 |
2,922 | 0.0459 | 0.0463 | 0.533 | 0.87 | 100 | 40 | 0.0459 | 0.042 | 0.67 | 0.38 | 100 | 54 | 0.0459 | 0.047 | 0.54 | 0.83 | 100 | 45 |
3,280 | 0.0804 | 0.0512 | 0.93 | 0.028 | 100 | 80 | 0.0804 | 0.053 | 0.95 | 0.02 | 100 | 85 | 0.0804 | 0.049 | 0.96 | 0.01 | 100 | 83 |
Table 4AUC, sensitivity, and specificity of different peaks for differentiation among malignant variants
Peaks | DCIS | METC | AUC | p-value | Sensitivity % | Specificity % | DCIS | IDC | AUC | p-value | Sensitivity % | Specificity% | METC | IDC | AUC | p-value | Sensitivity% | Specificity% |
---|
1,045 | 0.051 | 0.0611 | 0.65 | 0.22 | 70 | 46 | 0.051 | 0.064 | 0.31 | 0.05 | 62 | 23 | 0.0611 | 0.064 | 0.46 | 0.69 | 60 | 42 |
1,073 | 0.052 | 0.0597 | 0.65 | 0.24 | 70 | 46 | 0.052 | 0.063 | 0.35 | 0.11 | 62 | 26 | 0.0597 | 0.063 | 0.47 | 0.79 | 60 | 48 |
1,080 | 0.052 | 0.0607 | 0.62 | 0.32 | 70 | 46 | 0.052 | 0.063 | 0.36 | 0.13 | 62 | 39 | 0.0607 | 0.063 | 0.46 | 0.72 | 60 | 45 |
1,155 | 0.044 | 0.047 | 0.59 | 0.46 | 70 | 38 | 0.044 | 0.049 | 0.41 | 0.39 | 62 | 45 | 0.047 | 0.049 | 0.5 | 0.98 | 60 | 58 |
1,162 | 0.045 | 0.047 | 0.55 | 0.66 | 70 | 38 | 0.045 | 0.048 | 0.42 | 0.44 | 62 | 42 | 0.047 | 0.048 | 0.49 | 0.9 | 60 | 52 |
1,230 | 0.059 | 0.0605 | 0.5 | 1 | 70 | 23 | 0.059 | 0.062 | 0.44 | 0.54 | 69 | 29 | 0.0605 | 0.062 | 0.47 | 0.74 | 60 | 50 |
1,237 | 0.059 | 0.06 | 0.5 | 1 | 70 | 23 | 0.059 | 0.062 | 0.45 | 0.58 | 69 | 26 | 0.06 | 0.062 | 0.47 | 0.74 | 50 | 55 |
1,274 | 0.049 | 0.05 | 0.54 | 0.76 | 70 | 46 | 0.049 | 0.051 | 0.42 | 0.45 | 62 | 39 | 0.05 | 0.051 | 0.48 | 0.86 | 70 | 39 |
1,304 | 0.047 | 0.047 | 0.62 | 0.35 | 70 | 54 | 0.047 | 0.053 | 0.38 | 0.21 | 62 | 32 | 0.047 | 0.053 | 0.5 | 0.97 | 70 | 42 |
1,386 | 0.057 | 0.06 | 0.58 | 0.54 | 70 | 56 | 0.057 | 0.063 | 0.4 | 0.3 | 62 | 32 | 0.06 | 0.063 | 0.47 | 0.81 | 70 | 40 |
1,394 | 0.057 | 0.06 | 0.55 | 0.66 | 70 | 46 | 0.057 | 0.062 | 0.4 | 0.33 | 62 | 32 | 0.06 | 0.062 | 0.48 | 0.88 | 70 | 48 |
1,446 | 0.059 | 0.0585 | 0.48 | 0.85 | 70 | 30 | 0.059 | 0.06 | 0.47 | 0.77 | 60 | 45 | 0.0585 | 0.06 | 0.46 | 0.72 | 60 | 45 |
1,453 | 0.058 | 0.596 | 0.47 | 0.8 | 70 | 31 | 0.058 | 0.059 | 0.47 | 0.75 | 62 | 45 | 0.596 | 0.059 | 0.46 | 0.72 | 60 | 42 |
1,528 | 0.082 | 0.0802 | 0.45 | 0.66 | 70 | 31 | 0.082 | 0.082 | 0.47 | 0.77 | 62 | 48 | 0.0802 | 0.082 | 0.47 | 0.79 | 70 | 39 |
1,535 | 0.082 | 0.0789 | 0.45 | 0.66 | 70 | 31 | 0.082 | 0.082 | 0.48 | 0.85 | 62 | 45 | 0.0789 | 0.082 | 0.46 | 0.69 | 70 | 39 |
1,543 | 0.08 | 0.078 | 0.45 | 0.71 | 70 | 31 | 0.08 | 0.079 | 0.49 | 0.9 | 70 | 35 | 0.078 | 0.079 | 0.46 | 0.74 | 70 | 39 |
1,632 | 0.09 | 0.0879 | 0.45 | 0.66 | 70 | 31 | 0.09 | 0.088 | 0.52 | 0.83 | 70 | 46 | 0.0879 | 0.088 | 0.48 | 0.81 | 70 | 42 |
1,744 | 0.006 | 0.0105 | 0.72 | 0.04 | 70 | 100 | 0.006 | 0.011 | 0.24 | 0.008 | 54 | 13 | 0.0105 | 0.011 | 0.49 | 0.98 | 70 | 58 |
