Responsive monitoring of mitochondrial redox states in heart muscle predicts impending cardiac arrest

See allHide authors and affiliations

Science Translational Medicine  20 Sep 2017:
Vol. 9, Issue 408, eaan0117
DOI: 10.1126/scitranslmed.aan0117

Monitoring mitochondria to predict cardiac dysfunction

Mitochondria adapt to changing environmental conditions to try to meet the energy needs of a cell. Perry et al. used resonance Raman spectroscopy to monitor the mitochondrial redox state in the heart. Under low oxygen conditions, electron transport chain cytochromes within the mitochondria become progressively more reduced, which is detected on the basis of the spectral signal. The authors identified a threshold value that predicted the onset of cardiac arrest in rats under hypoxemic conditions and showed that they could monitor mitochondrial redox state during myocardial ischemia-reperfusion in pigs. Although the analysis was restricted to the heart, this approach could be useful for monitoring perfusion and viability of other tissues.


Assessing the adequacy of oxygen delivery to tissues is vital, particularly in the fields of intensive care medicine and surgery. As oxygen delivery to a cell becomes deficient, changes in mitochondrial redox state precede changes in cellular function. We describe a technique for the continuous monitoring of the mitochondrial redox state on the epicardial surface using resonance Raman spectroscopy. We quantify the reduced fraction of specific electron transport chain cytochromes, a metric we name the resonance Raman reduced mitochondrial ratio (3RMR). As oxygen deficiency worsens, heme moieties within the electron transport chain become progressively more reduced, leading to an increase in 3RMR. Myocardial 3RMR increased from baseline values of 18.1 ± 5.9 to 44.0 ± 16.9% (P = 0.0039) after inferior vena cava occlusion in rodents (n = 8). To demonstrate the diagnostic power of this measurement, 3RMR was continuously measured in rodents (n = 31) ventilated with 5 to 8% inspired oxygen for 30 min. A 3RMR value exceeding 40% at 10 min predicted subsequent cardiac arrest with 95% sensitivity and 100% specificity [area under the curve (AUC), 0.98], outperforming all current measures, including contractility (AUC, 0.51) and ejection fraction (AUC, 0.39). 3RMR correlated with indices of intracellular redox state and energy production. This technique may permit the real-time identification of critical defects in organ-specific oxygen delivery.


The adequate delivery of oxygen to mitochondria is vital to organ function. Abnormalities in oxygen delivery (DO2) are central to the pathology of many critical illness states, including sepsis, shock, and the systemic inflammatory response syndrome. Deficient DO2 not only leads to deficient energy production but may also lead to the formation of reactive oxygen species and establish the milieu for cellular injury and death (1). Currently, the adequacy of DO2 is assessed using the oxyhemoglobin saturation of venous blood (SvO2) (2, 3). However, this measure provides only global information and may be falsely reassuring in the sickest patients. For example, patients with tissue edema (4), altered erythrocyte rheology (5), or microvascular arteriovenous shunting (6) may experience impaired mitochondrial oxygen delivery yet have a normal or even elevated SvO2. In some situations, such as after vascular reconstruction or congenital heart surgery, understanding the adequacy of local DO2 is desirable.

Oxygen is used within the mitochondrion to produce energy via aerobic metabolism. It serves as the final electron acceptor at complex IV of the electron transport chain (ETC), where electrons at higher energy potential originating from NADH [the reduced form of nicotinamide adenine dinucleotide (NAD+)] and FADH2 successively lose energy. This energy loss drives proton pumps in complexes I, III, and IV to establish a membrane potential used in adenosine 5′-triphosphate (ATP) production (7). When oxygen delivery to the mitochondrion decreases below a critical level, the flux of electrons is diminished, causing upstream effects. One effect is that the concentration of proximal reducing agents, such as NADH, increases. Using its potent fluorescence properties, NADH concentration has been used to visualize areas of compromised tissue oxygenation in real time using animal models of coronary ischemia (8) and of cranial stimulation (9). Unfortunately, the presence of hemoglobin interferes with NADH fluorescence measurements, which has reduced the clinical utility of this tool.

Alternatively, deficient DO2 may be quantified through assessment of the mitochondrial redox state. It has long been recognized that the redox state of the active sites of electron flux within the ETC—mitochondrial cytochromes—is spectroscopically quantifiable and varies based on the availability of oxygen (10, 11); deficient DO2 results in the progressive reduction of cytochromes. Cytochrome a,a3 in complex IV is an ideal target for quantifying oxygen supply-demand relationships because it is oxidized directly by molecular dioxygen (1214) and accounts for >95% of cellular oxygen utilization (15). In the past, several groups have attempted to quantify the redox status of cytochrome a,a3 using absorbance spectroscopy (1620). However, these efforts have been limited by the overlap in the absorption spectra of the cytochromes of interest and other heme moieties that are present in vivo (for example, hemoglobin and myoglobin). This causes absorbance-based estimates of cytochrome redox status to vary artificially with changes in blood pressure and intravascular blood volume (21, 22). Furthermore, tissue edema and inhomogeneous tissue scattering complicate the determination of absolute concentrations of cytochrome components (22). More recently, refinements to this technique have permitted isolation of the cytochrome and hemoglobin signals, although these approaches are still limited to describing trends, rather than absolute values, of cytochrome redox status (20).

