Lung microvascular rarefaction impairs pulmonary gas exchange and exacerbates heart failure with preserved ejection fraction

61

Dyspnea and exercise intolerance are the primary clinical symptoms of heart 62 failure. Heart failure patients experience frequent hypoxemic episodes, yet underlying 63 mechanisms and relevance remain poorly understood. In a cohort of heart failure patients and 64 multiple animal models, we identify pulmonary capillary rarefaction driven by excessive 65 autophagy in endothelial cells as a novel mechanism of hypoxemia and cardiac disease 66 progression. 67

A cohort of heart failure with preserved ejection fraction ( HFpEF) patients was 68 analyzed for parameters of left ventricular (LV) dysfunction and pulmonary gas exchange. 69 Morphological and cellular mechanisms of impaired pulmonary oxygenation were assessed in 70 three animal models of heart failure, namely two HFpEF mod els, SU5416 -treated ZSF1 71 obese rats and high fat diet /L-NAME treated mice, and in rats subjected to aortic banding. 72 Lung microvascular rarefaction was quantified by micro -computed tomography, stereology, 73 flow cytometry and dye efflux. Cellular mechanisms o f capillary loss were analyzed by 74 single-cell transcriptomics, electron microscopy and immunofluorescence, and in mice with 75 endothelial-specific deletion of the autophagy gene Atg7 (Atg7EN-KO). 76

In 234 HFpEF patients, advancing NYHA class was associated with progressive 77 worsening of arterial oxygen saturation at rest and during exercise and a reduced lung 78 diffusing capacity. Impaired gas diffusion correlated with indices of LV diastolic dysfunction. 79 Impaired oxygenation and reduced exercise capacity were similarly evident in animal models 80 of left heart disease, which showed a distinct loss of pulmonary microvessels and capillaries. 81 Lung microvascular endothelial cells in HFpEF showed characteristics of increas ed 82 autophagic flux and apoptosis. Relative to their wild type HFpEF controls, Atg7EN-KO mice 83 had less capillary loss, restored normoxemia, improved exercise tolerance, and mitigated LV 84 diastolic dysfunction. Additional studies in HFpEF mice corroborated th e functional 85 relevance of impaired gas exchange for the progression of left heart disease by demonstrating 86 that additional hypoxia aggravated, whereas moderate hyperoxia improved LV function. 87

Our findings identify pulmonary microvascular rarefa ction as a novel 88 pathomechanism in heart failure that i) contributes to dyspnea and exercise intolerance, ii) 89 impairs pulmonary gas exchange and iii) accelerates LV disease progression. Strategies 90 targeting this axis such as moderate oxygen therapy may mitigate cardiopulmonary morbidity 91 in heart failure. 92 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint Clinical Trial Registration: Registered in the DRKS (Deutsche s Register für klinische 93 Studien) as trial# DRKS00032974 at https://drks.de/search/en/trial/DRKS00032974. 94 Key words: heart failure, HFpEF, gas exchange, capillary rarefaction, autophagy, apoptosis, 95 hypoxia, hyperoxia 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint

120 Heart failure with preserved ejection fraction (HFpEF), defined by a preserved left 121 ventricular (LV) ejection fraction (LVEF) ≥ 50%, accounts for >50% of all heart failure cases 122 globally1. It is characterized by diastolic LV dysfunction, increased LV myocardial stiffness 123 and abnormal ventricular-vascular coupling. HFpEF is commonly associated with and driven 124 by characteristic comorbidities including hypertension, metabolic syndrome (obesity, diabetes 125 mellitus, dyslipidemia ), chronic kidney disease as well as aging 2. In spite of the recent 126