2,922 | 0.042 | 0.0463 | 0.54 | 0.76 | 70 | 31 | 0.042 | 0.047 | 0.41 | 0.36 | 70 | 35 | 0.0463 | 0.047 | 0.47 | 0.79 | 60 | 48 |
3,280 | 0.053 | 0.0512 | 0.48 | 0.85 | 70 | 31 | 0.053 | 0.049 | 0.58 | 0.45 | 70 | 60 | 0.0512 | 0.049 | 0.54 | 0.72 | 50 | 45 |
Cut-off points and clustering patterns
The identified cut-off points were congruent with hierarchical clustering patterns, facilitating the identification of peaks with similar vibrational characteristics and chemical properties. For instance, peaks at 1,043, 1,073, and 1,080 cm−1 shared similar cut-off values (0.031, 0.032, and 0.032, respectively) between normal and carcinoma cases and among carcinoma categories, suggesting potential chemical similarity in their molecular structures, as shown in Table 5.
Table 5Discriminating peaks, cut-off points, and chemical assignments
Discriminating peaks | Average cut-off normal-carcinomas | Average cut-off carcinoma groups | Chemical assignment |
---|
1,043 | 0.031 | 0.053 | Phosphate (asym)/Glycogen |
1,073 | 0.032 | 0.055 | |
1,080 | 0.032 | 0.055 | |
1,155 | 0.025 | 0.046 | Protein (amino acids) |
1,162 | 0.031 | 0.045 | |
1,230 | 0.052 | 0.053 | Phosphate (asym) |
1,237 | 0.053 | 0.053 | |
1,274 | 0.038 | 0.046 | Protein (Collagen, Amide III) |
1,304 | 0.037 | 0.047 | |
1,386 | 0.046 | 0.055 | Lipids, Protein |
1,394 | 0.046 | 0.055 | |
1,446 | 0.053 | 0.053 | Protein (methyl group) asym |
1,453 | 0.055 | 0.052 | |
1,528 | 0.077 | 0.074 | Protein (Amide II) |
1,535 | 0.079 | 0.073 | |
1,543 | 0.079 | 0.069 | |
Peak ratio analysis
With Bonferroni correction, comparing breast carcinomas to normal breast tissue showed that the peak ratio (A1632/A1543), corresponding to protein levels associated with beta-sheet structures, increased in malignancy but lacked statistical significance for discrimination. However, phosphate (A1237/A1080) and glycogen (A1043/A1543) levels demonstrated statistically significant increases in carcinoma compared to normal (p < 0.01), though these could not be validated as diagnostic markers. Additionally, the nucleocytoplasmic index peak ratio was significantly elevated (p < 0.05). However, ROC curve analysis for these peak ratios (A1632/A1543, A1237/A1080, and A1043/A1543) revealed sub-par AUC values and abysmal specificity (less than 1%), indicating limited diagnostic relevance despite statistical significance for differentiating normal tissue from breast lesions and among malignancies. Notably, the A1080/A1632 ratio achieved statistical significance (p = 0.03) with an AUC of 1.0, sensitivity (∼100%), and specificity (∼100%) in differentiating normal from carcinomas and DCIS from invasive breast carcinomas (p < 0.001), underscoring its potential diagnostic value as shown in Table 6 and Figure S3. However, glycogen-related peaks yielded AUC values of 0.6–0.7 with ∼80% sensitivity and ∼40% specificity, suggesting only cautious consideration for delineating invasive ductal and metaplastic carcinomas (Fig. S4).