An alternative approach to the quantification of mitochondrial redox status is Raman spectroscopy, wherein the wavelength of light from a narrow band laser is shifted to lower energy by a precise quantity determined by the frequency of the vibrational modes of the molecules it encounters. The wavelength shift (Stokes shift) of inelastically scattered light can be separated from fluorescence to measure a redox state–specific spectral signature of a molecule. In the special case of resonance Raman spectroscopy (RRS), the optically excited state overlaps a strong electronic absorption line, resulting in orders of magnitude enhancement of the Raman cross section (23). Relevant to cellular energetics, the resonance Raman (RR) profiles of porphyrin structures (present in hemoglobin, myoglobin, and mitochondrial cytochromes) have been well described (2426) and are amplified by four to six orders of magnitude (enhancement factor) when excited near the Soret absorption band (400 to 450 nm) (27, 28). This enhancement makes the in vivo quantification of small quantities of such structures possible, even in a complex environment (2932). Using this approach, the redox state of mitochondrial cytochromes has been described in isolated mitochondria, in myocytes, and in bloodless tissues (3335). The technique has also been applied to the measurement of tissue oxyhemoglobin saturation in vivo (36). Quantification of mitochondrial redox state in vivo may represent a powerful and specific predictor of impending cardiac failure.

Here, we describe an RRS tool that quantifies the redox state of mitochondrial cytochromes on the epicardial surface in rodent and porcine models. We define a new measure, the RR reduced mitochondrial ratio (3RMR), and compare its sensitivity and specificity for predicting cardiac arrest, the cessation of cardiac function, as a manifestation of a terminal deficiency of myocardial energy production.


RRS instrument

We designed and constructed an RRS system for in vivo use with the following components (fig. S1). A compact, custom-built, 441-nm single-mode laser was thermally stabilized to prevent wavelength peak drift and wavelength mode hopping. Laser light was delivered to the probe head via a custom glass fiber optic bundle cable with laser cleanup and rejection filters integrated at the distal end. The laser spot on the tissue is 1.5 mm in diameter. The filtered RR photons are collected at the distal end, and the proximal end of the fiber bundle is coupled to the slit of a custom high-resolution spectrometer with a temperature-controlled two-dimensional charge-coupled device (CCD) detector array. The spectrometer is designed to minimize the effect of mechanical and thermal drift. Furthermore, a custom internal acetaminophen standard is referenced continuously and used to correct for spectrometer wavelength shift and laser wavelength drift. The Stokes shift accuracy is estimated to be ≤0.4 cm−1, with more accurate repeatability in the 700 to 1700 cm−1 spectral range.

Algorithmic approach

The collected RR spectrum was used to quantify each of the heme-containing components (table S1) using a regression algorithm and a spectral library of each component. First, we created a spectral library of the heme-containing components found in the myocardium, including isolated hemoglobin, myoglobin, and mitochondrial cytochromes, each in the oxidized/oxygenated and reduced/deoxygenated state (six components total) (Fig. 1). Then, we removed the fluorescence signals (Embedded Image and Embedded Image) from the collected and filtered raw spectrum, y, in two iterative steps, creating a refined estimate of the final RR spectrum (yR) (Fig. 2). The relative concentration of each component was determined by calculating the regression coefficient of each in an equation explaining the final RR spectrum (yR) as a weighted sum of each component’s spectrum. This allowed calculation of oxyhemoglobin saturation (SHbO2), oxymyoglobin saturation (SMbO2), and the ratio of reduced to total mitochondrial cytochromes. We summarized the latter of these metrics as the 3RMR, representing the ratio of the regression coefficient of reduced mitochondria to the sum of regression coefficients for oxidized and reduced mitochondria.

Fig. 1. Spectra of the oxygenated/oxidized and deoxygenated/reduced forms of hemoglobin, myoglobin, and mitochondria and their structural implications.

Raman scatter reflects vibrations, stretching, or bending of specific bonds within the porphyrin ring (A), a structure present in hemoglobin, myoglobin, and mitochondrial cytochromes. Shown here are enhancement factor-corrected spectra for mitochondria isolated from rodent myocardium (B), equine myoglobin (C), and purified rodent hemoglobin (D). In (B) to (D), blue lines represent deoxygenated/reduced species, and red lines represent oxygenated/oxidized species. The intensity of each spectrum is normalized to 1 arbitrary unit (AU) by correcting for the enhancement factor of each species. The identified peaks are in agreement with previously identified vibrational modes, as described by Spiro (54), Kitagawa et al. (55), Adar and Erecinska (56), and Argade et al. (43) for mitochondrial cytochromes and by Spiro and Strekas (44, 54) for myoglobin and hemoglobin.

Fig. 2. Spectral processing algorithm.

(A) Coarse baseline estimation: The processed, averaged, spectrum y gathered by the device includes a potent fluorescent baseline. A coarse estimate of this baseline, Embedded Image is obtained by jointly estimating fifth-order polynomial coefficients and library spectra coefficients using a linear regression. The first iteration of the true RR spectrum is thus given by Embedded Image. (B) Iterative baseline refinement: A composite regression curve ŷ0, which most closely fits y0, is then calculated from a linear combination of library spectra weighted by regression coefficients. A slow-varying cubic spline fit to the residual res0 = y0ŷ0, denoted by Embedded Image, is added to the initial baseline to include any slow wobbles in the baseline that were not captured by the polynomial fit. (B, inset) Components of ŷ0 include oxyhemoglobin (HbO), deoxyhemoglobin (Hb), oxymyoglobin (MbO), deoxymyoglobin (Mb), and oxidized and reduced mitochondrial spectra. (C) The iterative baseline fitting procedure is repeated thrice to obtain the final estimate (ŷR). The residual (res) is calculated by subtracting the final regression curve (ŷR) from the baseline-adjusted spectrum (yR).

