Label-Free Detection of Molecular Signatures in Heart Failure with Preserved Ejection Fraction Using Raman Micro-Spectroscopy

TL;DR

This study explores using Raman micro-spectroscopy to detect molecular signs of heart failure with preserved ejection fraction, offering a fast and label-free diagnostic method.

Contribution

The study introduces a novel spectroscopic data analysis strategy for HFpEF diagnosis with high classification accuracy.

Findings

Raman micro-spectroscopy achieved 92% classification accuracy in distinguishing HFpEF from control tissues.

The method uses protein-to-tryptophan ratios in cardiac tissue for reliable disease classification.

The approach minimizes preprocessing and enables organ-specific therapeutic response analysis.

Abstract

Heart failure with preserved ejection fraction (HFpEF) is a complex and heterogeneous syndrome characterized by delayed diagnosis and limited therapeutic options, contributing to poor clinical outcomes. In the present study, we investigated the applicability of Raman micro-spectroscopy (RmS) as a label-free, rapid, and cost-effective approach for identifying molecular signatures associated with HFpEF and enabling reliable disease classification. RmS was applied to evaluate disease-related biochemical alterations in cardiac and renal tissues obtained from a clinically relevant HFpEF model (ZSF1 rat). Furthermore, the effects of three pharmacological interventions were analyzed and classified (five experimental groups—36 animals in total), highlighting organ-specific therapeutic responses. We developed a spectroscopic data analysis strategy in which second-derivative Raman spectral features serve as quantitative inputs to a supervised classification model, enabling micro-spectroscopic discrimination of HFpEF versus control tissues and achieving a classification accuracy of 92% (sensitivity 93% and specificity 91%) based on the protein-to-tryptophan ratio in cardiac tissue, while minimizing the need for extensive data preprocessing. The spectroscopic markers used in this study were derived from prior multivariate discovery analyses and are evaluated here within a validation and translational classification framework. Collectively, these findings support the integration of RmS into molecular and translational research settings and suggest its potential utility for improving HFpEF diagnosis and treatment monitoring.

Figures

Full text

1. Introduction

Heart failure (HF), a severe cardiovascular condition, is a major global health burden, stemming from both structural and functional impairments that hinder the heart’s ability to properly fill with or eject blood [1]. Based on ejection fraction (EF), HF can be categorized into three main classes: HF with reduced EF (HFrEF; LVEF < 40%), HF with mid-range EF (HFmrEF; LVEF 40–49%), and HF with preserved EF (HFpEF; LVEF ≥ 50%) [2]. HFpEF arises from a complex interplay of pathophysiological processes, including hypertrophy, cardiac stiffness, and cardiac tissue remodeling. The prevailing view emphasizes a chronic pro-inflammatory state driven by multiple comorbidities, such as diabetes and obesity [3]. Arterial hypertension, which promotes pro-hypertrophic signaling in cardiomyocytes via the renin–aldosterone–angiotensin system (RAAS), can lead to vascular remodeling and endothelial dysfunction. Additionally, high blood pressure is a major factor in the onset of renal failure [4,5]. The interplay between upstream comorbidities such as diabetes, obesity, hypertension, and renal injury creates a vicious cycle, exacerbating the severity of each condition and further worsening the clinical prognosis for HFpEF patients [6,7].

Diagnosing HFpEF is challenging, primarily because its symptoms overlap with those of the associated comorbidities, leading to diagnostic bias. Conventional clinical diagnostic methods face substantial limitations [8], as they are often invasive and only detect HFpEF at advanced stages, when cardiac remodeling—manifested by changes in heart mass, volume, geometry, and stiffness—has already occurred. This underscores the urgent need for early-stage diagnostic tools.

Despite significant efforts to develop treatments for curing or slowing the progression of HFpEF, its pharmacological management remains challenging. Sodium–glucose cotransporter-2 (SGLT2) inhibitors, which reduce renal glucose reabsorption and exert favorable haemodynamic and metabolic effects, are currently the only drug class that has consistently reduced morbidity and mortality in patients with HFpEF [9]. However, the mechanisms underlying their cardioprotective actions are not fully understood and many patients continue to experience symptoms and adverse events despite SGLT2 inhibitor therapy [10]. Consequently, there is a pressing need for innovative diagnostic tools to assess HFpEF-related myocardial and renal alteration to monitor the effectiveness of both established and emerging therapies. In this context, Raman micro-spectroscopy represents a promising approach for elucidating mechanisms of action and for closely tracking therapeutic responses.