Table 6Biomarker mean differences and diagnostic performance between paired breast tissue types
Biomarkers | Breasts | | t-test | | AUC | | Sensitivity (%) | Specificity (%) | Cut-off |
---|
| Normal: Mean ± SEM | DCIS: Mean ± SEM | | p-value | | p-value | | | |
NC | 0.44 ± 0.09 | 0.56 ± 0.07 | 3.11 | 0.008 | 0.97 | 0.013* | 100 | 92 | 2.10 |
Phos | 0.71 ± 0.04 | 0.88 ± 0.04 | 1.90 | 0.08 | 0.08 | 0.03 | 100 | 0 | 0.03 |
Glyc | 0.45 ± 0.03 | 0.64 ± 0.04 | −2.20 | 0.05 | 0.03 | 0.013* | 100 | 0 | 0.001 |
| Normal: Mean ± SEM | METC: Mean ± SEM | | | | | | | |
NC | 0.44 ± 0.09 | 0.67 ± 0.05 | 7.14 | 0.03 | 1.00 | 0.01* | 100 | 100 | 0.19 |
Phos | 0.71 ± 0.04 | 1.01 ± 0.03 | 5.04 | 0.001 | 0.00 | 0.01 | 100 | 0 | 0.00 |
Glyc | 0.45 ± 0.03 | 0.78 ± 0.04 | 4.92 | 0.001 | 0.00 | 0.01 | 100 | 0 | 0.00 |
| Normal: Mean ± SEM | IDC: Mean ± SEM | | | | | | | |
NC | 0.44 ± 0.09 | 0.71 ± 0.05 | 5.68 | 0.0001 | 0.98 | 0.03* | 100 | 94 | 1.68 |
Phos | 0.71 ± 0.04 | 1.03 ± 0.02 | −5.05 | 0.0001 | 0.000 | 0.005 | 100 | 0 | −0.34 |
Glyc | 0.45 ± 0.03 | 0.83 ± 0.03 | −4.85 | 0.0001 | 0.011 | 0.02 | 100 | 0 | −0.58 |
| DCIS: Mean ± SEM | METC: Mean ± SEM | | | | | | | |
Pro | 1.13 ± 0.01 | 1.13 ± 0.01 | −0.26 | 0.80 | 0.45 | 0.71 | 54 | 50 | 1.12 |
NC | 0.56 ± 0.07 | 0.67 ± 0.05 | 2.98 | 0.007 | 0.83 | 0.008* | 85 | 80 | 1.53 |
Pho | 0.88 ± 0.04 | 1.00 ± 0.03 | −2.27 | 0.03 | 0.23 | 0.03 | 92 | 0 | 0.73 |
Gly | 0.64 ± 0.04 | 0.78 ± 0.04 | −2.62 | 0.02 | 0.22 | 0.02 | 92 | 0 | 0.49 |
| DCIS: Mean ± SEM | IDC: Mean ± SEM | | | | | | | |
Pro | 1.13 ± 0.01 | 1.14 ± 0.02 | −0.34 | 0.74 | 0.49 | 0.94 | 100 | 10 | 1.07: |
NC | 0.56 ± 0.07 | 0.71 ± 0.05 | 0.06 | 0.0001 | 0.86 | 0.0001* | 100 | 49 | 1.31 |
Pho | 0.88 ± 0.04 | 1.03 ± 0.02 | −3.58 | 0.001 | 0.21 | 0.003 | 100 | 0 | −0.28 |
Gly | 0.64 ± 0.04 | 0.83 ± 0.02 | −4.22 | 0.0001 | 0.18 | 0.001 | 92 | 5 | 0.49 |
| METC: Mean ± SEM | IDC: Mean ± SEM | | | | | | | |
Pro | 1.13 ± 0.01 | 1.14 ± 0.02 | 0.18 | 0.86 | 0.49 | 0.93 | 65 | 50 | 1.11 |
NC | 0.67 ± 0.05 | 0.71 ± 0.05 | −0.90 | 0.38 | 0.32 | 0.09 | 71 | 20 | 1.24 |
Pho | 1.00 ± 0.03 | 1.03 ± 0.02 | 0.48 | 0.63 | 0.52 | 0.88 | 68 | 20 | 0.95 |
Gly | 0.78 ± 0.04 | 0.83 ± 0.02 | 0.97 | 0.34 | 0.64 | 0.19 | 80 | 40 | 0.75 |
Discussion
Spectral analysis reveals distinct vibrational signatures differentiating normal breast tissue from carcinomatous counterparts, with normal tissue exhibiting a shift toward higher wavenumbers, suggesting diagnostic utility. Specifically, carcinoma samples displayed increased peak intensities at 1,045 cm−1, 1,075 cm−1, 1,080 cm−1, 1,151 cm−1, 1,162 cm−1, 1,274 cm−1, 1,237 cm−1, 1,308 cm−1, 1,386 cm−1, 1,394 cm−1, and 1,453 cm−1, while normal tissue showed enhanced intensities at 1,230 cm−1, 1,446 cm−1, 1,525 cm−1, 1,535 cm−1, 1,543 cm−1, 2,919 cm−1, 3,273 cm−1, and 3,284 cm−1.12,13,19
Normal tissue exhibited broader, more intense peaks at 3,273–3,284 cm−1 (O–H and Amide A stretching vibrations),19,24,25 while peaks at 2,916–2,919 cm−1 (CH2 asymmetric stretch of acyl lipids) in normal tissue showed a narrower band with lower intensity compared to carcinomas as shown in Table 2 This underscores the role of saturated lipids in membrane integrity and suggests carcinoma’s potential for progression and metastasis.20,26,41–43