of SGLT2 inhibitors as the first approved clinical therapy for HFpEF 127 (EMPEROR-preserved trial, DELIVER trial), morbidity and mortality remain high , stressing 128 a major unmet clinical need3,4. 129 Besides fatigue and impaired exercise capacity, dyspnea on exertion or even at rest is the 130 cardinal clinical symptom in HFpEF patients. Limitations in exercise capacity can be 131 evidently attributed to a reduced ability to increase cardiac output in HFpEF patients , which 132 is often normal at rest 5. Yet, dyspnea is not a direct consequence of a reduced cardiac reserve 133 but rather a sign of impaired arterial blood oxygenation and/or impaired lung inflation as 134 measured by arterial chemosensors and pulmonary stretch sensors, respectively. Indeed, 135 several clinical studies show evidence of systemic hypoxemia in HFpEF patients. 136 Specifically, oxygen desaturations with SaO2 values <90% are 3 times more frequent and last 137 5-6 times longer in HFpEF patients as compared to healthy controls 6. In a study of 539 138 HFpEF patients undergoing invasive cardiopulmonary exercise testing, exertional hypoxemia 139 with SaO 2 <94% was observed in 25% of patients, and particularly prominent in older and 140 more obese patients 7. Further r egression analyses revealed a direct relationship between 141 increased pulmonary capillary pressure and lower arterial oxygen tension, especially during 142 exercise7. Importantly, impaired arterial oxygenation cannot be attributed to LV-failure 143 associated hemodynamic changes per se, as reduced cardiac output and the resulting lower 144 blood flow velocity across the pulmonary microvasculature will improve , rather than impair, 145 oxygen uptake, while pulmonary congestion increase s vascular surface area and thereby 146 facilitates enhanced diffusion. Rather, systemic hypoxemia likely reflects impaired oxygen 147 uptake in the lung. Consistent with this view, clinical st udies show a marked increase in the 148 alveolar-arterial oxygen difference in patients with HFpEF 7 or acute heart failure 8, 149 respectively. Similarly, the lung´s diffusing capacity for carbon monoxide (DLCO) is 150 characteristically reduced in HFpEF patients 9,10 suggesting impaired alveolo -capillary gas 151 exchange. Clinical HFpEF is frequently associated with extensive vascular remodeling in 152 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint both pulmonary arteries and veins 11,12, yet, HFpEF -associated structural changes at the level 153 of the alveolo -capillary interface as the primary functional unit of pulmonary gas exchange 154 remain largely unclear. In the present study, we hypothesized that HFpEF is not only 155 associated with pulmonary arterial and venous remodeling, but also with extensive structural 156 remodeling of pulmonary microvessels resulting in reduced alveolo -capillary gas exchange. 157 The ensuing hypoxemia will in turn aggravate right ventricular (RV) and LV dysfunction, 158 thereby propagating HFpEF deterioration in a positive feedback loop. 159 160

161 Patient data 162 234 patients with HFpEF were prospectively recruited into the clinical cohort of the 163 Collaborative Research Center 1470 (CRC1470), a single-center consortium supported by the 164 German Research Foundation (DFG, Deutsche Forschungsgemeinschaft). The study 165 complied with the Declaration of Helsinki and was approved by the Ethics Committee of 166 Charité – Universitätsmedizin Berlin (Berlin, Germany; approval number EA1/224/21). All 167 participants were older than 18 years and provided written informed consent. A positiv e 168 diagnosis of HFpEF according to the current American College of Cardiology (ACC) criteria 169 was required for inclusion in the cohort, and defined as follows: (i) symptoms and/or signs of 170 heart failure corresponding to at least New York Heart Association (N YHA) class II prior to 171 medication; (ii) a left ventricular ejection fraction (LVEF) exceeding 50%; (iii) objective 172 evidence of cardiac structural and/or functional abnormalities indicative of diastolic 173 dysfunction or increased left ventricular filling pres sures. Structural abnormalities were 174 