3RMR performance during ex vivo ischemia

To first determine the effect of ischemia on 3RMR within a bloodless system (removing hemoglobin interference), we studied isolated crystalloid-perfused hearts using a Langendorff model of graded ischemia. Rodent hearts were explanted and retrograde-perfused, and coronary flow rate was decreased stepwise from 100 to 0% of baseline. Myocardial tissue oxygen tension (tPO2) measured by fluorescence quenching (37) decreased from a baseline of 97.3 ± 5.6 to 1.4 ± 1.3 mmHg at 0% flow. Baseline 3RMR was 22 ± 3% (about 20% of the resonantly enhanced cytochromes represented in the RR spectrum were in the reduced state and 80% in the oxidized state when myocardial DO2 was baseline); 3RMR changed to 54% reduced and 46% oxidized during ischemia (Fig. 3A). In addition, increased myocardial 3RMR correlated with reduced myocardial contractility, as measured by a left ventricular pressure balloon (Fig. 3B). Decreasing tissue oxygen tension (tPO2) also correlated with decreased contractility (Fig. 3C).

Fig. 3. 3RMR response to graded ischemia ex vivo.

(A) As coronary flow rate (CFR) was incrementally decreased from baseline in a flow-controlled retrograde Langendorff perfusion experiment, myocardial tissue oxygen tension (tPO2, gray bars) decreased from a baseline of 97.3 ± 5.6 mmHg (mean ± SD) to 1.4 ± 1.3 mmHg [P < 0.0001, repeated-measures analysis of variance (ANOVA)]. Contemporaneously, the 3RMR increased from a baseline of 22 ± 3% during full perfusion to 54 ± 14% during complete ischemia (P < 0.001, repeated-measures ANOVA). Data are means, and error is SD. (B) Increases in 3RMR were nonlinearly (single-phase decay) associated with decreases in contractility (dP/dTmax). Data are means of each 10-min observation period. The line is a single-phase decay line with 95% confidence interval (CI) of regression line; P < 0.0001, Spearman correlation coefficient r = 0.88; r2 = 0.69. (C) Contractility (dP/dTmax) was linearly associated with tPO2. Data are means of each 10-min observation period. Line is a linear regression line with 95% CI of regression line; P < 0.0001, Pearson correlation coefficient r = −0.88; r2 = 0.71; n = 10 rodents.

3RMR performance during in vivo ischemia

We then tested the ability of the RRS system to detect a change in mitochondrial redox state in vivo using a model of inferior vena cava (IVC) occlusion. 3RMR was measured on the epicardial surface via an open sternum of rodents as the IVC was snared and occluded. This created an abrupt decline in cardiac output and blood pressure (Fig. 4A). During the 10-min period of IVC occlusion, tPO2 measured using an embedded fluorescence quenching probe as above significantly decreased (P = 0.0006) and 3RMR significantly increased (P = 0.0039, Fig. 4B). The change in 3RMR from baseline was associated with a progressive decline in myocardial contractility as measured using a conductance catheter (Fig. 4C).

Fig. 4. 3RMR response during in vivo occlusion of the IVC.

(A) After complete IVC occlusion (IVCO), tissue oxygen tension (tPO2, gray line) decreased from a baseline of 25 to 10 mmHg within 2 min. The increase in 3RMR signal (red line) was delayed from this decrease in tPO2 by 2 to 3 min (integration time, 3 min). Mean arterial blood pressure (mABP, black line) decreased abruptly after IVCO as is typical for acute changes in myocardial preload. n = 1 representative sample. (B) After 10 min of IVCO (post), myocardial tPO2 decreased significantly (P = 0.0006) and 3RMR increased significantly (P = 0.0039, paired t test). (C) Acute IVCO caused an abrupt decrease in myocardial preload, which significantly decreased contractility (defined here as dP/dTmax). From this new (that is, after occlusion) baseline, 3RMR significantly increased (P < 0.0001, red bars) and was associated with a significant decrease in contractility (P = 0.0027, blue bars, linear trend repeated-measures ANOVA). (B and C) Data are means, and error is SEM. n = 8 rodents.

3RMR performance during prolonged hypoxemia

We then examined the effects of hypoxemia on epicardial 3RMR, interrogating its ability to predict impending myocardial dysfunction and cardiac arrest. Rats were instrumented to include an open sternum for placement of the RRS probe over the epicardial surface, an indwelling tissue oxygen tension probe, and a left ventricular pressure-volume conductance catheter under positive pressure mechanical ventilation. After stabilization, rodents were exposed to severe hypoxia (inspired oxygen fraction between 5 and 8%) for a period of 30 min. 3RMR exhibited a nonlinear correlation with tPO2 that we approximated using a hyperbolic function (Fig. 5A). Above a tPO2 of 10 mmHg, 3RMR was nearly always less than 40%. The specificity of 3RMR > 40% for severe tissue hypoxia (tPO2 < 10 mmHg) was 97% (fig. S2). RRS-based tissue oxyhemoglobin (Fig. 5B) and oxymyoglobin (Fig. 5C) concentrations also correlated with tPO2.

Fig. 5. Correlations between tPO2 and RRS-based 3RMR, oxyhemoglobin, and oxymyoglobin concentrations in rodents.

(A) After baseline measurements on 100% oxygen, arterial hypoxemia was induced by ventilation with 5 to 8% oxygen. As tissue oxygen tension (tPO2) decreased, 3RMR increased from baseline levels of ~25 to >40%. Among 1859 data points (1-min medians), 3RMR was <40% at all but 12 points in which tPO2 was >10 mmHg, making a 3RMR < 40% clinically reassuring (specific diagnostic test). When tPO2 < 10 mmHg, 3RMR exceeded 40% in most data points. (B) RRS-based tissue oxyhemoglobin saturation (SHbO2) was related to tPO2 using hyperbolic fit. (C) Tissue myoglobin saturation (SMbO2) and tPO2 were also correlated using a hyperbolic fit. (A to C) Line is a nonlinear regression line in the form y = xn/(k + xn), as expected for oxygen binding to heme. (A) n = 2.8, r2 = 0.59; (B) n = 1, r2 = 0.62; (C) n = 1, r2 = 0.7.