Conversely, higher expression of the lipid peak at 1,744 cm−1 (C=O stretching vibrations from esters and phospholipids) was observed in malignant tissue,19,28,41,42 positioning it as a key differential marker for metaplastic versus DCIS. This increased expression aligns with reduced hydrogen bonding, potentially augmenting de novo lipogenesis in carcinoma, indicating metastatic potential, and highlighting a redox environment in metaplastic carcinoma similar to aggressively proliferating ductal carcinoma.28–30,39,42
Normal tissue displayed sharper, more intense features at 1,627–1,632 cm−1 (β-sheet Amide I stretching vibrations) and 1,529 (β-sheet Amide II) and 1,543 cm−1 (α-sheet Amide II),13–17,33,35–36,44–47 suggesting β-sheet proteins as markers distinguishing normal from breast carcinoma,36 despite conflicting evidence regarding β-sheet protein reductions in metaplastic carcinoma. Protein peaks were fairly conserved across all malignant breast tissues.10,16,32,42,46
Normal breast tissue exhibited a more pronounced peak at 1,446 cm−1 (CH2 bending of lipids, triglycerides, and C–H vibrations),21,48 supporting observations for unsaturated lipid bands and emphasizing membrane fluidity for cellular health. In contrast, the peak at 1,453 cm−1 (asymmetric CH3 vibrations and protein deformation) showed less distinct characteristics.12,26,29,42
Metaplastic carcinoma showed increased intensities in peaks at 1,386 cm−1 and 1,394 cm−1. Peaks at 1,394 cm−1 is identified with fatty acids due to COO- symmetric stretching of amino acids side bonds vibration and 1,308 cm−1 involving protein symmetric CH3 bending assigned to collagen/Amide III vibrations,28,35 highlighting fatty acids and amino acid involvement in neoplastic transformation.29,30,33,35,36 Increased collagen deposition in carcinoma tissues suggests heightened tumor aggressiveness.24,25,32,36,49,50
Peaks at 1,230 cm−1 and 1,232 cm−1 (nucleic acid contributions) showed sharper features in normal tissue,11,32,33,34–38,51 aligning with literature indicating differential intensities in carcinoma tissues due to genetic mutations and nuclear enlargement, particularly in metaplastic and ductal carcinomas, which exhibited increased intensity compared with normal breasts. Elevated peaks at 1,308 cm−1 and 1,274 cm−1 assigned to collagen- a matrix tissue due to metalloproteinases- were found prominently in carcinomas,30,32,35–38,51 suggesting fibroblast activation and heightened collagen deposition in metastatic conditions.16,31,32,41,50
Peak 1,151 cm−1 (C–O glycogen stretches) was more prevalent in malignant samples,33,36,38,52 especially in invasive ductal carcinoma, followed by metaplastic carcinoma, pre-IDC, and normal breast tissue.29 This is corroborated by other carbohydrate peaks, reflecting escalated cell activity in the G1 phase and heightened energy needs during DNA synthesis,7,15,34,37,39,40 although interpretations are rare in the literature, especially concerning metaplastic carcinomas.
Peak 1,080 cm−1 (nucleic acids) indicated shifts favoring normal tissue,28,34,35,51 paralleling documentation of asymmetric phosphate vibrations linked to malignancy markers.11,12,19,36,37,40 Meanwhile, peak 1,162 cm−1 (collagen mechanics) showed intensified patterns in metaplastic as well as ductal carcinomas versus normal breast tissue.