Raman micro-spectroscopy (RmS)—known for its sensitivity, specificity, non-destructiveness, and label-free nature—offers a promising alternative to traditional diagnostic methods [11,12]. This study explores the application of RS in the context of cardiovascular diseases, with a particular focus on HFpEF. The primary objectives are to identify distinct spectroscopic features that provide insights into pathophysiological mechanisms at the molecular level, while paving the way for innovative diagnostic and therapeutic strategies.

While previous studies have demonstrated the ability of Raman spectroscopy to detect heart failure-induced chemical alterations in cardiac tissue and have established its feasibility for identifying disease-related spectral fingerprints [13,14,15,16,17,18,19], its application has largely remained descriptive or limited to proof-of-concept analyses. The specific aim of the present work is to extend these earlier findings by correlating RmS analyses of both the heart and kidney, which are directly affected by HFpEF-related dysfunctions [20], to provide a systemic spectroscopic assessment of HFpEF that goes beyond single-organ characterization. The type of information provided by Raman spectroscopy in biological applications is molecular in nature and refers to the chemical composition of the sample in terms of the relative content of molecules or, more often, classes of molecules (such as proteins, lipids, nucleic acids, and carbohydrates), rather than the unambiguous identification of individual molecular species. In this sense, the spectroscopic markers, defined as the intensity ratios of selected Raman bands, were used to capture the biochemical changes accompanying the development of the pathological state. These marker bands were derived from a prior exploratory multivariate Principal Component Analysis (PCA) performed on myocardial tissue, in which PCA loading vectors (see Section SM2 of the Supplementary Materials) identified the spectral regions most strongly associated with HFpEF-related biochemical remodeling [19]. While PCA represents the preferred approach for exploratory discrimination and biochemical interpretation, the present study focuses on the validation and translational robustness of these PCA-derived spectral features through a targeted, threshold-based classification strategy applicable at the individual-animal level. Using RmS combined with statistical analyses, we examined the left ventricle (LV) and both the cortex and medulla of the kidneys. Moreover, the technique was used to distinguish between the effects of different promising treatments for HFpEF, including the linear atrial natriuretic peptide proANP_31–67_, and Entresto, a widely used combination of sacubitril (a neutral endopeptidase inhibitor) and valsartan (an angiotensin II receptor blocker). Although proANP_31–67_, also known as Vastiras (Vas), has demonstrated biocompatibility and preliminary evidence of efficacy in treating HF [21,22,23], it is still under investigation as a therapeutic option for HFpEF. Entresto (En) is recommended as a first-line treatment for HFrEF. However, its effectiveness in treating HFpEF remains debated. In the present study, we used RmS to examine the early effects of proANP_31–67_ alone or in combination with Entresto (En + Vas), in a rodent model of HFpEF. Importantly, the scientific objective was not limited to group-level comparisons but focused on quantitative classification of individual, drug-treated animals as HFpEF-positive or HFpEF-negative, in order to directly assess therapeutic efficacy at the single-animal level. To achieve this, we adopted a targeted, marker-based quantitative strategy, which enables threshold-driven classification of individual animals and allows assessment of treatment-related biochemical modifications, including partial normalization in specific tissues, within a systemic cardiorenal context.

2. Results

Our analysis focused on detecting molecular changes in the left ventricle, kidney cortex, and kidney medulla, with the goal of identifying early biochemical signatures that conventional diagnostic approaches may miss. For the kidney tissue, cortex and medulla were analyzed separately because these regions differ substantially in cellular composition and because region-specific HF-related spectral markers have been observed in other animal models [23]. Tissue sections from the LV were specifically selected because this chamber is the primary site of structural and functional remodeling in HFpEF, as also indicated by our earlier comparative analysis of left and right ventricular alterations, performed only on the control and HF diseased animals [19]. We first verified that the HFpEF model displayed the expected physiological alterations, including elevated blood pressure, impaired diastolic function, and early signs of cardiac and renal remodeling. These baseline measurements confirmed that the animals developed a reproducible HFpEF phenotype suitable for evaluating treatment effects [22].