We then evaluated the performance of a single–point-in-time 3RMR measurement as a diagnostic test for impending cardiac arrest. We examined incremental 3RMR values at 10 min of hypoxia for each animal, finding that a threshold of 40% provided maximal sensitivity and specificity as a diagnostic tool for predicting cardiac arrest. Animals were, thus, separated into two groups: 3RMR > 40% or 3RMR ≤ 40% at 10 min of hypoxia. Animals in the first group exhibited an early and sustained elevation in 3RMR, whereas 3RMR remained below 40% for the majority of the experiment in the other group (Fig. 6A). These two groups were similar at baseline (table S2). At clinically extreme and comparable degrees of hypoxemia (arterial oxygen tension [PaO2] at 10 min 27.5 ± 4.4 mmHg versus 32.6 ± 6.1 mmHg; P = 0.0162), the group with elevated 3RMR developed a progressively lower tissue oxygen tension (Fig. 6B), a lower subsequent contractile state (Fig. 6C), and a higher incidence of sudden cardiac arrest within 30 min (Fig. 6D). 3RMR provided better risk stratification than did the current best standards, including tPO2 and hemoglobin or myoglobin oximetry (fig. S3). When assessed at 10 min of hypoxemia, 3RMR > 40% was 95% sensitive and 100% specific for subsequent cardiac arrest (Fig. 6E), which was predicted more accurately by 3RMR [area under the curve (AUC), 0.98] than by tPO2 (AUC, 0.95), hemoglobin saturation (AUC, 0.86), myoglobin saturation (AUC, 0.85), contractility (AUC, 0.51), or ejection fraction (AUC, 0.38) (Fig. 6F). Intracellular ATP concentration in freeze-clamped tissues was lower, and NADH concentration was higher in the 3RMR > 40% group than in the 3RMR ≤ 40% group (fig. S4). Although RRS-based tissue oxyhemoglobin saturations correlated with tPO2, they were not significantly different between the high and low 3RMR groups (P = 0.697; fig. S5).

Fig. 6. Raman-based oxygenation and hemodynamic data of two groups of rodents in vivo, separated by 3RMR at 10 min.

(A) 3RMR values over time in animals in which 3RMR at 10 min exceeded 40% (red line, n = 12) and those in which it was at or below 40% at 10 min of hypoxia (black line, n = 19) were significantly different (P < 0.0001, repeated-measures ANOVA with Bonferroni correction). (B) In animals with higher 3RMR, myocardial tPO2 was significantly lower than in animals with a low 3RMR at 10 min (P = 0.001, repeated-measures ANOVA with Bonferroni correction). (C) Myocardial contractility (dP/dTmax) was similar between groups at 10 min (P = 0.065) but was significantly lower in the high 3RMR group by 30 min (P < 0.0001), suggesting the inability to increase contractility in response to hypoxemia. (A to C) Data are means, and error (shaded) is SEM. (D) Animals in the higher 3RMR group exhibited a significantly higher incidence of cardiac arrest within the 30-min observation period compared with those in the lower 3RMR group (P < 0.0001, log-rank test). (E) Sensitivity (solid black) and specificity (dotted black) for 3RMR values at 10 min, as predictive of cardiac arrest within 30 min, were jointly maximized at a threshold 3RMR value of 40%. (F) Receiver operating characteristics plot of 3RMR (solid red, AUC 0.98), tissue oxygen tension (dotted black, AUC 0.93), tissue oxyhemoglobin saturation (blue dot-dash, AUC 0.82), contractility (dotted green, AUC 0.51), and ejection fraction (orange dot-dash, AUC 0.39) measurements at 10 min as diagnostic tests predicting impending cardiac arrest in the following 20 min.

3RMR performance during myocardial ischemia in swine

Finally, we examined the performance of 3RMR measurements in swine on cardiopulmonary bypass. Yorkshire swine (n = 5) underwent cannulation for cardiopulmonary bypass, aortic root cross-clamping, a 75-min period of ischemia after administration of cold blood cardioplegia solution, followed by a period of reperfusion. We found that baseline 3RMR measured over the left ventricle, ~6 to 7 mm in thickness, was primarily oxidized at baseline (16.8 ± 6.2%), similar to values obtained in rodents. During the ischemic period, 3RMR progressively increased to 54.7 ± 15.1% (P = 0.0108, paired t test). Within 3 min of reperfusion, 3RMR decreased to baseline levels (11.7 ± 3.5%), potentially lower than baseline because oxygen provision was copious and metabolic activity was diminished because of hypothermia (Fig. 7, A and B). RRS-based tissue oxyhemoglobin saturation (Fig. 7C) and oxymyoglobin saturation (Fig. 7D) also changed during the ischemic period.

Fig. 7. 3RMR performance in swine undergoing aortic cross-clamping and subsequent reperfusion.

(A) In swine, 3RMR was 16.8 ± 6.2% at baseline, increased to 54.7 ± 15.1% after 75 min of ischemic time, and then decreased to 11.7 ± 3.5% 3 min after reperfusion (P = 0.0015, repeated-measures ANOVA with Dunnett’s correction). (B) After myocardial reperfusion with cardiopulmonary bypass, 3RMR decreased to near-baseline levels within 3 min (n = 1, representative sample). (C) RRS-based tissue oxyhemoglobin saturation (SHbO2) decreased during ischemia, also returning to baseline levels during reperfusion (P = 0.005), as did oxymyoglobin saturation (SMbO2, P = 0.03; D). (A, C, and D) *P < 0.05 and **P < 0.01. All data are means, and error is SEM. n = 5 swine.


