The study was conducted at the Bipenggou valley in Lixian County, Sichuan, China (31°13′47”–31°20′17” N, 102°52′19”–102°57′41” E), which belongs to Qionglai Mountains located at the eastern margin of Hengduan Mountains—Southeastern China, as the watershed of Minjiang River and Daduhe River. Bipenggou is part of the Miaro Nature Reserve, the eastern part of Bipenggou is adjacent to Wolong National Nature Reserve, and the southern part is adjacent to Siguniangshan Scenic Area in Xiaojin County, which is also part of the “World Heritage Site of Giant Panda Corridor.” Bipenggou has a complex topography, large relative elevation, and complex climate, which is mainly influenced by three air masses from the Pacific Ocean, Indian Ocean, and Qinghai–Tibet Plateau; the climate is rich in precipitation. The altitude in this region ranges from 2,015 to 5,922 m, with a steep vertical drop that provides several vertical zones over a small area, making it an ideal location to study plant adaptation to a changing environment.

The forest type is coniferous subalpine forest in the range of 2,200–3,600 m in this region. Studies show that in Bipenggou the dominant species between 2,900 and 3,200 m are Picea asperata and Picea likiangensis, under which are deciduous trees, such as Sorbus koehneana, Betula albosinensisi, and Acer laxiflorus, and the main shrubs are Sophora davidii, Berberis veitchii, Daphne odora, Viburnum betulifolium, and Cotoneaster multiflorus; in the range of 3,100–3,300 m, the dominant species are Betula albosinensisi, Abies recurvata, Picea asperata, and Cupressus chengiana; the main shrubs are Rhododendron violaceum, Prunus plusinensis, Rosaceae sp., Rhododendron setosum, Lonicera tangutica, and Sophora davidii; in the range of 3,400–3,600 m, the dominant species are Larix masteriana, Cupressus chengiana, Abies faxoniana, and Abies squamata; the main shrubs are Rhododendron violaceum, Sorbus koehneana, Prunus plusinensis, Berberis veitchii, and Rhododendron cephalanthum. In the range of 2,800–3,600 m, consistent with our study area, the soil is brown loam with pH 5.5–7.0, with high gravel content and shallow soil layer in the area (Li et al., 2010). The study site and overview of the sampling plots are shown in Figure 1.

Lonicera L. is an important genus of the family Caprifoliaceae, which comprises about 180 species (Jachuła et al., 2019) and is a representative genus in the alpine region. Most species have labial crowns and are biflorous, with a pleasant color and a beautiful posture, making it an important genus of plants used for landscaping (Zhang et al., 2004; Liu et al., 2015), and some Lonicera species have been listed among valuable nectar- and pollen-yielding plants (Bożek, 2007). There are 99 species widespread in China (FRPS) which provide a good material for studying the diversity of the plant mating system for its rich flower color and morphological variation; however, most of the studies on Lonicera spp. are focused on the medicinal value of the genus (Shang et al., 2011; Ge et al., 2018). There are very limited studies conducted on the biological characteristics and utilization of the genus in landscaping. For this study, we selected L. nervosa (Figure 2), as it shows characteristics like bright color, luxuriant branches and leaves, beautiful shape, and strong adaptability. The corolla of L. nervosa is mauve or purplish red, and the color varies in different habitats at varying altitudes, which provide abundant material for resource utilization.

The reproduction pattern of L. nervosa is seed breeding (Wang et al., 2000), with limited preliminary cutting experiments on the breeding of L. nervosa (Feng et al., 2015); however, there is no study on the adaptability of L. nervosa. Thus, our study will be of great significance to fill the research gap on the environmental adaptability of L. nervosa and will be useful to the breeding and utilization of the species.