We then collected Raman spectra from each tissue region and compared profiles across control, HFpEF, Vas-treated, En-treated, and En + Vas groups. This approach allowed us to detect subtle shifts in chemical composition that reflect early pathological changes. While Principal Component Analysis was previously applied to myocardial spectra in this HFpEF model to demonstrate global spectral separation between control and diseased hearts [19], kidney tissues had not been previously investigated. In this study, a marker-based quantitative approach was therefore adopted as a translational validation strategy, enabling classification of individual treated animals and assessment of treatment-related biochemical normalization in both cardiac and renal tissues. Statistical analyses were used to determine which spectral features best distinguished HFpEF tissues from controls and to identify whether either therapy, alone or in combination, partially restored these molecular patterns.

2.1. The Spectroscopic Markers of Cardiac Damage Induced by HFpEF

RmS provided an objective histological assessment of heart tissues in the ZSF1 HFpEF animal model. We previously found several spectroscopic markers—such as lipids, carbohydrates, and glutamate bands—in the cardiac ventricles due to the comorbidities associated with the disease, including obesity and diabetes [19]. Also, abnormal collagen cross-linking and a decrease in tryptophan were observed, both of which were linked to ventricular stiffening and inflammation, conditions favorable for HFpEF development [19].

Based on this earlier analysis, we selected specific marker bands from the Raman spectra (Figure 1) to classify individual rats according to the left ventricular changes induced by HFpEF, such as wall thickening, diastolic dysfunction, and changes in ventricular volume and mass [22]. These marker bands correspond to the spectral regions identified by PCA loading vectors (Figure S5) in the exploratory analysis reported in Ref. [19], and were selected here to enable quantitative, threshold-based classification at the individual-animal level. The intensity ratios of 1655 cm^−1^ to 1577 cm^−1^ (H1), indicative of the protein-to-tryptophan (Trp) concentration ratio [24,25,26], and 1304 cm^−1^ to 1340 cm^−1^ (H2), reflecting the lipid-to-collagen concentration ratio [27,28], were estimated from second-derivative profiles and used as markers for HFpEF positivity in cardiac tissue.

We calculated the two intensity ratios H1 and H2 from the Raman spectra of murine hearts. Figure 1 shows that the mean intensity at 1655 (1304) cm^−1^ is higher, while the intensity at 1577 (1340) cm^−1^ is lower in the spectrum of the tissue from left ventricle of obese rats (LV_Ob) compared to that of lean rats (LV_Ln). Consequently, a larger H1 (H2) is expected due to HFpEF. Other spectral changes were observed when comparing data from obese (Ob) rats to those from lean (Ln), such as the 1000–1150 cm^−1^ region, which is characteristic of carbohydrates and lipids, and the 1450 cm^−1^ peak, assigned to proteins and lipids, as shown in Figure 1 [27,28]. These bands were not considered due to their association with multiple molecular species.

Figure 2a,b illustrate the distributions of H1 and H2 ratios for Ln and Ob Raman spectra. Both parameters exhibit broader distributions in Ob compared to Ln data, with values for obese animals generally being higher than those of lean rats.

We used these distributions to determine the highest F-score, following standard classification methods [29,30], and identified the optimal cut-off value for each ratio.

The data presented in Table 1 indicate that both markers demonstrate high levels of accuracy, sensitivity, and specificity, with the H1 ratio showing the highest accuracy in distinguishing positivity for HFpEF. This is associated with H1 and H2 values exceeding their respective cut-off thresholds.

2.2. The Spectroscopic Markers of Renal Damage Induced by HFpEF

The glomeruli of the renal cortex and the tubules of the renal medulla were studied separately to assess the effects of HFpEF on different functional regions of the renal parenchyma. Unlike the cardiac markers, which were derived from a prior PCA-based exploratory analysis, renal spectroscopic markers were defined here to capture HFpEF-related biochemical alterations in functionally distinct renal regions using the same molecular rationale and quantitative classification framework. Representative mean spectra of the glomeruli of renal cortex and tubules of renal medulla from Ln and Ob rats are presented in Figure 3a,b, along with their corresponding standard deviations.