Comparative evaluation of peak ratios revealed that nucleic acid and glycogen levels were statistically elevated in metaplastic carcinoma versus normal breast tissue, consistent with studies validating nucleic acid increases in carcinoma development.38–40,47 Elevated glycogen levels in other carcinomatous tissues are similarly documented,36,52 although some studies suggest glycogen is relatively lower in normal tissues.12,16,35 Elevated glycogen levels in metaplastic tissues could suggest rapid cell cycle shifts.40,44 The nucleocytoplasmic index ratio also proved significantly relevant (p = 0.03) for discriminating between normal and all metaplastic carcinomas as well as between ductal carcinoma in situ and invasive breast carcinomas, reflecting carcinoma’s need to compensate for its metabolic demands through excessive nuclear growth relative to the cytoplasm.34,53
Hierarchical clustering, one of the supervised learning models,19,29,47,54–58 coupled with ROC curve analysis for cutoff determination, elucidates potential synergistic chemical relationships between spectral peaks and establishes diagnostic links between clustering patterns and optimal classification thresholds. This analysis revealed three distinct clusters capable of differentiating breast tissue types. Notably, hierarchical clustering associated less-utilized peaks (e.g., 1,394 and 1,386 cm−1) with established carcinoma biomarkers, suggesting a shared chemical basis despite differing vibrational modes. The observed spectral proximity, quantified by linkage and Euclidean distances, implies closer functional relationships than conventionally recognized, potentially harboring diagnostic significance. Similarly, the co-clustering of the 1,162 cm−1 peak (typically assigned to carotenoids) with other compounds suggests a more complex origin than previously considered.54,55 Furthermore, the clustering of peaks at 1,535, 1,528, 1,543, and 1,632 cm−1, all indicative of amide protein origins,19,35,44,46 underscores their inherent similarities and highlights underlying chemical associations potentially relevant to metaplastic carcinogenesis.31,33 Differential gene expression resulting in specific protein conformations likely contributes to the discriminatory power of these protein-related spectral features.34,35,46
Figure 3a–f delineate spectral clusters, pinpointing vibrational modes and enabling chemical assignments via spatial relationships. Phosphodiester vibrational peaks demonstrate a strong correlation, whereas Amide A and lipid peaks show marked divergence from other clusters. Spectral peak clustering analysis reveals disease-associated shifts in biochemical composition. Despite variability in peak patterns across matched breast carcinoma cases, the fundamental chemical identities and spectral signatures consistently differentiate breast carcinoma subtypes. Notably, initially dissimilar peaks converged within shared clusters upon refined discrimination. Moreover, protein peaks exhibit autonomous vibrational characteristics, making them robust markers for investigating protein conformational changes,25,32,36,45,47 especially during the progression from DCIS to metaplastic carcinoma, and from DCIS to IDC.