We set five sampling points at an interval of 200 m along the altitude from 2,950 m to around 3,650 m (2,945, 3,102, 3,326, 3,505, and 3,657 m). Depending on the distribution of L. nervosa at each point, three to five sampling sites were identified, where the Lonicera species bloom during May–June. During the late spring to the early summer of 2019, at each sampling site, we randomly selected three to five mature L. nervosa individuals at reproductive stages (June 18, 23, 25, and 29 and July 3), 10–20 individuals at each altitude point in total. From each individual, five twigs with flowers in full-bloom phase were randomly selected with no significant loss of leaves or flowers and no controlled variation in flower color or floral size within the twigs. After we cut the twigs from the branches, we quickly put them into Ziplock bags to keep them fresh for further indoor measurements, which included photography and trait measurements.

With the advancement in camera technology, the sensor of a digital camera can record highly detailed (high pixel), pure (low ISO and low image noise), and more colorful (deep color depth) images that contain more optical and color information. Meanwhile, it is easy and convenient to operate and calibrate a camera. Thus, with proper use, digital cameras serve as useful and relatively inexpensive tools to capture images and quantify color and pattern (Stevens et al., 2007). In this study, we selected a Nikon D750 DSLR camera (Nikon Corporation, Tokyo, Japan) which had a 35.9 × 24-mm CMOS sensor (6,016 × 4,016 pixels) with 14-bit color depth, 100–25,600 ISO range. It showed full regulation of exposure and metering as recommended for unbiased data acquisition (Stevens et al., 2007; White et al., 2015; Del Valle et al., 2018).

The properties of color images are device dependent, and the display color of the object will vary when images are captured at different color temperatures and camera settings (Hong et al., 2001; Stevens et al., 2007), so we put the flowers in the miniature studio after sampling and captured them with the DSLR to ensure that the color temperature and other settings were consistent. The experimental data for the definition of typical sunlight are derived from the measured sunlight, and their relative spectral power distribution is similar to the measured sunlight. Compared with other standard lighting bodies, the chromaticity points of typical sunlight are more consistent with the actual sunlight. Therefore, the International Commission on Illumination recommended that the measured sunlight should be replaced by typical sunlight D55 (5,503 K), D65 (6,504 K), and D75 (7,504 K) and, as far as possible, typical sunlight D65 (Liang et al., 2013). In terms of the photographing conditions for this study, based on actual observation, we set the white balance of the camera to 5,500 K as the color at the color temperature of 5,500 K is closest to the natural situation. With all these pre-adjusted settings, we fixed the camera on the tripod and equipped it with an NIKKOR 24–85-mm autofocus lens. Then, we manually adjusted the focal length to 35 mm which was suitable to the miniature studio, and the aperture was set to f/8 and the integration time to 1/200 s. We deliberately underexposed all photographs by 0.3 f-stop to prevent color “clipping” or saturation (Stevens et al., 2007; Del Valle et al., 2018). The position between the camera and the platform to place the flowers was fixed. When capturing, we put a flannelette on the platform before we put the flowers on it to prevent shadow and excessive brightness (Kendal et al., 2013). All images were taken in RAW format, which contained unprocessed images that may be linearized using a specialized software. We used the ImageJ software (Schneider et al., 2012) for image processing. We selected the petal center as the region of interest in each image and extracted the mean value of red–green–blue (RGB) channels which was converted into hue–saturation–lightness (HSL) value afterward.

Color parameters can fully express pigment content so that measuring plant color parameters and computing by equation can provide a fast and accurate method to estimate pigment content (Qi et al., 2019). Digital cameras have become a common tool for studying plant colors. With the help of DSLR and image processing tools, Del Valle et al. (2018) built a fast, non-invasive method to estimate anthocyanin pigment concentration in reproductive and vegetative plant tissues, which provided reliable measurement results.