Two intensity ratios were examined to monitor the most significant spectral changes induced by HFpEF. These ratios were evaluated at 1655 cm^−1^ and 1577 cm^−1^, as well as at 1655 cm^−1^ and 1263 cm^−1^, for both glomeruli and tubules (see Figure 4a–d). The K1 and K2 markers were used to represent the protein-to-Trp ratio and the total protein-to-non-collagenous protein concentration ratio, respectively [24].

Based on the previous assignment, the K1 intensity ratio was indicative of the concentration ratio between proteins and Trp. In contrast, the K2 marker reflected the concentration ratio of total proteins to non-collagenous proteins.

Data presented in Figure 4 and Table 2 and Table 3 indicate that the two spectral markers demonstrate higher accuracy in classifying the effects of HFpEF on the medullary region of the renal parenchyma. Notably, the cut-off values of corresponding markers (M_K1 with respect to C_K1, and M_K2 compared to C_K2) are not significantly different. The highest levels of accuracy, sensitivity, and specificity are achieved specifically with the M_K1 ratio. The low accuracy of C_K1 and C_K2 parameters reflects the large overlap of Ln and Ob distributions shown in Figure 4c,d.

We analyzed heart and kidney tissues from the entire set of ZSF1 rats (Ln, Ob, and treated animals) and evaluated the median value of each spectral marker for each rat. We discovered a linear correlation between H1 and H2, as well as between K1 and K2. This correlation was also observed in data from treated rats, although the spectroscopic markers were identified by comparing spectra from only Ln and Ob rats. Figure 5 presents these findings, showing that the proteins/Trp concentration ratio (estimated by H1 in the heart and K1 in the kidneys) spans different ranges depending on the tissue type (cardiac or renal). Similar values are obtained for specific functional units (medulla or cortex) of the kidney. Interestingly, no cross-correlations were observed between cardiac and renal markers, suggesting that the chemical changes in cardiac tissue occurred through different mechanisms compared to those in the renal tissue.

2.3. Classification of Cardiorenal Damages in Drug-Treated Samples

The most accurate markers described above were used to classify cardiorenal damages, with reference to heart tissue and both the medullar and cortical areas of kidneys. This classification of individual rats was effectively applied to rationalize the effects of pharmacological treatments, and thus the median values of M_K1 and H1 markers from spectra of drug-treated rats were used to recognize HFpEF positivity in cardiac and renal tissues. Drug-treated rats were divided into three groups: one group for those treated with Vastiras (Ob + Vas), one group for those treated with Entresto (Ob + En) and one group for rats treated with both Entresto and Vastiras (Ob + En & Vas).

Figure 6 presents a four-panel classification of data from each rat based on the composition of heart and kidney tissues. The panels are divided according to the cut-off values of H1 and M_K1 on the left, and H1 and C_K1 on the right. The upper-right panel (pink area) highlights HFpEF positivity in both organs, while the bottom-left panel (green area) indicates HFpEF negativity for both the heart and kidney. The white areas in the upper-left and bottom-right panels represent cases of only renal damage and only cardiac damage, respectively. Most of the data points fall within a range of HFpEF negativity in renal tissue, while still showing HFpEF positivity in cardiac tissue.

A few exceptions, such as EZ6, DZ4, and DZ8, deviate from this trend, suggesting potential individual variability in treatment response. These cases may indicate either incomplete recovery of kidney function or a partial therapeutic effect on the heart and warrant further investigation.

3. Discussion

In the present study, RmS has been successfully employed to enable precise assessment of HFpEF, demonstrating the potential of this experimental approach to provide rapid, reliable, and label-free molecular-level information for tissue histopathology related to HFpEF phenotypes. Spectroscopic markers revealed molecular changes associated with HFpEF and its comorbidities in both organs, enabling the classification of individual rats based on alterations in cardiac and/or renal chemical compositions. This novel classification method of organ damage demonstrates high accuracy, specificity, and sensitivity.