We developed an RR spectroscopic system that detects changes in the mitochondrial redox state in vivo. Myocardial redox state changed responsively, within 2 to 3 min, in models of both ischemia and hypoxemia. These changes not only correlated with myocardial tissue oxygen tension and tissue oxyhemoglobin saturations but also predicted impending decrements in cardiac contractility. In both ischemia (IVC occlusion) and hypoxemia, the decrement in myocyte contractile function temporally followed the increase in myocardial 3RMR, suggesting that an elevation in 3RMR signals a symptomatic deficiency in electron transport. In the model of hypoxemia, 3RMR served as an early warning indicator for impending cardiac arrest: Changes present at 10 min permitted risk stratification for cardiac arrest in a highly sensitive and specific way. In the setting of cardiopulmonary bypass, 3RMR increased progressively during the ischemic period and became reoxidized to baseline levels within minutes of reperfusion, demonstrating the temporal responsiveness of the 3RMR measurement.

This technique represents a new paradigm to assess the adequacy of DO2 to a tissue. Current methods to assess this vitally important end point focus either on measures of tissue oxygenation or on metrics of organ function. tPO2 may be the purest metric of tissue oxygenation but is rarely used because of its invasive nature. Even so, tPO2 and its surrogates, such as mixed venous oxyhemoglobin saturation, are unable to determine when a given degree of tissue hypoxia has led to impaired mitochondrial electron transport. Markers of organ function (serum creatinine), organ injury (hepatic transaminase), or anaerobic metabolism (serum lactic acid) are at best delayed surrogate indicators of inadequate DO2 and are primarily useful to identify trends over time. In contrast, 3RMR provides a window into electron flux by measuring an effective redox state of mitochondria; when oxygen (electron sink) delivery is insufficient compared to the electron flux, cytochromes within the ETC become reduced. It is possible that beyond a given 3RMR threshold, the cell is unable to manufacture sufficient energy to maintain function. In support of this, we found that elevations in 3RMR beyond 40% at a single point in time heralded a decrement in contractility followed by cardiac arrest. In addition to predicting organ dysfunction, here manifesting as cardiac arrest, identification of mitochondrial redox status may also predict apoptosis and cellular injury, vital processes that are thought to be initiated by abnormal mitochondrial redox states (38, 39).

Several technical aspects of our approach permitted this quantification of mitochondrial redox state in vivo. First, we chose a laser excitation wavelength (441 nm) in the Soret band region, which is known to have resonance with heme-containing moieties, including mitochondrial cytochromes. At this wavelength, it is known that cytochrome a,a3 exhibits strong resonant enhancement (25, 4043) and therefore likely contributes considerably to our composite mitochondrial signal. Given that the amplitude of resonant Raman enhancement increases with proximity to the Soret absorption maximum (44), we expect that cytochrome b-c1 and cytochrome c, which both exhibit Soret absorption maxima between 400 and 450 nm (table S1), also contribute to the 3RMR spectrum. In the future, the inclusion of additional wavelengths and expansion of our library to include individual cytochromes may permit functional interrogation of electron transport. Second, in contrast to absorbance techniques, the uniqueness of the RR fingerprint of the 3RMR components permitted the precise identification of low concentrations of mitochondrial species even in the presence of abundant hemoglobin and myoglobin. We found that neither the 3RMR value nor the signal intensity changed acutely after IVC occlusion despite concomitant changes in blood pressure (a limitation found in absorbance spectroscopic approaches), suggesting that the 3RMR signal was distinct from that of hemoglobin. Third, we used a temperature-stabilized laser with an internal reference to correct for optical drift, which enhanced the integrity of the measured spectra. Fourth, we used a regression approach with iterative baseline smoothing followed by modeling to quantify the fit coefficient of each component, which corresponded to the contribution of each known substance to the whole signal. Previous descriptions of RRS relied upon individual peak assignment to identify and quantify a substance (45). Instead, our process normalizes signal strengths and allows quality checking of the fractional coefficient estimate through assessment of the randomness of the residual and confidence estimates for each value. In all experiments, we found that the observed RR spectrum from the myocardium is well explained by the linear addition of the spectra of each component with almost shot-noise–limited (theoretical limited) residual spectrum. We therefore have statistical confidence in our estimates of hemoglobin and myoglobin oxygen saturation and the value of 3RMR, with the following assumptions. (i) We assume that the components of the spectral library are the only substantial contributors to the tissue spectra. We believe this assumption to be acceptable given the low amplitude and random nature of the residua observed. (ii) We assume that the iterative baseline correction does not meaningfully change the results. We believe this to also be an acceptable assumption because changes in the spline analyses did not significantly alter the final results. (iii) We assume that the mitochondrial spectrum can be completely explained across the entire range of redox states by a combination of the purely oxidized and purely reduced mitochondrial spectra. We find this assumption to be acceptable because it explains the collected spectrum with a random final residual.


First, although this technique provides a measurement of local mitochondrial redox state, we did not examine its utility as a marker of global perfusion. Second, at the current excitatory wavelength, the depth of penetration in vivo may be as little as 0.2 to 0.5 mm (table S3). Therefore, the device requires direct optical access to tissues, and its application is therefore limited to environments in which tissues are directly accessible, such as the operating room, or through the use of an endoscopic probe. Finally, we found the device to be sensitive to the presence of blood at the interface between the probe head and the tissue under examination, a space that must remain clear of chromophores to maintain spectral integrity. This is addressed by cleaning the probe head and tissue interface before measurements. Similarly, the probe is sensitive to interference from ambient light and functions optimally in a dark environment. Although this is possible in a surgical setting (as for echocardiograms), the incorporation of a probe cover may obviate the need for a dark working environment. Interference from either blood or light artifact can be identified through specific spectral features, which may serve as the basis for a signal quality indicator.

