Samples were acquired from the Zixi mountain nature preservation area, Baize Horticulture Co., Ltd., and Xinlu Horticulture Co., Ltd. in Chuxiong City, Yunnan, China. They include thirty-nine varieties of C. reticulata Lindl., one variety of C. japonica L., one variety of C. oleifera Abel, and one variety of C. sasanqua Thunb., as shown in Table 2. Among them, twenty varieties belong to the Chuxiong population of C. reticulata Lindl., and two varieties belong to the Dali population of C. reticulata Lindl., and six varieties belong to the Tengchong population of C. reticulata Lindl., and four are traditional species that grow in Yunnan Province.

DNA was extracted using the improved SDS-CTAB method. Fresh leaves (0.15 g) were taken and fully ground. The leaves were mixed with 750 μ L DNA extraction buffer (200 mmol· L^-1 Tris-HCl, 100 mmol· L^-1 EDTA, 12.6 mmol· L^-1 NaCl, and 10% CTAB 2% solution [100 mmol· L^-1 Tris-HCl, 20 mmol· L^-1 EDTA, 1.4 mol· L^-1 NaCl]), 300 μ L 10% SDS, and 100 μ L ethanol. The mixture was incubated for 30 minutes in a 65 ℃ water bath. The supernatant was removed, then an equivalent volume of phenol-saturated water was added and the mixture was shaken for 5 minutes. The supernatant was removed again, and replaced with an equivalent volume of chloroform-isoamyl alcohol (24: 1 V/V chloroform: isoamyl alcohol). The phenol and chloroform steps were repeated three times to denature and remove all proteins. Isopropyl alcohol was mixed with the extract and stored at 4 ℃ overnight to precipitate DNA. To remove RNA, 3 μ L RNase A enzyme (10 mg· mL^-1) was added. Finally, the precipitate was washed twice with 75% ethanol to obtain a clean DNA sample.

The quality and purity of the DNA precipitate were confirmed by determination of the 260/280 nm absorption ratio using a DMS 200 spectrophotometer (Varian Techtron Pty. Ltd., Australia). The 260/280 ratio of DNA sample was 1.83, which confirmed that the precipitate is composed of genomic DNA. Genomic DNA solutions with a concentration of 20~50 ng· L^-1 were chosen and lyophilized using a rotary SpeedVac (ALPHA1-2LD, Christ Co., Germany) for 3 h.

One hundred and sixty-four genomic DNA samples from forty-one varieties were acquired and scanned by FTIR spectrometer.

FTIR spectra from all the one hundred and sixty-four genomic DNA samples were obtained using an FTIR spectrometer (Spectrum 100, PerkinElmer Co., and Britain). Samples were prepared for scanning by blending and grinding 0.5 mg DNA with 150 mg KBr, then compressing the resulting pellet into disks. Scans were repeated 16 times in the scanning range of 4 000~400 cm^-1 at a resolution of 4 cm^-1. During detection, atmospheric absorption correction was employed and consistent force was applied to the sample.

The FTIR data of DNA were pretreated using smoothing, normalizing, and the first derivative method. Forty principal components were extracted and used to cluster and identified by using hierarchical cluster. The data were managed with Origin 8.0, MATLAB R2010a, and PASW 18.0 software Origin 8.0 was used to plot the spectrum, and MATLAB R2010a was used to extract the principal components of each spectrum, and PASW 18.0 was used to cluster and analyze anova.

In this paper, we used principal component analysis (PCA) and hierarchical clustering analysis (HCA). In this research, a covariance matrix was used to extract the principal components, and regression was used to calculate a factor score. Ward’ s method was applied to the hierarchical clustering analysis. The clustering distance limits were based on the squared Euclidean distances.

DNA average FTIR spectra from forty-one Camellia varieties were acquired; these are shown in Figure 1. Each spectrum stands for one cluster category respectively. The result shows that there are three outstanding absorbance peaks. The first peak is in the regions of 1 639~1 559 cm^-1. The second peak is at 1 410 cm^-1. The third peak is in the regions of 1 081~1 050 cm^-1. There are some peaks at 1 325, 1 258, 988, 928, 837, and 774 cm^-1 (Table 1). The leading distinctions between varieties are in the regions 1 410~1 075, and 1 400~900 cm^-1. Some varieties do not have an obvious peak in the above two regions; some have one peak; others have two peaks. Some peaks are obvious in some varieties, whereas others are not so obvious.