defined as a left ventricular mass index (LVMI) > 95 g/m² in females or > 115 g/m² in males, 175 a relative wall thickness (RWT) > 0.42, or a left atrial volume index (LAVI) > 34 mL/m². 176 Functional abnormalities included an E/e′ ratio at rest > 13 or a pulmonary artery systolic 177 pressure (PASP) > 35 mmHg. Inclusion further required plasma levels of N-terminal pro–B-178 type natriuretic peptide (NT -proBNP) > 125 pg/mL. Exclusion criteria included a life 179 expectancy of less than one year, a history of heart failure with mid -range or reduced ejection 180 fraction, acute coronary syndrome within the preceding 30 days, cardiac surgery within the 181 past three months, or end -stage re nal disease with or without hemodialysis. Further 182 exclusions comprised severe valvular heart disease, hypertrophic cardiomyopathy, 183 amyloidosis, congenital heart disease, sarcoidosis, and constrictive pericarditis. Patients with 184 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint extracardiac disorders that could account for their symptoms, such as chronic obstructive 185 pulmonary disease (COPD) GOLD stage > 2, pulmonary arterial hypertension, or moderate 186 to severe anemia (hemoglobin < 10 g/dL in males and < 9.5 g/dL in females), were also 187 excluded. All particip ants underwent a comprehensive clinical assessment, including two -188 dimensional echocardiography (2DE) of both ventricles, spiroergometry, diffusion single 189 breath analysis, and standard blood work. 190 Animals 191 All animal experiments were performed in compliance with the Directive 2010/63/EU of the 192 European Parliament, the German Animal Welfare Act, and the Guide for the Care and Use of 193 Laboratory Animals (NIH Publication No. 85 -23, revised 1985). Experimental procedures 194 were approved by the local regulatory autho rity (Landesamt für Gesundheit und Soziales, 195 Berlin, Germany; protocol numbers G0008/22, G0025/24, G0030/18 and G0085/23) and 196 adhered to the Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines. 197 All animals were housed for at least 1 week to acclimatize to laboratory conditions before 198 starting experimental procedures. Two to four animal s were housed per cage, divided by sex 199 and treatment. Animals were kept in a 12-h:12-h light/dark cycle with free access to food and 200 water in individually v entilated cages and pathogen -free rooms. All animals underwent 201 weekly body weight and daily health monitoring. 202 C57BL/6J mice and Sprague-Dawley rats were purchased from Janvier -Labs. ZSF1 rats were 203 purchased from Charles River Laboratories. Mice with an endothelial cell (EC) -specific 204 deletion of the essential autophagy gene Atg7 were generated by crossing Atg7flox/- mice with 205 mice constitutively expressing Cre recombinase under control of the VE -cadherin promoter 206 (Cdh5-Cretg/+). Atg7flox/flox; Cdh5-Cretg/+ mice were used as EC -specific Atg7 knockout mice 207 (Atg7EN-KO), with Atg7flox/flox; Cdh5-Cre+/+ (Atg7EN-WT) mice serving as corresponding 208 controls. Ear biopsies were collected for genotyping by PCR. Animals were randomly 209 assigned to either control, heart failure, or intervention groups specified below. 210 Murine HFpEF model. Eight week-old C57BL/6J mice of both sexes were divided into four 211 experimental groups: i) naive mice receiving standard diet (chow, 9% fat, Ssniff 212 Spezialdiäten GmbH, Soest, Germany) and regular drinking water serving as controls, ii) 213 mice receiving a rodent diet with 60% fat (high -fat diet - HFD, Ssniff Spezialdiäten GmbH, 214 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint Soest, Germany) and N[ ω]-nitro-l-arginine methyl ester (L -NAME, 0.5 g/L, N5751 -10G, 215 Sigma-Aldrich) for t welve weeks to induce HFpEF 13, iii) mice receiving HFD, L -NAME 216 diluted in drinking water and additional exposure to chronic hypoxia (10% O2) for two weeks 217 from week 10 to assess the effects of hypoxemia on LV function, iv) mice receiving HFD, L-218 NAME diluted in drinking water and additional exposure to moderate hyperoxia (40% O 2) 219 for two weeks from week 10 to restore normoxemia in HFpEF mice and assess effects on RV 220 and LV function. All groups were age- and sex-matched and run in parallel. 