This work provides a platform for the future interrogation of mitochondrial function in human tissues, with applications in diverse fields of medicine. The technique may be used to monitor tissue viability during surgery (including cardiac surgery), the protection of organs explanted for transplantation, or the identification of critical limb ischemia. Monitoring 3RMR on other tissues, such as the tongue, esophagus, or bladder, may provide a global assessment of the adequacy of DO2 and a new target for resuscitation. It may open new avenues of investigation relating changes in mitochondrial redox state over time to subsequent ischemia-reperfusion injury. Mitochondrial function is also central to cancer biology, myopathies, aging, and mitochondrial diseases, for which a noninvasive mitochondrial probe could lead to new avenues of research and facilitate drug discovery. Expansion of the Raman spectral library and number of excitatory wavelengths may permit quantification of the redox status of individual mitochondrial electron transport complexes, further enhancing the specificity and utility of this technique.


Study design

The following experiments were designed to examine whether 3RMR can be quantified in vitro and in vivo, and whether a 3RMR threshold exists beyond which cardiac arrest ensues. We compared 3RMR with measures of myocardial function and mortality in several models: An isolated heart model of graded ischemia was used to describe the 3RMR measurements in the absence of blood perfusion; an in vivo inflow occlusion model was used to describe 3RMR changes during abrupt, severe ischemia which included blood perfusion; a hypoxemic ventilation model was used to examine the ability of 3RMR to predict cardiac arrest in the setting of hypoxemia; and, finally, 3RMR was measured on the epicardial surface of swine during aortic cross-clamping to demonstrate the feasibility of measurement in a pediatric-sized heart. Because all animals received the same treatments and there were no treatment groups, there was no randomization or blinding. All experiments were performed in compliance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals.

RRS system

Device description. The device comprises a laser pump source, a source/collection fiber optics probe, and a high-resolution/high-throughput spectrometer. A multimode laser diode was wavelength-stabilized to a single line (441 nm, 4 mW continuous power laser source) using a custom-designed, temperature-controlled external cavity geometry. A fiber optic probe was built that included a 1.5-m-long, jacketed glass fiber bundle to deliver the laser light and collect the Raman signal back to the spectrometer. The laser line was filtered at the distal end to eliminate glass-related Raman and fluorescence artifact and projected onto the tissues, resulting in a 1.5-mm-diameter sample size. The captured Raman signal was filtered to eliminate elastic laser scatter before coupling into the fiber bundle. Spectra were measured by a custom-built spectrometer with a full width at half maximum resolution of 8 cm−1, with an internal, real-time Stokes shift calibration reference (acetaminophen), producing an absolute Stokes shift accuracy of <0.4 cm−1. Cosmic rays were eliminated from CCD pixels. The CCD image sensor was temperature-controlled to maintain constant dark current. The resulting spectra were captured approximately every 1 s and signal-averaged over a rolling 180-s window to create the raw spectrum, y, for analysis (fig. S6).

Creation of RRS library. Excitation at a wavelength of 441 nm allowed us to create clean Raman spectra for heme-containing structures, because these resonance-enhanced spectra are known to be dominated by porphyrin vibrational modes (7) with electronic transitions (π-π*) polarized in the porphyrin plane. This leads to characteristic Raman peaks at 1000 to 1700 cm−1 arising from in-plane stretching vibrations of the ring (46). Furthermore, use of RR the excitation wavelength near the Soret absorption band (400 to 450 nm) results in a dominant RR spectral peak, ν4, between 1350 and 1380 cm−1, which has been established as a spectral marker of oxygenation/deoxygenation or oxidized/reduced state of the iron center of hemoproteins, including cytochromes (44, 4749). We created the RRS spectra for the following hemoproteins found in tissues, verifying their iron atom oxidative state using the ν4 band.

(1) Hemoglobin: Spectra of fresh rodent blood were measured in a cuvette, and oxyhemoglobin saturation, SHbO2, was incrementally varied between <1 and >99% using oxygen and nitrogen mixtures. The RR spectra of oxygenated and deoxygenated blood were scaled to yield a 1:1 linear correlation between saturation values obtained using RRS and co-oximeter (fig. S7).

(2) Myoglobin: Commercially available, high-purity equine myocardial myoglobin extract (Sigma-Aldrich) was prepared as a surrogate for biologically active rodent myoglobin. Oxygen and nitrogen flow and sodium dithionite (Sigma-Aldrich) were used to vary oxymyoglobin saturation (SMbO2) from 0 to 100%, verified by ν4 Raman marker peaks at 1355 and 1376 cm−1.

(3) Mitochondrial cytochromes: Fresh mitochondria were isolated from homogenized rodent (Charles River Laboratories) hearts and purified by a process of serial differential centrifugation (50). Oxidized conditions were established by exposing mitochondria to hyperoxia (100% oxygen flow) in an abundance of substrate [succinate and adenosine 5′-diphosphate (ADP), Sigma-Aldrich] and verified on the basis of the absence of a reduced RR marker peak at 1357 cm−1. Fully reduced conditions were established by exposure to 100% nitrogen with dissolved sodium dithionite and verified by the absence of the oxidized marker peak at 1371 cm−1 within a sealed cuvette.

3RMR was displayed in real time on a user interface (fig. S8). Details of the mathematical regression algorithm can be found in Supplementary Materials and Methods, figs. S9 to 12, and tables S4 and S5.

