2.1.

Quantification and Classification of Striation Patterns in SHG Images of Muscle

We used the SHG microscope to capture digital image volumes from skeletal muscle tissue from a range of normal and diseased sources. In healthy myofibers, bands of sarcomeric SHG are straight and evenly spaced, with each band lying virtually orthogonal to the long axis (axis of contraction) of the cell. Damaged cells display a range of visible deviations from this norm, sometimes localized to subdomains within a giant myofiber: hypercontraction or hyperextension of striation spacing; misorientation, skewing, or contortion of bands; diminution or complete loss of SHG intensity (Fig. 1 ). An individual region of contractile apparatus can be found to display any one or a combination of these defects. To measure these changes and to assign criteria for scoring normal and abnormal structure, we applied the Helmholtz equation for wavenumber to calculate the local striation spacing and angle of orientation, thus creating new image maps for each of these metrics, in addition to the primary map of SHG intensity [Figs. 2a and 2b ; see Section 4 for details of the algorithm]. The deep optical sectioning power of the SHG microscope produced multislice image stacks that typically comprised more than $5\times {10}^6\mu {\mathrm{m}}^3$ of tissue. Thus, the fine spatial resolution of this sarcomere pattern quantitation (SPQ) algorithm allowed automatic assessment, by three distinct numerical criteria, of each of the tens of thousands of sarcomeres within the group of muscle fibers that were sampled by a single image volume. Unfortunately, the nonlinear dependence of measured SHG intensity upon sample thickness deterred us from assigning reliable abnormal/normal cut-off levels to make diagnostic distinctions based on image brightness. However, by reference to the literature on the physiology and ultrastructure of mammalian sarcomeres, and by empirical fitting to control images, we defined thresholds for scoring of abnormal sarcomere length ( $<1.6\mu \mathrm{m}$ or $>2.8\mu \mathrm{m}$ , a range characteristic of normal needle biopsy samples²⁷) and striation angle ( $<70°$ or $>110°$ ). These threshold measurments were independent of specimen-depth effects on image intensity, and they excluded consideration of any region lacking SHG pattern, therefore focusing entirely on myosin-containing structure within myofiber cells. Acquiring these objective SPQ measurements permitted detailed statistical analysis of the effects of specific MDs upon the structure of sarcomeres throughout the imaged volume.

Fig. 1

SHG imaging of normal and diseased muscle. Single optical sections of gastrocnemius muscle dissected from (a) $5\text{-}\text{week}$ -old control, (b) $md{x}^{-∕-};ut{r}^{+∕+}$ , and (c) $md{x}^{-∕-};ut{r}^{-∕-}$ animals, respectively. Yellow arrowheads mark gaps between myofibers; yellow and blue arrows label splits and ruptures, respectively, within the contractile lattice of myofibers; green arrows show myofiber areas with bright unstriated SHG. $\text{Scale}=20\mu \mathrm{m}$ . (Color online only.)

Fig. 2

Sarcomere pattern quantification and correlation of sarcomere-length distribution with severity of muscular disorders. Single SHG optical sections (left panels) and corresponding sarcomere length maps (center panels) and angle maps (right panels) from (a) $5\text{-}\text{week}$ -old control and (b) $md{x}^{-∕-};ut{r}^{-∕-}$ mice. Pseudocolor scale shows sarcomere lengths in $\mu \mathrm{m}$ (center) and angle of striations, relative to the long axis of the muscle fiber, in degrees (right). (c) Sarcomere length distributions affected by both mild and severe forms of muscular dystrophy. Histograms of sarcomere length were assembled from the full collection of SHG image stacks for each condition (see Section 4 for numbers of animals, specimens, and images). Fractions of hypercontracted sarcomeres $(<1.6\mu \mathrm{m})$ are shown; distributions of non-hypercontracted sarcomeres are filled with color matching the pseudocolor scale shown in (a). (d) Distribution of sarcomere length correlates with severity of various muscle disorders. Two characteristics of the length histograms (FNH and MNHL) were averaged for each entire image stack. Mean and standard deviation were calculated from the full collection of image stacks for each condition (see Section 4 for numbers of animals, specimens, and images). Asterisks $\left({}^{*}\right)$ indicate data significantly different from control $(p<0.05)$ , as measured by a $t$ -test for independent samples (Statistica 6.0, Statsoft, Inc.), with each image stack treated as an independent sample.