Color information detected by the DSLR was in RGB color space, which was not intuitive, and its process of distinguishing color aberrations was nonlinear (Chien and Tsai, 2014), so RGB color space was not suitable for color recognition. HSL color space, however, was more intuitive and consistent with human visual characteristics (Mizunuma et al., 2014; Lin et al., 2015; Qiao et al., 2016) and is typically used in color recognition applications. In the HSL color mode, H [hue, H∈ (0°, 360°)] was defined to characterize the type of color, and the value of hue could be used to represent the color warmth—the hue value looped from 0° to 360° and defined 0° and 180° as the y axis and 90° and 270° as the x axis; the chromophore was divided into four quadrants. When the hue value was in the first and fourth quadrants, the color was considered as warm color, where 0° (360°) was the warm pole, and when the hue value was in the second and third quadrants, the color was considered cold, where 180° was the cold pole. S [color saturation, S∈ (0, 1)] described the discrimination of color purity in the case of the same hue and lightness (Weeks et al., 1995). L [color lightness, L∈ (0, 1)] spanned the full range of the selected hue from black to white in the HSL color space, so HSL is suitable for observing the color lightness.

We converted the RGB channels into HSL value; the equations are listed below (Joblove and Greenberg, 1978):

H = R blue - R red ( R max - R min ) × 60 + 120 , if ⁢ R green = R max = R red - R green ( R max - R min ) × 60 + 240 , if ⁢ R blue = R max = R green - R blue ( R m ⁢ a ⁢ x - R m ⁢ i ⁢ n ) × 60 + 360 , else ⁢ if ⁢ R green < R blue = ( R green - R blue ( R max - R min ) × 60 ) , otherwise ; L = 1 2 × R max + R min ⁢ min 255 ; S = R max - R min R max + R min , if ⁢ L ≤ 0.5 = R max - R min 2 = 255 - ( R max + R min ) , if ⁢ L > 0.5 = 0 ⁢ if ⁢ R max = R min

After capturing the images for color measurement, we quickly measured the fresh mass, and then we scanned the floral display area and leaf area with a scanner. After this, the samples were put into the oven, set at a temperature of 75°C, to dry for 48 h for the dry mass measurement (LY/T 1211-1999). The leaf and flower traits were determined separately both in fresh mass and dry mass. The trait measurements are provided in Table 1—based on the TRY plant trait database (Kattge et al., 2020).

Data was checked for the normality and homogeneity of variance before analyses. Data analyses were performed by using SPSS 18.0 (SPSS Inc., Chicago, IL, United States) and Canoco 5.0 (Ithaca, NY, United States). One-way ANOVA was performed to find the differences in floral traits along the altitudinal gradient, and Duncan’s multiple-range tests were employed to detect significance among means at p ≤ 0.05. Pearson’s correlation coefficients and redundancy analysis (RDA) were calculated to determine the relationship between floral color, floral display, leaf traits, and altitude. Floral water content (FWC) was a strong predictor to imply the variation of the ambient humidity (Roddy et al., 2019), and specific leaf area (SLA) was a common indicator for the intensity of illumination, so we substitute FWC and SLA for ambient humidity and light intensity as environmental factors in calculating the RDA. Simple linear regression was used to examine these relationships; however, sometimes the community species diversity or functional pattern change in response to environmental characteristics was not linear, and sudden changes in ecological processes may occur in the response. A stable response state was present to a certain environmental gradient but, after the environment stress reaches a threshold, another state of response may occur with severe fluctuations. Obviously, different ecological processes were reflected in different gradient ranges on both sides of the critical point of the environment. Thus, piecewise regression was performed to confirm the altitudinal trends of the traits. R studio (R Core Team, 2020) was used to perform the data visualization; we used the R package ggpubr (Kassambara, 2020) to plot boxplot figures and the packages Hmisc (Harrell and Dupont, 2021) and PerformanceAnalytics (Peterson and Peter, 2020) to plot the correlation matrix.