Langendorff experiments

Sprague-Dawley rats (n = 8, 385 to 565 g, Charles River Laboratories) were anesthetized with 1 to 2% isoflurane, inhaled via a nose cone, and anticoagulated with heparin sodium [1000 U/kg, intraperitoneally (ip); Sagent Pharmaceuticals]. The chest was widely opened, and the heart and lungs were excised en bloc, immersed in ice-cold Krebs-Hensleit buffer [made in-house as previously described (51)], and then retrograde-perfused using an isolated heart system (Harvard Apparatus IH51-B). The left atrium was opened, and a calibrated pressure balloon was placed into the left ventricle for isovolemic developed pressure measurements. The end-diastolic pressure was maintained at 10 mmHg using a microsyringe. A fluorescence-quenching oxygen tension probe (OxyLite, Oxford Optronix) was placed into the ventricular septum via apical puncture to a depth of 2 to 4 mm. This probe contains a needle-based sensor including a tris ruthenium(III) chloride–based fluorophore enclosed in a silicone matrix. Light pulses were generated using a blue light–emitting diode to induce fluorescence from the ruthenium luminophor. The lifetime of the fluorescent pulse is inversely proportional to the oxygen tension at the probe’s tip. The RRS probe was placed 3 to 4 mm from the epicardial surface. Care was taken to clean the RRS probe head before each experiment using acetone. Under constant coronary perfusion pressure (80 mmHg), coronary flow rate was measured by timed collection of effluent. Thereafter, the perfusion was changed to a volume-controlled mode, and 100% of baseline flow was provided for a 30-min equilibration time. Graded ischemia was completed by decreasing the coronary flow rate to 50, 40, 30, 20, 10, and 0% of baseline flow (10 min each). All endpoints were continuously recorded in LabChart 8.0. 3RMR and dP/dTmax were calculated as 1 min medians during the final minute of each 10-min increment and compared over time.

IVC occlusion experiment

Sprague-Dawley rats (n = 8, 520 to 690 g) were induced for anesthesia using ketamine (20 mg) and xylazine (2.5 mg) administered intraperitoneally, and then anaesthetized using inhaled isoflurane (1 to 2%) via a ventilator. Rats were tracheally intubated by direct laryngoscopy and ventilated (CWE, SAR-1000) with positive end-expiratory pressure 5 cm H2O and tidal volumes of 8 to 10 ml/kg. Temperature was maintained at 38°C via a heating plate. FiO2 was maintained at 1.0 throughout the experiment. The chest was widely opened, the pericardium was resected, and the IVC was isolated and snared. After calibration, an arterial pressure catheter (SPR-671, 1.4 French, Millar Inc.) was inserted into the femoral artery via arteriotomy, and a pressure volume conductance catheter (SPR-869, 2 French, Millar Inc.) was placed into the left ventricle via apical puncture. A separate arterial catheter was placed in the contralateral femoral artery for phlebotomy. A fluorescence-quenching oxygen tension probe was placed into the ventricular septum via apical puncture. Finally, the RRS probe was positioned 3 to 4 mm above the epicardial surface. After instrumentation, all the animals underwent a 30-min baseline period. Thereafter, complete IVC occlusion took place by affixing a 5-g weight to the IVC snare for a 10-min observation period. Endpoints included 3RMR, myocardial tPO2, and contractility (dP/dTmax, calculated from the first derivative of the pressure-time curve), each of which was continuously recorded (LabChart 8.0) and exported as Q1 minute medians.

Hypoxic ventilation experiment

Sprague-Dawley rats (n = 31, 480 to 710 g) were anesthetized and instrumented as described above, including a tracheal tube, femoral arterial catheter, and left ventricular pressure-volume conductance catheter. FiO2 was maintained at 1.0 during instrumentation and baseline period, and then titrated to between 0.05 and 0.08 using an FiO2 blender (nitrogen-oxygen blender, CareFusion) for a 30-min experimental period. During this time, acute hemodynamics (including blood pressure, heart rate, and contractility), RRS-based measurements (SHbO2, SMbO2, and 3RMR), and tPO2 were continuously recorded. Arterial blood was sampled at baseline and every 10 min for blood gas analysis and co-oximetry. Venous oxyhemoglobin saturation during the baseline period was assessed by right ventricular puncture. At 30 min of hypoxia, surviving animals underwent acute freeze-clamping of the heart muscle and subsequent immersion in liquid nitrogen for 30 min. [NADH]/[NAD+] and [ATP]/[ADP] ratios in these tissues were measured by fluorometric enzyme assay (Sigma-Aldrich and ThermoFisher, respectively) according to the manufacturers’ instructions and compared between 3RMR assignments. Endpoints included 3RMR, oxyhemoglobin and oxymyoglobin saturations, contractility (dP/dTmax), and the time to cardiac arrest (defined as pulse pressure <5 mmHg). All endpoints were downsampled to 1-min medians, which were compared between 3RMR assignments using generalized estimating equations.

Aortic cross-clamping bypass experiment

Neonatal Yorkshire swine (n = 5, 5 to 12 days, 3.8 to 4.3 kg, chosen to approximate the myocardial thickness of an infant or toddler) were anesthetized using telazol and xylazine, and then anesthetized using inhaled isoflurane (1 to 2%) via an anesthesia machine (Drager Apollo). After neuromuscular blockade and placement of central arterial and venous monitoring catheters, a full median sternotomy and a subtotal thymectomy were performed, and the sternum and pericardium were retracted. After the administration of systemic heparin (75 U/kg, intravenously), the right atrium (18 French) and aortic root (8 French) were cannulated, and cardiopulmonary bypass flows were initiated using a neonatal cardiopulmonary bypass circuit primed with donor blood. Each swine was cooled to 25°C (rectal) over 30 min using a pH stat strategy (52), after which the aortic root was cross-clamped and del Nido cold blood cardioplegia (20 ml/kg) (53) was administered. During this time, total body circulatory arrest (perfusion pump was discontinued and blood was drained from the body) was instituted for 75 min, and hypothermia was maintained using a cooling blanket and ice packs on the myocardium, head, and flanks. Ice was removed for 3 min during RRS measurements at 10, 30, and 75 min. After the ischemic period, the aortic cross clamp was removed, and whole-body perfusion (including myocardium) was restored. The RRS measurements took place over the left ventricle before cooling, during ischemia, and during reperfusion.