2.2.

Correlation of Sarcomere Pattern Scores with Damage Due to MDs in Mice

We tested the applicability of the SPQ scoring algorithm to detecting and quantifying disease in muscle. We imaged specimens of gastrocnemius muscle by SHG, comparing SHG image sets from control mice against SHG image data acquired from mice with each of three established models of muscular disorders: disuse-induced atrophy, hereditary muscular dystrophy, and aging (see descriptions of disease models in Section 4). The selected models each incur a characteristic range of damage within the full spectrum of phenotypic severity, allowing us to rate the sensitivity of SHG image analysis in detecting injury. Efforts to combine binary thresholds for both SPQ criteria (length and angle) to measure the abundance of normal versus abnormal sarcomeres provided rather reliable differentiation between healthy and diseased muscle samples (data not shown). However, we found that a detailed analysis focusing specifically on the distribution of sarcomere lengths within specimens produced even better distinctions. A consistent negative correlation between the severity of a disease and the mean sarcomere length was revealed for all disorder models we tested [Fig. 2d]. While sarcomere lengths spanned from $1.4\text{to}2.4\mu \mathrm{m}$ in biopsies of healthy muscles, both mild and severe forms of MDs showed a noticeable reduction in the range of sarcomere lengths and an increased fraction of hypercontracted sarcomeres with lengths below $1.6\mu \mathrm{m}$ [Figs. 2c and 2d].

To employ these differences in sarcomere length as a diagnostic marker, we extracted three distinct characteristics of the distribution of sarcomere lengths for each sample volume: the overall mean length (ML), the fraction of all sarcomeres that were not hypercontracted below $1.6\mu \mathrm{m}$ in length (FNH), and the mean length of the fraction of sarcomeres that were not hypercontracted (MNHL). Clear discrimination between healthy and disordered muscle could be achieved using either ML or FNH alone when control mice were compared to $mdx∕utr$ double mutants, which lack functional dystrophin and utrophin proteins and develop a severe form of Duchenne-like dystrophy [Fig. 2d].^{28, 29} Similarly, ML showed a significant difference between controls and animals subjected to atrophy by hindlimb suspension (HLS) [Fig. 2d], a protocol that decreased femur bone mineral density by an average of 12% and muscle fiber size by 13% (see Fig. 3 ). Consistent, but not statistically significant, alteration of sarcomere pattern relative to controls was detected in biopsies affected by other milder muscular disorders (mice with $mdx$ single mutations lacking only dystrophin, wild-type mice recovering from atrophy, or wild-type mice of advanced age). These milder forms of disease also displayed a broader range of FNH scores among the acquired image volumes, suggesting a mixture of normal and abnormal sarcomere pattern throughout the tissue of these animals.

Fig. 3

Muscle fiber cross-sectional areas from control, atrophied, and recovering mice. Data were gathered by transverse reslicing of muscle fibers within digital SHG image volumes of muscle. Each myofiber was outlined and the area in pixels was measured in ImageJ. Plot shows average and standard deviation for area measurements from myofibers in both gastrocnemius muscles of all mice included in each group (7 control, 6 HLS, 5 $\mathrm{HLS}+\text{reloaded}$ ). $n=884$ measured fibers for control, $n=686$ for HLS, $n=472$ for $\mathrm{HLS}+\text{reloaded}$ .

To more carefully test the diagnostic efficacy of SPQ scores, we performed receiver operating characteristic (ROC) analysis—both nonparametric and using multivariate logistic regression—to assess the specificity and sensitivity of SPQ measures in distinguishing groups by a clinically relevant statistical test (Table 1 ).^{30, 31, 32} As expected, evaluation of muscle damage based on a single sarcomere pattern parameter (FNH or MNHL) was sufficient to reliably detect severe MDs, such as dystrophy in $mdx∕utr$ mutants and muscle atrophy in hindlimb-suspended mice [ROC areas under the curve (AUC) $⩾0.89$ , with 95% confidence $>0.790$ for both FNH and MNHL in both comparisons; Table 1], at late stages when other signs of disease were clearly observed. Yet the diagnostic power of either of these parameters alone was not sufficient for clear discrimination of any moderate or mild form of MD from control samples (ROC AUC $⩽0.755$ , 95% confidence $⩽0.620$ for all cases; Table 1).