Fig.1

	Figure Option ViewDownloadNew Window
	Fig.1 DNA average FTIR spectra of nine typical Camellia varieties 1: Dalicha; 2: ZiPao; 3: Yunzhen; 4: Hentiangao; 5: Songziling; 6: Luchengchun; 7: Ziqiang; 8: Tongzimian; 9: Jiaoyuan

Fig.2

	Figure Option ViewDownloadNew Window
	Fig.2 First derivative average spectra of nine typical Camellia DNA FTIR 1: Dalicha; 2: ZiPao; 3: Yunzhen; 4: Hentiangao; 5: Songziling; 6: Luchengchun; 7: Ziqiang; 8: Tongzimian; 9: Jiaoyuan

The first derivative confirms the accuracy of the peaks’ wavenumbers and emphasizes distinctions among the categories (Figure 2).

In the regions of 1 000~800 cm^-1, IR peaks reflect sugar conformation. The peaks in the 1 250~1 000 cm^-1 regions are assigned to vibrations along the sugar-phosphate backbone of the DNA molecule. Peaks in the 1 500~1 250 cm^-1 range reflect base-sugar vibrations. The 1 760~1 480 cm^-1 regions are sensitive to base-pairing and base-stacking effects (Table 1).

Table 1

Table 1

Table 1 Wavenumber and assignations of Camellia DNA FTIR in the regions of 1 800~700 cm-1 Wavenumber /cm-1 Assignment 1 665 C2=O2 Str. Cytosine[16, 18] 1 647 C4=O4 Str. thymine[16] 1 634 N— H guanine[15] 1 616 C4— C5=C6 Str. Cytosine[15, 18, 19] 1 599 C4=C5, C5— C6 Str. guanine[16, 18] 1 559 C6=O6, C5— C6, C4=C5 Str. Guanine[15, 16, 18, 19] 1 395 cytosine ring[15, 16, 18, 19] 1 340 N— H Thymine[15, 16, 18, 19] 1 248 PO2 asym. Str. (Phosphate Ⅰ )[19] 1 110 P— O— C Sym., str.[15, 16, 18, 19] 1 060 C— O Str. Deoxyribose[15, 16, 18] 1 050 C— O Str. Deoxyribose[15, 16, 18, 19] 857 N-type marker[18] 831 S-type marker[18]

Table 1 Wavenumber and assignations of Camellia DNA FTIR in the regions of 1 800~700 cm-1

Pretreatment of the spectra eata is necessary due to random noise, baseline drift, concentration differences, asymmetry of sample and environment, etc. Pretreatment can effectively get!rid of artifacts. Smoothing, normalizing, plotting the derivative, and selecting spectra range are common pretreatment methods. In this research, FFT Filter smoothing was used to remove random noise. First derivatives were used to clear up influence from instrument background and baseline drift. The 1 800~700 cm^-1 wavenumber range was selected to simplify the mathematics. Figure 3(a) shows the mean, standard deviation, standard error, minimum, and maximum of the FTIR data for each genomic DNA sample from the forty-one varieties of Camellia after pretreatment.

Anova shows that the mean difference is significant between groups (p< 0.05). LSD multiple comparisons reveal further details: the difference is significant when Variety 3, 5, 7, 9, 10, 16, and 28 are compared with Variety 20 or Variety 21 (p< 0.05). The difference is not significant among other varieties.

Fig.3

	Figure Option ViewDownloadNew Window
	Fig.3 (a) The standard Error of forty-one varieties after pretreatment, (b) 3D scatter plot of the principal component 1, principal component 2 and principal component 3 Anova significance analysis is necessary. It can effectively display data’ s characteristic, and provides scientific helps and foresees for further analysis.

The correlation matrix was used to extract the principal components, and regression was used to determine the factor score coefficient matrix for PCA.

The results show that there is an obvious turning point at the fortieth component, so forty principal components were extracted, and used to further cluster and identify the forty-one varieties. The cumulative percentage of the first forty principal components is 100%.