ROC curve analysis revealed similar chemical properties across spectral peaks, suggesting their potential to differentiate normal, in situ, and invasive breast tissues. However, some peaks exhibited preferential elevation in specific carcinoma subtypes, indicating possible utility in subtyping. Despite achieving 100% sensitivity on average, low specificity and AUC values below 0.5 indicate limited diagnostic accuracy for most peaks in this study. Contrary to prior reports highlighting carbohydrates, their diagnostic relevance was diminished in our findings. Nucleic acid/phosphate peaks showed inconsistent diagnostic potential, likely reflecting tumor heterogeneity. Amide II peaks associated with β-sheet proteins and Amide I peaks demonstrated considerable diagnostic promise (AUC ≈ 0.7, sensitivity ≈ 100%).14–17,35–38 The significant elevation of β-sheet proteins in normal tissue compared to metaplastic samples suggests their potential as diagnostic markers, although prior conflicting observations highlight complex metabolic influences. The peak ratio corresponding to the nucleocytoplasmic ratio (AUC = 1.0, 100% sensitivity and 100% specificity) compares well with ratios used in previous investigations, differentiating normal breast from tumorous phenotypes.14,21,29,57
Furthermore, peak 3,280 stood out for its high discriminatory power (AUC = 0.93), challenging assumptions about its limited value due to water interference,29,35,56,57 and displaying the ability to distinguish normal from histological types and monitor treatment responses.13,14,29,53,58 The increased levels of Amide A alongside other amide bands suggest a possible connection to oncogene-driven dysregulation, warranting further investigation.24,33,39 Ongoing exploration of these spectral features is crucial for improving carcinoma diagnostics, with a focus on enhancing specificity while maintaining high sensitivity. Discrepancies with previous studies reporting exceptional diagnostic performance for certain peaks may stem from lipid peroxidation.42,54 In contrast, peak 1,744’s limited diagnostic power (AUC = 0.133) may have resulted from lipid loss during tissue processing.26,42,48,53
FTIR spectroscopy offers a powerful diagnostic alternative for breast tissue analysis, showing comparable efficacy to diffuse optical spectroscopy, intrinsic fluorescence spectroscopy, and diffuse reflectance spectroscopy. While diffuse optical spectroscopy, intrinsic fluorescence spectroscopy, and diffuse reflectance spectroscopy effectively quantify key biochemical markers—oxyhemoglobin, β-carotene, methemoglobin, tissue hemoglobin, collagen, lipids, and redox state —with exceptional diagnostic accuracy,49,50,54,55 FTIR, particularly through analysis of the Amide A band, achieves similarly high levels of sensitivity and specificity in differentiating breast tissue types.29,56,57 This confirms existing research and solidifies FTIR as a valuable tool in breast tissue characterization.10,19,22,29,39,40,46,53,57,58
Despite the value in investigating the unique characteristics of rare carcinomas, this study acknowledges its limitations. The small sample size, dictated by the carcinoma’s infrequent occurrence during the study period, limits the broad applicability of the conclusions. Furthermore, the absence of comprehensive patient biodata, details regarding presentation patterns (days/year), and follow-up protocols prevented potentially valuable survival analysis. Future research should prioritize the inclusion of such data. Finally, the exclusion of other breast lesion subtypes beyond ductal carcinomas and normal tissue hinders a complete understanding of the specific biochemical and molecular differences defining metaplastic carcinoma within the spectrum of breast malignancies.