Table 1

Discrimination of control and diseased mouse muscle by SPQ measurements. Receiver operating characteristic area-under-the-curve (ROC AUC) values are shown in bold face for comparisons among control, mild, and severe disorders. Confidence intervals for each of these calculated values were derived by 1000-fold bootstrap resampling with replacement. The 95% confidence interval for the ROC AUC of each comparison is shown in parentheses under the measured ROC AUC value.

Remarkably, however, bivariate logistic regression combining both FNH and MNHL substantially improved the impact of SPQ scoring and allowed us in several cases to efficiently discriminate intermediate forms of MDs from either control samples or severely affected individuals. For example, in a test for the subtle changes brought on by aging, we achieved quite effective differentiation of $24\text{-}\text{month}$ -old mouse muscle from $10\text{-}\text{month}$ -old specimens (ROC AUC 0.866, 95% confidence 0.786; Table 1). This contrast was detected despite our finding of no significant difference in either agility or the percentage of fat-free body mass between breed-matched animals of these approximate age groups (Fig. 4 ) and the absence of any visibly recognizable alterations of sarcomere pattern. Even more strikingly, both $mdx$ (mild) and $mdx∕utr$ (severe) dystrophic animals were absolutely discriminated from wild-type controls by a logistic regression model combining all three measures of sarcomere length distribution (ROC AUC 1.000, 95% confidence 1.000; Table 1). The definitiveness of this result stands out, especially with regard to $mdx$ , because the pathology of $mdx$ mice is known to be subtle in the early weeks of life.^{29, 33, 34, 35}

Fig. 4

Mobility performance and body composition of mature adult versus old mice. (a) In agility testing on the rotarod apparatus, each animal was subjected to a warm-up run followed by five sequential mobility tests as described in Section 4. Each of the five bars (left to right) represents the mean value ± SEM of the rotational velocity at which mice fell during each of the five trials: trial 1 $(0–24\mathrm{rpm})$ , trial 2 $(0–36\mathrm{rpm})$ , trial 3 $(0–36\mathrm{rpm})$ , trial 4 $(12–48\mathrm{rpm})$ , trial 5 $(12–48\mathrm{rpm})$ . $n=5$ for $10\text{-}\text{month}$ -old group, $n=10$ for $21\text{-}\text{month}$ -old group. (b) Body composition measurements from X-ray absorptiometry analysis. $n=10$ and 20 for mature and aged animals, respectively; all were male C57BL6 mice purchased from NIA-monitored aging colonies. Results were analyzed using one-way ANOVA and, when significant, the Holm–Sidak procedure for pairwise multiple comparisons was performed. $\left({}^{\&}\right)$ represents statistically significant differences $(p<0.05)$ when compared to values in $10\text{-}\text{month}$ -old animals.

To examine the applicability of SPQ scoring in monitoring muscle healing after damage, we analyzed the alteration of sarcomere pattern associated with reloading of HLS-atrophied muscles. Two days of recovery did not change the cross-sectional diameter of muscle fibers in the gastocnemius (Fig. 3), and has been shown in the literature to yield no improvement in either muscle mass or bone mineral density in adult C57BL6 mice (Ref. 36 and our own unpublished data). Nonetheless, we found a clear improvement of muscle SPQ values within hindlimb-reloaded muscle tissue imaged by SHG. In fact, recuperation of pattern within just two days of reloading was sufficient to reliably segregate recovering animals from the fully atrophic (ROC AUC 0.866, 95% confidence 0.799 in the trivariate regression model), and recovering limbs were actually more accurately discerned from HLS-atrophy specimens than from controls (Table 1). This result is remarkable, for it reveals both the speed of contractile adaptation to newly regained mobility and the sensitivity of SHG/SPQ to the early onset of musculoskeletal remodeling during physiological recovery from disease.

2.3.

SPQ of Changes in Human Aging

To consider whether these methods were also applicable in human muscle, and to ask if SPQ would correlate with functional performance, we recruited groups of unrelated young adult and elderly volunteers who underwent tests of physical performance and provided a needle biopsy of the quadriceps muscle, vastus lateralis. We acquired SHG images and calculated SPQ values from each of the biopsies. As was seen for mice, SPQ appeared to efficiently resolve specimens from the young and old groups, even by univariate analysis of individual scores [ROC AUC values $\mathrm{FNH}=0.866$ , $\mathrm{MNHL}=0.839$ , $\mathrm{ML}=0.797$ ; see Fig. 5b ]. Small sample sizes for this study led to decreased lower bounds for confidence in ROC analysis of any of these individual scores. However, we found for human specimens that bivariate and trivariate logistic regression (combining FNH, MNHL, and/or ML) substantially improved the impact of SPQ scoring, to quite closely parallel the results seen for aging in mice [Fig. 5c].