A 3D scatter plot of PC1, PC2, and PC3 reveals that the forty-one Camellia varieties separate well (Figure 3. b); the points do not cluster well. A 3D scatter plot from the principal components that were significantly different in ANOVA shows that center points are more concentrated, but there is still no informative cluster effect. This indicated that clustering these forty-one Camellia varieties is complicated, and cannot be accomplished directly by PCA. Further analysis was needed.

2.4.1 Standard hierarchical cluster model

One hundred and sixty-four genomic DNA samples from forty-one Camellia varieties (four DNA samples from each variety) were extracted and scanned by FTIR. Three genomic DNA samples were used to establish average spectra for clustering analysis, and one sample was used for identification analysis.

Fig.4

	Figure Option ViewDownloadNew Window
	Fig.4 Standard hierarchical cluster model of forty-one Camellia varieties based on DNA FTIR

The standard hierarchical cluster model was established by using forty extracted principal components. Ward’ s method was applied, and the metrics are based on squared Euclidean distance. The cluster results show that these forty-one Camellia varieties group into nine categories at 1.0 cluster distance (Figure 4). The first category is composed of the varieties Shizitou, Dalicha, Hong Lu Zhen, Chuxiong, Guomei, and Xiangguocha. The second category is composed of Zidie, Ziwei (Mt. Zixi), Yanzhi, ZiPao, Chudie, Chuxionghong, Zidai, Honghuayoucha, Ziyun, Zuihongyan, and Chamei. The third category is composed of Jiejiegao, Zhushazipao, Yunzhen, MuDanKui, Fentianjiao, Lifang and Yunzhen. The fourth category is composed of Seben, Hentiangao, Zhixia, Zhaoqinggongzhu, Xuejiao and Zibao. The fifth category is composed of Songziling and Donglin. The sixth category is composed of Zehe and Luchengchun. The seventh category is composed of Ziqiang alone. The eighth category is composed of Hedinghong, Linfeng, Tongzimian, Zhinan and Shisangzhichun. The ninth category contains only Jiaoyuan.

Table 2

Table 2

Table 2 The result of clustering and identified of forty-one Camellia samples by Hierarchical Cluster Category Variety Species Cluster result Clustering Right or False Right Rate/% Identifying Right or False Right Rate/% 1 Sizitou C. reticulata L. Right 92.68 Right 100 Dalicha C. reticulata L. Right Right Hongluzhen C. japonica L. False Right Chuxiong C. reticulata L. Right Right Guome C. reticulata L. Right Right Xiangguo C. reticulata L. Right Right 2 Zidie C. reticulata L. Right Right Ziwei C. reticulata L. Right Right Yanzhi C. reticulata L. Right Right ZiPao C. reticulata L. Right Right Chudie C. reticulata L. Right Right Chuxionghong C. reticulata L. Right Right Honghuayoucha C. ofeifera Abel False Right Zidai C. reticulata L. Right Right Ziyun C. reticulata L. Right Right Zuihongyan C. reticulata L. Right Right Chamei C. sasanqua Thunb. False Right 3 Jiejiegao C. reticulata L. Right Right Zhushazipao C. reticulata L. Right Right Yunzhen C. reticulata L. Right Right MuDanKui C. reticulata L. Right Right Fentianjiao C. reticulata L. Right Right Lifang C. reticulata L. Right Right Yunfen C. reticulata L. Right Right 4 Seben C. reticulata L. Right Right Hentiangao C. reticulata L. Right Right Zhixia C. reticulata L. Right Right Zhaoqinggongzhu C. reticulata L. Right Right Xuejiao C. reticulata L. Right Right Zibao C. reticulata L. Right Right 5 Songziling C. reticulata L. Right Right Donglin C. reticulata L. Right Right 6 Zehe C. reticulata L. Right Right Luchengchun C. reticulata L. Right Right 7 Ziqiang C. reticulata L. Right Right 8 Hedinghong C. reticulata L. Right Right Linfeng C. reticulata L. Right Right Tongzimian C. reticulata L. Right Right Zhinan C. reticulata L. Right Right Shisangzhichun C. reticulata L. Right Right 9 Jiaoyuan C. reticulata L. Right Right

Table 2 The result of clustering and identified of forty-one Camellia samples by Hierarchical Cluster