Petal H, petal L, and petal S vary significantly within the entire altitudinal range; however, at altitudes above 3,300 m, there was no significant difference for petal L and petal S (Table 2). The petal H was between 330° and 340° (Figure 1A), and regression analysis showed a linear relationship between petal H and elevation (Figure 3A). Altitude had a significant positive effect on petal S and a significant negative effect on petal L; piecewise regression showed that the gradient change of petal L and petal S had a clear breakpoint at around 3,300 m (Figures 3B,C). Below 3,300 m, petal S increased with an increase in altitude and reached its maximum value at 3,300 m, while a further increase in altitude had no significant effect on petal S. On the contrary, below 3,300 m, petal L showed a significant decreasing trend with an increase in altitude, while at above 3,300 m, no significant difference was observed. Moreover, the correlation analysis showed that both petal H and petal L had a significant negative correlation with altitude (Figures 4A–C)—specifically, the correlation between petal H and altitude was consistent within all altitudes, while the correlation between petal L and altitude was mainly found at a lower altitude (below 3,300 m). Within the entire altitudinal range, there was no significant correlation between petal S and altitude, but a positive significant correlation existed below 3,300 m.

Altitude had a significant effect on individual floral display area (IFDA), total floral display area (TFDA), individual Leaf area (ILA), total leaf area (TLA), individual floral fresh mass (IFFM), total floral fresh mass (TFFM), individual floral dry mass (IFDM), total floral dry mass (TFDM), floral number per twig (FN), and FWC (Table 2). Within the altitudinal gradient, no significant correlation was observed between altitude and FN, but at 3,300 m, FN was significantly higher than at other altitudinal gradients (Figure 4). As for the floral biomass, both the IFFM and the IFDM had a significant correlation with altitude (Figures 4A,B). However, the variation trend by segment was different for IFFM and IFDM; the breakpoint was at around 3,300 m, below which the IFFM and the IFDM both had a significant positive correlation with altitude. The difference occurred when altitude was above 3,300 m; the IFFM no longer changed significantly after 3,300 m, but the IFDM showed a descending trend (Figures 3D,E). The gradient variation of TFFM was similar to that of IFFM. It showed a significant correlation with altitude, and below the breakpoint 3,300 m, the TFFM was increasing with an increase in altitude. Afterward, the TFFM showed a slight drop compared to that at 3,300 m (Figure 3D). The TFDM had a similar gradient variation tendency with IFDM, but no significance was observed (Figure 3E).

The IFDA had no significant correlation with altitude (Figure 4A), while altitude had opposite effects on TFDA at both sides of 3,300 m (Figures 4B,C)—below 3,300 m, TFDA increased with an increase in altitude. However, at altitudes above 3,300 m, the TFDA decreased with an increase in altitude (Figure 3G).

The piecewise regression showed a breakpoint in the gradient change of FWC at around 3,300 m (Figure 3F). Below 3,300 m, the altitude had a significant negative effect on FWC, while the reverse was observed at above 3,300 m. The relationship between FWC and color performance could be determined in separated altitude ranges, with a breakpoint at around 3,300 m. The FWC had a significant positive effect on petal L and a significant negative effect on petal S up to 3,300 m, with no significant effect on petal H (Figure 4B). However, at above 3,300 m, the FWC showed a significant negative effect on petal H but no significant effect on either petal L or petal S (Figure 4C). As for the floral biomass, FWC had a significant negative correlation with IFFM and TFFM (Figures 4A–C). Otherwise, the floral biomass also had a separate correlation with color performance at different altitude ranges, and the floral biomass had a significant negative correlation with petal L while it had a positive correlation with petal S; both correlations were highly significant below 3,300 m (Figures 4A–C).

From the RDA analysis, within the whole range of altitude (2,950–3,650 m), the eigenvalues of each axis were 0.1620, 0.0466, 0.0134, and 0.4258, and the explained variations of each axis were 16.20, 20.86, 22.20, and 64.78%; at 2,950–3,300 m, the eigenvalues of each axis were 0.3964, 0.0364, 0.0153, and 0.3061, and the explained variations of each axis were 39.64, 43.28, 44.80, and 75.41%; at 3,300–3,650 m, the eigenvalues of each axis were 0.1405, 0.0762, 0.0337, and 0.4475, and the explained variations of each axis were 14.05, 21.67, 25.03, and 69.78%, respectively (Figure 5).