Fig. 5

Quantification of human sarcopenia of aging by SPQ analysis. (a) Means of stack-average values of FNH and ML for biopsies from four young adult (22 image stacks) and 11 old human volunteers (39 image stacks). Error bars show standard deviation. (b) ROC curves for distinction between muscle biopsies from human volunteers. Areas under the curve for individual measures of the sarcomere length distribution were $\mathrm{FNH}=0.866$ , $\mathrm{ML}=0.797$ , $\mathrm{MNHL}=0.839$ (MNHL not shown). (c) Discrimination of control and diseased mouse muscle by SPQ measurements. Receiver operating characteristic area-under-the-curve (ROC AUC) values are shown for comparisons between young healthy adult and elderly volunteers. 95% confidence intervals for the AUC are shown in parentheses. Abbreviations: $\mathrm{ML}=\text{mean}$ sarcomere length; $\mathrm{FNH}=\text{fraction}$ of all sarcomeres that were not hypercontracted; $\mathrm{MNHL}=\text{mean}$ length of non-hypercontracted sarcomeres; $\mathrm{BVLR}=\text{bivariate}$ logistic regression. $\mathrm{TVLR}=\text{trivariate}$ logistic regression. Values highlighted in gray have lower limits for 95% confidence which are above 0.74, indicating a good diagnostic test.

4.1.

Animal Models of Atrophy, Dystrophy, and Sarcopenia

Muscle atrophy was induced in $6\text{-}\text{month}$ -old C57BL6 mice by a $14\text{-}\text{day}$ course of HLS, as described in the following section. HLS in both mice and rats has been shown to result in substantial atrophy of both muscle and bone within the hind legs during this time span. Returning animals to walking on all fours induces a recovery of muscle force, mass, and fiber size, as well as bone mass recovery, at a rate that is comparable to or slower than the initial rate of atrophy. ^{42, 43, 44, 54, 55, 56, 57, 58} Mice of the $md{x}^{-∕-}$ genotype lack functional dystrophin protein and show continual necrosis and regeneration of muscle fibers starting at about $20\text{days}$ of age and extending throughout the full course of life.^{33, 34, 35, 39} Yet $md{x}^{-∕-}$ animals show normal muscle strength through much of life and can achieve close to 80% of the life spans of controls.^{60, 61} In contrast, $md{x}^{-∕-};ut{r}^{-∕-}$ mice, which lack both dystrophin and the paralogous protein utrophin, display much more rapid incidence of muscle fiber damage and have a maximum lifespan of $20\text{weeks}$ .²⁹ Sarcopenia, the progressive loss of muscle mass and strength with aging, has been found to affect animals across many phyla. To evaluate the effectiveness of SHG imaging in recognizing sarcopenic changes in the mouse, we examined muscle from mice at mid-adulthood ( $10\text{months}$ old) and at an advanced age $\left(24\text{months}\right)$ close to the average lifespan of the wild type.⁶⁰ All animal protocols were approved by the Animal Care and Use Committee of the University of Connecticut Health Center (UCHC), Farmington, CT.

4.2.

ROC Analysis and Bootstrapped Estimation of Error

The receiver operating characteristic curve compares the sensitivity and specificity of a classifier system as the threshold for discrimination is varied, indicating the true-positive and false-positive rates for classification of members of two distinct groups.^{30, 31, 32} A test with no power in distinguishing between two groups yields a value of 0.5 for the area under the ROC curve (AUC), while a perfect classifier yields 1.0 for the ROC AUC. We calculated ROC curves and AUC values for primary data using STATA and Microsoft Excel. For estimation of error from our relatively small samples (and taking into account dependencies between image volumes from the same animal/patient, leg, or biopsy), we generated 1,000 model data sets for each class of specimen by bootstrap selection of data points with replacement, using custom-written C++ code. Logistic regression models were generated and ROC curve errors were calculated from the bootstrapped data sets using STATA.