Analysis of heart tissue sections revealed that the most effective indicator of the presence of HFpEF (marker H1) is associated with a reduction in Trp content, reflecting the tissue’s inflammatory state [19,31]. This was indicated by signal intensity at 1577 cm^−1^. Additionally, conformational rearrangement of collagen, part of the cross-linking phenomenon, was evidenced by the narrowing of the amide I band and the increased intensity at 1655 cm^−1^. These spectral variations were clearly detected by RmS, which demonstrated greater sensitivity than conventional histological analyses, the latter revealing only a slight increase in interstitial fibrosis in Ob samples compared to Ln [19]. A similar effect was observed in renal tissue, where the marker K1 provided the highest accuracy in indicating the effects of HFpEF, particularly in the medullary region.

Despite the significant changes in tissue composition caused by comorbidities —evidenced by spectroscopic findings of increased lipid and carbohydrate concentrations in samples from Ob rats [19]—the ability of vibrational spectroscopy to simultaneously monitor all major tissue components without additives provides a significant advantage in understanding the complex clinical presentations observed in this animal model.

The results of the present work differ significantly from those previously observed in the DSS model of HFpEF [18], where spectroscopic markers in cardiac tissue were primarily associated with abnormal collagen deposition, highlighting interstitial fibrosis, and with the presence of microcalcifications indicative of advanced pathology. Moreover, HFpEF-induced alterations in renal tissue were more pronounced in the cortex rather than the medullary region of the kidney [23].

This underscores the variability in pathological manifestations depending on the HFpEF model used. The differential deposition of collagen and the specific localization of tissue alterations suggest that the underlying mechanisms of HFpEF may vary, necessitating model-specific diagnostic and therapeutic strategies. Such differences emphasize the need for a more nuanced understanding of HFpEF’s pathophysiology. Tailored interventions may need to be developed depending on the unique comorbidities and features present.

We found that the protein-to-Trp spectroscopic marker (H1 = K1) is the only one shared between heart and kidney tissues. In contrast, the H2 and K2 markers appear to be more tissue-specific, capturing the distinct effects induced by HFpEF in these organs. The K2 marker reflects changes in the protein composition of renal tissue, while the H2 marker is associated with an increase in lipid content in samples from obese animals. The identification of both shared and tissue-specific markers may reflect different trajectories of organ involvement in HFpEF. The shared H1/K1 marker likely reflects systemic inflammation, while tissue-specific changes may arise from local metabolic or structural remodeling processes. These findings support the development of spectroscopic marker panels capable of detecting disease presence and localizing its effects.

The ability to identify and compare markers within and across different tissues, both qualitatively and quantitatively, arose from the application of a straightforward classification method for spectroscopic results. This method enabled the determination of the optimal cut-off value for each parameter, along with its associated accuracy, sensitivity, and specificity.

A notable advantage of this approach lies in its simplicity. Importantly, this simplicity does not replace upstream multivariate discovery approaches, but rather represents a deliberate downstream choice aimed at improving robustness, reproducibility, and applicability at the individual-sample level. Estimating these parameters requires minimal processing of spectroscopic data—only derivatization of spectral profiles is necessary. This step eliminates the need for background signal subtraction, and intensity ratios can be calculated without normalization. Such simplicity reduces the number of processing steps, minimizing potential errors and enhancing overall efficiency. By avoiding complex preprocessing stages, the method not only saves time but also ensures greater consistency and reproducibility, making it particularly well-suited for high-throughput analyses and clinical diagnostic applications.

The strength of this targeted, marker-based statistical classification method lies in its ability to effectively distinguish between groups while optimizing the balance between sensitivity and specificity. Rather than targeting absolute concentrations of individual biomarkers, which would require dedicated biochemical assays, our approach leverages relative, multivariate spectral signatures that capture site-specific tissue remodeling processes. The identified cut-off value facilitates the classification of individuals based on chemical changes induced by HFpEF. This straightforward approach achieves robust classification without the need for advanced machine learning models, making it particularly suitable for use by those who are not specialists in spectroscopic techniques. This approach complements multivariate exploratory analyses by enabling prospective classification without the need to re-derive dataset-specific models.

We used this method to classify rats subjected to three pharmacological treatments (Vas, En, En & Vas) and to investigate their effects on normalizing cardiac and renal tissue composition. The ability to examine drug action in situ, with a focus on specific functional structures within the organ—such as the kidney—proves particularly valuable, as it can help clarify the treatment’s mechanism of action.