221 Rat HFpEF model. To induce HFpEF in rats, male obese ZSF1 rats (8 - to 10-weeks of age) 222 received a single 100 mg/kg subcutaneous injection of the vascular endothelial growth factor 223 receptor-2 antagonist Sugen (Semaxinib, SU5416, MedChemExpress, HY-10374, Monmouth 224 Junction, NJ) dissolved in CMC buffer (0.5% sodium carboxymethyl cellulose, 0.4% 225 polysorbate 80, 0.9% sodium chloride, and 0.9% benzyl alcohol) as described 14. Lean ZSF-1 226 rats served as controls. Endpoint analyses were performed after 14 weeks. 227 Rat heart failure mod el. As additional model, we induced left h eart failure in 4 - to 8-weeks 228 old male Sprague–Dawley (SD) rats ( ∼100 g body weight) by aortic banding (AoB) surgery 229 under ketamine/xylazine (87/13 mg/kg body weight) anesthesia as previously described 15,16. 230 In brief, a metal clip with an open inner diameter of 0.8 mm was surgically implanted across 231 the aorta just after the coronary arteries branch off. Sham animals were subjected to the same 232 surgical procedure without clip placement. Endpoint examinations were p erformed at weeks 233 3 and 9 after surgery. Researchers performing endpoint and post -mortem analyses were 234 blinded for groups and treatment. 235 Treadmill exercise 236 Mice were initially acclimatized to a treadmill (CaloTreadmill, TSE Systems, Germany) by 237 running uphill (20°) at a constant speed (5 m/min) for 5 min for 3 consecutive days. After 238 acclimatization, an exhaustion test was performed by running uphill (20°) on the treadmill at 239 warm-up speed (5 m/min) for 4 min and then increasing the speed to 14 m/min for 2 min. 240 Every 2 min thereafter, the speed was increased by 2 m/min until the animal was exhausted. 241 Exhaustion was defined by two separate criteria: (i) the animal was exposed to the shock grid 242 at the back of the treadmill for ≥10 seconds or (ii) the animal sp ent less than 2 s on the 243 treadmill before contacting the shock grid again (i.e. lack of motivation). Running distance 244 and the maximum rate of oxygen consumption (V̇ O2 max) were assessed using PhenoMaster 245 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint software V7.1.6 (TSE Systems, Germany). Arterial oxygen saturation (SaO ₂) was measured 246 non-invasively immediately before and after the exercise protocol using a MouseOx Plus 247 collar sensor system (STARR Life Sciences, Oakmont, PA, USA). 248 Echocardiography 249 Progressive cardiac dysfunction was monitored by non -invasive small animal 250 echocardiography according to the standards defined by the American Society of 251 Echocardiography (ASE). To this end, animals were anesthetized with 1 -2% (mice) / 1.5 -3% 252 (rats) isoflurane inhalation anesthesia. Eye ointment was applied to protect the animals´ eyes 253 from dryi ng. Echocardiographic imaging was performed using ultrahigh frequency linear 254 array transducers (MX400, 20 –46 MHz for mice; MX250, 15–30 MHz for rats ) on a Vevo® 255 3100 high-resolution imaging system (FUJIFILM VisualSonics Inc., Toronto, ONT, Canada). 256 In brief, animals were fixed in supine position on a heated table and the thorax was depilated 257 (Veet depilatory cream and – for rats – shaving). The heart was visualized parasternally in the 258 short axis, the long axis and in apical four-chamber view. In addition, measurements were 259 taken in brightness (B)- and motion (M)-Mode (short axis and long axis) to calculate the LV 260 and RV chamber dimensions at the level of the papillary muscles. Doppler imaging was 261 recorded for the LV and the pulmonary artery, and speckle tra cking echocardiography was 262 performed for LV strain analyses. All images were stored as raw data in DICOM format for 263 offline analysis. 