Statistical methods

For the ex vivo graded ischemia experiment, tPO2 and 3RMR were compared between CFRs using repeated-measures ANOVA, with Bonferroni correction used to account for the repeated measures. The association between contractility and 3RMR was assessed using a single-phase decay, which fitted the data more closely than did a linear regression line, and the significance was examined using a Spearman’s correlation test. The association between contractility as tPO2 was assessed by linear regression (GraphPad Prism version 7.00), and the significance was assessed using a Pearson’s correlation test. Time points for which all data were not available were excluded. For the IVC occlusion experiment, changes in tPO2 and 3RMR were assessed between two time points by paired t test. Changes in contractility (from baseline to different time points after IVC occlusion) were compared over time using linear trend repeated-measures ANOVA with Greenhouse-Geisser correction. For the hypoxemic ventilation experiment, the relationships between 3RMR, SHbO2, and SMbO2 with tPO2 were assessed using second-order polynomial (quadratic) functions for each relationship, models that were empirically chosen to represent cooperative binding between oxygen and heme, and supported by favorable curve fits. Subsequently, 3RMR, tPO2, and contractility were downsampled to 60-s medians over 30 min and compared between the animals in whom 3RMR exceeded 40% at 10 min of hypoxia and those in whom it did not using repeated-measures ANOVA with a Bonferroni correction for multiple comparisons (IBM SPSS Statistics for Windows version 22.0). Baseline values of each variable were used as covariates in the analysis. Estimated marginal means, along with SE and 95% CI, were calculated. For the aortic cross-clamping experiment in swine, 3RMR, SHbO2, and SMbO2 were compared between time points using repeated-measures ANOVA with Dunnett’s correction to compare each timed value with that during reperfusion. Alpha levels of 0.05 were considered significant for all experiments. Individual subject-level data are reported in table S6.


Materials and Methods

Fig. S1. Schematic of the RRS system.

Fig. S2. Sensitivity and specificity plots of 3RMR to detect severe tissue hypoxia.

Fig. S3. Survival curves over time for rodents during the hypoxia experiment when risk-stratified at 10 min using tissue PO2, tissue oxyhemoglobin saturation, or tissue oxymyoglobin saturation.

Fig. S4. Intracellular ATP and NADH concentrations after hypoxia experiment.

Fig. S5. Oxyhemoglobin and oxymyoglobin saturations in rodents separated by the 3RMR value.

Fig. S6. Raw RR spectra after baseline correction, the best-fit line of the explained spectrum, and the residual.

Fig. S7. Correlation between oxyhemoglobin saturations based on RR spectral analysis and traditional co-oximetry.

Fig. S8. Screenshot of the user interface.

Fig. S9. Quantile-quantile plots of the residual from the in vivo experiments.

Fig. S10. Measured spectra and residua from the in vivo experiments.

Fig. S11. Simultaneous changes in a single 3RMR and oxymyoglobin concentration reading are independent of changes in the total mitochondria–to–total myoglobin ratio.

Fig. S12. Distribution of bootstrap estimates from single–point-in-time measurements during baseline and severe hypoxia.

Table S1. Spectroscopically active components of myocardium.

Table S2. Baseline characteristics of rodents included in the hypoxia experiment.

Table S3. Depth of myocardial tissue sampling within a clinically relevant range of hemoglobin concentration and oxyhemoglobin saturation.

Table S4. Scale factors for hemoglobin, myoglobin, and mitochondria.

Table S5. Confidence limits of SHbO2, SMbO2, and 3RMR from a single instantaneous measurement during baseline and hypoxic conditions in a single animal.

Table S6. Individual subject-level data.

References (5761)


  1. Acknowledgments: We acknowledge O. Shirahai, associate professor at Boston University School of Medicine, for assistance with the mitochondrial isolations; E. Miller, head of the Department of Electrical Engineering at Tufts University, for a critical appraisal of the regression algorithm; New Health Sciences Inc. for assistance with the oxyhemoglobin measurements; and J. Mayer and the perfusion and veterinary staff at Boston Children’s Hospital for assistance with the swine experiments. We also acknowledge K. Ward for his early work and collaboration on the measurement of hemoglobin oxygen saturation using RRS. Funding: This work was supported by the American Heart Association Innovative Research Grant (14IRG18430027), the Department of Defense (DOD) Advanced Technology/Therapeutic Development Award (W81XWH-15-1-0544), the DOD Basic Research Award (W81XWH-11-2-0041), the Smith Family President’s Innovator Award, the Hess Family Philanthropic Fund, and Pendar Technologies. Author contributions: D.V. and P.C. designed and manufactured the device. D.V., K.K., and P.C. developed the spectral model and regression algorithm. P.C., D.V., and L.M.T. built the spectral library. D.A.P., L.M.T., and J.N.K. completed the in vitro experiments. D.A.P., J.W.S., and J.N.K. completed the in vivo experiments. D.A.P., J.W.S., D.V., K.K., P.R., P.C., B.D.P., F.X.M., and J.N.K. processed the data and wrote the manuscript. D.A.P., P.R., D.V., and J.N.K. reviewed and processed all the primary data. All the authors reviewed the manuscript in final form. Competing interests: P.C. is a principal scientist, K.K. is the senior algorithm developer, P.R. is the director of clinical development, and D.V. is the president and CEO of Pendar Technologies. Pendar Technologies holds an exclusive license to U.S. patent 7,113,814 entitled “Tissue Interrogation Spectroscopy.” All the other authors declare that they have no competing interests. Data and materials availability: Please direct requests for materials or information to D.V. (dvakhshoori{at} and J.N.K. (john.kheir{at}

Stay Connected to Science Translational Medicine

Navigate This Article