All forty-one varieties were clustered into three larger categories at 15.0 cluster distance. The first and the second category above belong to one. The third and the fourth category belong to another. Those from the fifth to the ninth category belong to the third. The clustering accuracy rate is 92.68% (Table 2). Three varieties that belong to the species C. japonica L., C. oleifera Abel, and C. sasanqua Thunb. should be clustered independently from C. reticulata Lindl. according to traditional classification. They were not isolated from C. reticulata Lindl., so they were thought to be clustered in error. However, it is not impossible that they are naturally associated with C. reticulata Lindl., since they are used in grafting offspring of C. reticulata Lindl. varieties.

2.4.2 Relationship analysis of forty-one Camellia varieties based on FTIR spectra data of their genomic DNA

Table 2 shows that the nearest relationships among the varieties examined are between three pairs: Shizitou and Dalicha, MuDanKui and Fentianjiao, and Yanzhi and ZiPao. Three other pairs have cluster distances less than 1.0: Jiejiegao and Zhushazipao, Guomei and Xiangguo, and Lifang and Yunfeng. The following pairs have cluster distances less than 2.0: Chudie and Chuxionghong, Shizitou and Hongluzhen, Yunzhen and Jiejiegao, Chudie and Zidai, Seben and Hentiangao, Lifang and MuDanKui, Zidie and Ziwei, Yanzhi and Chudie, and Zhixia and Seben. The cluster distances are less than 10.0 (Table 3).

Table 3

Table 3

Table 3 The cluster distance coefficients schedule of Standard Hierarchical Cluster Model of forty-one Camellia varieties of DNA FTIR Stage Cluster Combined Stage Cluster Combined Cluster 1 Cluster 2 Coefficients Cluster 1 Cluster 2 Coefficients 1 Sizitou Dalicha 0.201 21 Zhaoqinggongzhu Xuejiao 19.716 2 MuDanKui Fentianjiao 0.496 22 Lifang Yunzhen 21.859 3 Yanzhi ZiPao 0.794 23 Yanzhi Ziyun 24.235 4 Jiejiegao Zhushazipao 1.153 24 Zehe Luchengchun 26.678 5 Guomei Xiangguo 1.523 25 Zhinan Shisangzhichun 29.425 6 Lifang Yunfeng 1.926 26 Zhaoqinggongzhu Zibao 32.197 7 Chudie Chuxionghong 2.378 27 Yanzhi Zuihongyan 34.990 8 Sizitou Hongluzhen 2.846 28 Yanzhi Chamei 38.188 9 Yunzhen Jiejiegao 3.397 29 Chuxiong Guomei 41.485 10 Chudie Zidai 3.989 30 Songziling Donglin 46.077 11 Seben Hentiangao 4.784 31 Hedinghong Zhinan 50.774 12 Lifang MuDanKui 5.805 32 Zhixia Zhaoqinggongzhu 56.044 13 Zidie Ziwei 6.895 33 Chuxiong Yanzhi 62.766 14 Yanzhi Chudie 8.007 34 Zehe Ziqiang 70.035 15 Zhixia Seben 9.189 35 Jiaoyuan Donglin 78.492 16 Hedinghong Linfeng 10.430 36 Zehe Jiaoyuan 88.379 17 Yanzhi Honghuayoucha 11.947 37 Zhixia Lifang 103.507 18 Hedinghong Tongzimian 13.854 38 Songziling Zehe 119.278 19 Yanzhi Zidie 15.771 39 Songziling Zhixia 204.373 20 Chuxiong Sizitou 17.709 40 Chuxiong Songziling 346.242

Table 3 The cluster distance coefficients schedule of Standard Hierarchical Cluster Model of forty-one Camellia varieties of DNA FTIR

One of the four samples taken from each variety was used for identification analysis, and was labeled with the letter “ x” . FTIR spectra data of these “ mystery” samples were put into the standard hierarchical cluster model. When “ mystery” samples were between two adjacent varieties, the cluster distances between the sample and the two adjacent varieties were compared, and the sample was assigned to the variety with the smaller cluster distance. The hierarchical cluster plot and the cluster distances correctly identified every sample (Figure 5); the accuracy rate is 100% (Table 2).