264 Image analysis was performed using the VevoLAB Version 5.7.1 (FUJIFILM VisualSonics). 265 Ejection fraction, stroke volume, en d-diastolic volume, and heart rate were determined from 266 B-mode images of the LV in the parasternal long -axis view using the LV Trace tool. End -267 diastolic diameter was obtained from M -mode images in the same view with the semi -268 automatic LV Trace function. Two -dimensional speckle -tracking echocardiography for 269 assessing global longitudina l peak strain (GLS) was carried out using VevoStrain 2.0 270 software (FUJIFILM VisualSonics) and a speckle -tracking algorithm. Semi-automated 271 endocardial border tracing was carried out on three consecutive cardiac cycles per animal, 272 selected from a B -mode cine loop obtained in the parasternal long -axis view at end -systole. 273 B-mode images were chosen based on frame rates above 200 fps and clear visualization of 274 endocardial and epicardial LV borders. Peak GLS values were derived from six independent 275 anatomical LV segments. 276 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 277 Invasive hemodynamics 278 Mice were anesthetized by intraperitoneal (i.p.) injection of ketamine (120 mg/kg body 279 weight) and xylazine (16 mg/kg body weight), and neck and chest were depilated as 280 described above. Mice were then placed on a specially prepared, sterile operating table 281 equipped with a heating mat. After checking the interdigital reflex to ensure a sufficient depth 282 of anesthesia, the jugular vein and the carotid artery were sequentially catheterized with a 1F 283 microtip Millar catheter and the catheter was advanced into the RV and LV for measurement 284 of RV systolic pressure (RVSP) and LV systolic pressure (LVSP), respectively. 285 Rats were anaesthetized by i.p. ketamine (87 mg/kg body weight) a nd xylazine (13 mg/kg 286 body weight), neck and chest were depilated as described above, and animals were placed on 287 a sterile operating table as reported for mice. After ensuring sufficient depth of anesthesia by 288 checking the inter-phalangeal reflex, rats were tracheotomized and mechanically ventilated as 289 follows: The trachea was visualized with a cervical midline incision and cannulated. Lungs 290 were perioperatively ventilated at a breathing rate of 90 breaths/min and a tidal volume of 8.5 291 mL/kg body weight. Fo llowing thoracotomy by vertical sternotomy, cardiac catheterization 292 was performed directly through the apex of the heart, and LVSP and RVSP were recorded via 293 a Millar catheter positioned in the left and right ventricle, respectively. Stable pressure 294 profiles were recorded for at least 2 min by PowerLab 4/35 connected to LabChart v. 8 295 software (ADInstruments, Sydney, Australia) and analyzed off -line by an independent 296 observer blinded to the respective protocol. LVSP and RVSP were calculated as the mean 297 systolic pressure measured over 10 s in hemodynamically stable recordings. Arterial and 298 venous blood samples were obtained for blood gas analyses. Animals were then euthanized 299 under deep anesthesia by exsanguination, and blood and tissue samples were collected. 300 Gravimetry 301 Following euthanasia, the heart was excised after perfusion through the left ventricle with 10 302 mL ice-cold 1X PBS, the ventricles were separated from the atria and dissected into RV and 303 LV including septum (LV+S). RV and (LV+S) weight were determined by gravimetry and 304 normalized to body weight (for rats) and tibia length (for mice), respectively. 305 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 306 Flow cytometry 307 Lung tissue was subjected to enzymatic digestion using collagenase I (450 U/mL), 308 collagenase XI (60 U/mL), DNAse I and hyaluronidase (60 U/mL) (Sigma -Aldrich) buffered 309 in HEPES buffer (Corning) an d 1x PBS for 30 min at 37°C and 700 rpm . Cells were filtered 310 through a 40-µm nylon mesh for mouse lungs and a 70-µm nylon mesh for rat lungs, washed, 311 and centrifuged (5 min, 350xg, 4°C) to obtain single -cell suspensions. Red blood cells were 312 lysed with 1x red blood cell lysis buffer (BioLegend). 