In the previous experiment on DSS rats, treatment with Vastiras had already demonstrated its efficacy in mitigating the effects of HFpEF and normalizing renal tissue composition [23]. With respect to the ZSF1 model, the graphs in Figure 6 indicate that all three treatments yield similar effects. Specifically, they are effective in normalizing the conditions of damaged kidneys but fail to fully restore the altered composition of heart tissue. This pattern suggests that, while pharmacological treatments successfully reverse renal damage associated with HFpEF, they are less effective in addressing cardiac alterations.

This discrepancy underscores the organ-specific complexities of HFpEF, where kidneys appear to respond more readily to treatment than the heart. The clustering of data in regions indicative of HFpEF negativity in renal tissue, juxtaposed with persistent positivity in cardiac tissue, also suggests possible differences in drug bioavailability, target engagement, or regenerative capacity between the organs. The kidney’s enhanced responsiveness may be due to its higher vascularization and faster turnover, which makes it more amenable to molecular normalization. These insights highlight the urgent need for more targeted therapies to specifically address cardiac dysfunction, even when there is a significant renal improvement. The proposed methodology is not intrinsically limited to HFpEF and, in principle, can be applied to any heart failure phenotypes, included HFrEF, and HFmrEF.

4. Materials and Methods

4.1. Study Protocol

The study was conducted according to the Norwegian Animal Welfare Act, which conforms with the “European Directive 2010/63/EU on the protection of animals used for scientific purposes” and was approved by the Norwegian Food Safety Authority (FOTS approval #15886; approval date: 19 June 2018). The detailed protocol and all cardiac function-related, histology, and biochemical analysis data have been previously reported [19]. Briefly, four-week-old male ZSF1 rats were divided into five groups: eight lean rats (Ln), eight obese (Ob), eight obese and treated with pro-ANP_31–67_ (Ob + Vas), six obese and treated with Entresto (Ob + En), and six obese and treated with a combination of both Entresto and pro-ANP_31–67_ (Ob + En & Vas). Pharmacological treatments were administered by subcutaneous (s.c.) infusion of drugs via an Alzet osmotic mini pump. Till the age of 25 weeks, animals were tracked using echocardiography for HFpEF development. Power analysis was performed using LV end systolic volume based on previously published data [22]. At the study endpoint, animals were humanely euthanized under deep anesthesia with 5% isoflurane, in accordance with the 2020 Guidelines of the American Veterinary Medical Association (AVMA) for the Euthanasia of Animals. Following euthanasia, organs were collected for subsequent biochemical and histological analyses [19]. The current study serves as a post hoc analysis of kidney and heart tissues using spectroscopic methods. Murine cardiac ventricular sections were snap-frozen after harvesting for analysis. Kidney samples were fixed in 10% formalin for 24 h at room temperature followed by dehydration through graded ethanol (70%, 80%, 95%, and 100%; 30–60 min per step), prior to paraffin infiltration. Then, sections (10 µm thickness) were deparaffinized by consecutive xylene washes before Raman micro-spectroscopic analysis.

4.2. Samples for Spectroscopic Analyses

Raman spectroscopy was employed to examine ventricular sections from six replicates of both Ln and Ob tissues. Given that heart tissue remains spatially uniform at the resolution achieved in these experiments, the average spectra representing different tissue types (Ln, Ob, Ob + Vas, Ob + En and Ob + En & Vas) were compared. This assumption is supported by the low spectral variance observed across sampling points in the current dataset, as reflected by the small standard deviation around the mean spectra in Figure 1. Renal sections from the same samples were analyzed by RmS after being deparaffinized.

4.3. Raman Measurements

Raman scattering spectra were obtained using an Olympus IX73 inverted confocal microscope connected to the MonoVista CRS+ spectrometer from S&I Spectroscopy & Imaging GmbH (Warstein, Germany), directly on snap-frozen cardiac tissue sections. For each anatomical region examined in this study—namely, the myocardial region of the left ventricle, glomeruli of the renal cortex, and tubules of the renal medulla—at least 20 randomly selected point measurements were performed. This procedure was repeated for each sample, resulting in approximately 150–170 spectra per tissue region for each experimental group (Ln, Ob, Ob + Vas, Ob + En, Ob + En & Vas). For both cardiac and kidney samples, measurements were conducted in backscattering mode using a 785 nm incident laser wavelength at a power of 13.6 mW and 10x objective lens. Each spectrum was generated by averaging 60 accumulations, with a 10 s integration time per accumulation, covering the spectral range of 400 cm^−1^ to 1942 cm^−1^ and achieving a spectral resolution of approximately 2 cm^−1^.