313 Cell suspensions from mouse lungs were stained at 4°C for 30 min with CD45 -Brilliant 314 Violet 711 (clone 30 -F11, 1:300, 103147, BioLegend), CD31 -Alexa Fluor 700 (clone 390, 315 1:600, 102443, BioLegend) antibodies and Ale xa Fluor 647 -conjugated isolectin griffonia 316 simplicifolia IB4 (GS -IB4, 1:1000, I32450, Invitrogen) with Fc block by CD16/CD32 317 antibody (clone 2.4G2, 1:600, BD Bioscience, 553141). Viable cells were detected as 318 negative for DAPI staining (BD Bioscience, 564907). 319 Cell suspensions from rat lungs were stained at 4°C for 30 min with CD45-PerCP/Cyanine5.5 320 (clone OX-33, 1:800, 202318, BioLegend), CD31 -PE (clone TLD-3A12, 1:800, 555027, BD 321 Pharmingen) antibodies and Alexa Fluor 647 -conjugated isolectin griffonia si mplicifolia IB4 322 (GS-IB4, 1:1000, I32450, Invitrogen) with Fc block by CD32 antibody (clone D34 -485, 323 1:1000, BD Pharmingen, 550271). Viable cells were detected as negative for 7-AAD viability 324 staining (BioLegend, 420404). 325 Endothelial cells were identified a s CD45 -CD31+ cells, and lung microvascular endothelial 326 cells were subsequently gated as CD31 +GS-IB4+ cells. Autophagic flux was assessed with an 327 Autophagy detection kit (Abcam, ab139484) using monodansylcadaverine as fluorescent 328 marker for autophagic vacuo les. All flow cytometric data were recorded on a Symphony A5 329 flow cytometer with FACSDiva 8.0.1 software and analyzed with FlowJo 10.8.1 software 330 (BD Biosciences). 331 Stereology 332 Post-mortem, murine and rat lungs were perfused with fixative (1.5% glutaraldehyd e and 333 1.5% paraformaldehyde (PFA) in 0.15 Mol/L HEPES buffer; pH 7.35) at an initial inflation 334 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint pressure of 25 cmH 2O and then at 20 cmH 2O during fixation. Lungs and heart were removed 335 and transferred to the same fixative solution and then further processed for stereological 336 imaging. After separation of heart and lungs, volumes of the left and right lung [V(lung)] 337 were determined separately by fluid displacement according to Archimedes’ princ iple17. The 338 lung tissue was embedded into 4% agar and then cut into slices with an isotropic uniformly 339 random (IUR) protocol using an orientator 18. Every second section was embedded in glycol 340 methacrylate (Technovit ® 7100, Kulzer GmbH) as described previou sly19. The remaining 341 tissue slices were further cut into 1 mm 3 tissue blocks and embedded in Epon resin (Roth, 342 Karlsruhe, Germany) as described before20. Tissue sections embedded in Technovit® were cut 343 into 1.5 µm thick sections, stained with toluidine blu e and imaged on a Axio Scan.Z1 slide 344 scanner (Zeiss, Göttingen, Germany). From these, the volume of the lung parenchyma was 345 assessed using the newCAST stereology software (Visiopharm, Horsholm, Denmark). An 346

lens magni fication of 20x with a test grid consisting of 3 x 3 points was used to 347 count the number of points hitting parenchyma (P Par) and nonparenchymal (PNonPar) structures 348 and to calculate the total volume of parenchyma of the lung (V(par,lung)) by multiplying PPar 349 with V(lung). 350 The total number of capillaries was quantified according to a standard protocol described by 351 Willführ and colleagues21 by estimating the Euler number of capillary networks. To this end, 352 the disector principle was applied 21 and 1 µm thick consecutive sections were cut from Epon 353 embedded tissue blocks stained with toluidine blue for parenchyma and nonparenchymal 354 structures. Sections were again digitalized with the AxioScan Z.1 slide scanner at 40 x 355 magnification. The Euler -Poincaré characteristic/Euler number of capillary networks w as 356 calculated by counting the topological constellations: new, isolated parts (islands); enclosed 357 cavities within an existing pro file (holes); and in the majority of cases in this study, extra 358 connections between two isolated structures (bridges). The first and third sections were 359 analyzed using the disector method 22, and stereological estimation was performed by 360 Visiopharm software. In brief, tissue sections were digitally aligned, and profile events were 361 estimated within an unbiased counting frame with an area of 9123.42 μm2 (A(frame)). The 362

volume for the profile estimation V(dis) was calculated from the number of 363 counting frames (P(ref)) that were assessed and the height of the disector pairs (h, 2 µm) as 364 shown below 365 [𝑉(𝑑𝑖𝑠)] = ℎ × (𝑃(𝑟𝑒𝑓)) × 𝐴(𝑓𝑟𝑎𝑚𝑒) .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint From the raw counts, the Euler number of capillaries [ χ(cap)] was determined as: 366 𝜒(𝑐𝑎𝑝) = ∑(𝑖𝑠𝑙𝑎𝑛𝑑𝑠) − ∑(𝑏𝑟𝑖𝑑𝑔𝑒𝑠) + ∑(ℎ𝑜𝑙𝑒𝑠). 367 The numerical density [Nv(cap/lung)] was then calc ulated using χ(cap) as well as the total 368 number of capillaries per lung [NV(cap,lung)]: 369 𝑁V (cap,lung) = − ∑ 𝜒(𝑐𝑎𝑝) 𝑉(𝑑𝑖𝑠) × 𝑉(𝑙𝑢𝑛𝑔) 370 In a subset of experiments (specifically, in the SU5416-treated ZSF1 obese rats and their 371 corresponding controls), image quality and the resolution at light microscopic level was not 372 sufficient to allow for the stereological approach outlined above. Instead, we utilized TEM 373 images from these lungs to calculate total capillary volume density within the inter -alveolar 374 septa of the lung as previously described 20. In brief, a point grid test system was 375 superimposed onto the digitized TEM images to estimate volume densities of the respective 376 compartments according to design -based stereological principles. By point counting, the 377 epithelial, capillary endothelial, interstitial and capillary lumen (P capillary lumen) compartment 378 were quantified using STEPanizer software 23.The total number of test points hitting the 379 alveolar septum (Ptotal) was defined as the sum of points hitting alveolar epithelium, capillary 380 endothelium, interstitium, and capillary lumen. Total capillary volume density within the 381 septum was calculated as: Vv(cap/sept) = Pcapillary lumen / Ptotal. 382 Transmission electron microscopy 383 For transmission electron microscopic (TEM) analysis, ultrathin sections of 70 nm were cut 384 from Epon embedded tissue blocks collected on pioloform -coated copper grids and stained 385 with uranyl acetate and lead citrate60. TEM images were acquired by a Zeiss E M 906 386 electron microscope at 80 kV acceleration voltage (Carl Zeiss, Oberkochen, Germany) 387 equipped with a 2K CCD camera (TRS, Moorenweis, Germany). To assess autophagic stress 388 in EC, autophagosomes were identified as double membrane structures containing 389 undegraded cytoplasmic components such as endoplasmic reticulum membranes, 390 mitochondria, or ribosomes as previously described24. 391 Micro-computed tomography 392 To prepare lung tissue for micro-computed tomographic (µCT) imaging of the microvascular 393 bed, the contrast agent Vascupaint™ (SKU: MDL -121, MediLumine , Canada) was infused 394 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint immediately post -mortem into the RV , from where it distributed through the pulmonary 395 vascular system. Once the contra st agent emerged from the LV, perfusion was maintained for 396 an additional 15 min to allow for optimal distribution of the contrast agent across the 397 pulmonary vasculature. Then, the pulmonary artery and the opening in the left ventricle were 398 ligated and the entire body of the animal was placed in the refrigerator (4°C) overnight to 399 harden the Vascupaint inside the lung. The next day, the lungs were transfered to 4% PFA for 400 storage and subsequent scanning by µCT. 401 Lungs were imaged on a SkyScan 1276 µCT imager (Bruker, Belgium) using the vendor’s 402 software for image acquisition (v.1.4) with the following parameters: X -ray source voltage of 403 100 kV and source current of 80 µA, step and shoot mode, 360° acquisition, Cu 0.25 mm 404 filter, rotation step of 0.3°, pixel si ze of 8 µm, exposure time of 2,000 ms, and frame 405 averaging of 4. Flat field correction was always applied. 