4.4. Data Processing and Statistical Analysis

An exploratory multivariate PCA was previously applied to myocardial Raman spectra from this HFpEF model to identify disease-associated spectral features and biochemical signatures [19]. In the present study, PCA was not recalculated; instead, a targeted, marker-based quantitative strategy was intentionally adopted as a downstream validation and translational approach aimed at individual-level classification. All subsequent processing steps were conducted in OriginPro 2023 software from OriginLab (Northampton, MA, USA). After obtaining the RmS spectra, a frequency calibration was performed, setting the intense phenylalanine peak at 1003 cm^−1^. Second-derivative spectra were obtained after using a second-order polynomial with a 9-point Savitsky–Golay smoothing filter. Spectra were processed using the second derivative primarily to accelerate analysis: this approach reduces the time required to process each spectrum from minutes (needed for baseline subtraction and deconvolution) to a fraction of a second, which is particularly important given that each sample group contains hundreds of spectra. Derivative-based metrics can provide quantitative results comparable to conventional peak-fitting approaches while substantially simplifying preprocessing workflows. This simplification of preprocessing does not replace multivariate exploratory analyses such as PCA, but rather supports the implementation of a robust and computationally efficient marker-based framework suitable for prospective and high-throughput applications. In the present study, we directly compared second-derivative intensities with conventional band-area analysis, and the results—demonstrating strong agreement between the two approaches—are reported in Section SM1 of the Supplementary Materials.

The distribution of values for H1, H2, K1, and K2 spectral markers (absolute intensity of second-derivative profiles at the selected frequencies) was derived from spectra of Ln and Ob rats, allowing the determination of true positive, false negative, true negative, and false positive fractions for each marker. These were used to estimate accuracy, sensitivity, specificity, and F-score data corresponding to various thresholds between the Ln and Ob distributions. F-score value was determined as the parameter to evaluate the model performances using the standard formula F = (2∙sensitivity∙specificity)/(sensitivity + specificity) [29,30]. The highest F-score value identified the optimal cut-off for classifying HFpEF positivity. These spectral markers were also evaluated from the spectra of treated rats. For each animal, the median value of H1, H2, K1, and K2 was calculated to assess the effects of pharmacological treatments in relation to recovery from cardiac and/or renal damage.

5. Conclusions

Raman investigations of renal and cardiac tissues in the context of HFpEF have provided valuable insights into the intricate biochemical alterations associated with this unique heart failure phenotype. The versatility and sensitivity of this spectroscopic technique reveal chemical signatures specific to morphological and functional structures of interest in biological samples. This generates a deeper understanding of HFpEF pathophysiology at the molecular level. The spectroscopic markers employed here originate from prior multivariate discovery analyses and are evaluated in this work within a simplified, translational framework aimed at individual-level disease classification and treatment monitoring.

This approach not only enhances the diagnosis of complex pathologies like HFpEF but also aids in evaluating therapeutic responses to drug treatments across different morphological structures. The experimental workflow can be further optimized by reducing spectrum acquisition time through the use of thinner tissue sections. With its simplicity in biopsy analysis and minimal data pre-processing requirements, vibrational spectroscopy stands out as a promising tool in biomedical research, with considerable potential for clinical applications.

While the present study demonstrates RmS efficacy in an animal model, future work must focus on validation in human tissues to assess clinical translatability. Raman spectroscopy has already been explored in human oncology and neurodegenerative disease diagnostics, where its label-free and minimally invasive nature makes it ideal for intraoperative or bedside applications. A similar translational pipeline could be envisioned for HFpEF, beginning with ex vivo analysis of human biopsy samples, followed by development of Raman-compatible diagnostic devices for clinical pathology labs. RmS can be seamlessly integrated into diagnostic laboratories, complementing existing clinical workflows due to its affordability and straightforward operation. Furthermore, specialized technicians can be trained to identify spectral markers, provided that sufficient experimentation with human samples is conducted.