406 Images were reconstructed to 8-bit files using NRecon (v.1.7.4.6, Bruker, Belgium). Beam -407 hardening correction of 12% and ring -artefact correction of 2 were applied. Image analysis 408 was performed with CTAn (v.1.20.8.0, Bruker, Belgium) starting with a delineation of a 409 volume of interest (VOI) to separate the tissue from the surrounding elements in the image. 410 Next, multiple analysis steps were applied to calculate the total organ volume, the volume of 411 the vascular tree and the distributions of the diameters of the segmented blood vessels as 412 follows: Gaussian filtering 3D (radius = 2), global thresholding (11 -255), morphological 413 opening 3D (r=3), sweeping a ll except the largest object (3D), morphological closing 3D 414 (r=3), removing pores 3D, morphological closing 2D (r=8), removing pores 3D, 415 morphological opening 2D (r=8), removing speckles 3D (<5,000 voxels), saving bitmap, 3D 416 analysis, reloading image, median filtering 3D (r=1), global thresholding (62 -255), removing 417 speckles 3D (<30 voxels), saving bitmap, and finally 3D analysis with computation of the 418 trabecular thickness distribution. In a few cases the global threshold for segmentation of the 419 total orga n volume was set at 15 instead of 11 to achieve the same accuracy. Total tissue 420 volume and total vascular volume were calculated. To specifically capture changes in the 421 pulmonary microvasculature, vascular volume of vessels <250 µm in size was quantified; a 422 vessel size of 110 µm was defined as lower cutoff for reliable resolution of single vessel 423 segments. 424 Pulmonary capillary surface area 425 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint In a subset of AoB experiments, pulmonary capillary surface area was determined by the blue 426 dextran (BD, Sigma -Aldrich, Germany) efflux method 25. In brief, lungs were isolated from 427 AoB or sham rats, perfused with Krebs -Ringers buffer containing 5% dextran, and then 428 loaded with BD by addition of 0.67 mg/mL BD to the perfusate. Samples were collected from 429 the outflow reservoir at 1 min intervals, then circulatory flow was re -started with BD -free 430 Krebs-Ringers buffer containing 7% bovine serum albumin (BSA) with sample collect ion 431 from the outflow reservoir in 30 s intervals. The perfusion volume was recorded, and the BD 432 concentration in each sample was spectrophotometrically determined at λ=630 nm. As the 433 pulmonary capillaries will bind BD during dye loading as a function of th eir total surface 434 area, the subsequent efflux of dextran during BD -free perfusion can be used to calculate how 435 much BD did initially bind to the capillary bed, and thus to estimate the pulmonary capillary 436 surface area. 437 Lung immunohistology 438 Lung tissue sect ions were fixed in 4% PFA and stored in 1x PBS. After embedding in 439 paraffin, 5 µm thick sections were cut, dried at 65°C and deparaffinized by Xylol (2X 5 min, 440 Roth) and a decreasing alcohol series (2X 100%, 90%, 80%, 70%) finished with distilled 441 water (5 min each step). Samples were boiled in a microwave at 600W for 10 min in Tris -442 EDTA buffer, then cooled for 20 min. After rinsing sections for 5 min in PBS, samples were 443 blocked by PBTB (0.2% Triton X -100, 5% normal goat serum, 0.2% BSA in PBS) at room 444 temperature for 1 h and finally incubated with primary antibodies at 4°C overnight. Primary 445 antibodies used for immunostaining of lung tissue sections and respective dilutions were: 446 PECAM-1 (1:400; goat polyclonal; #AF3628, R&D Systems, Minneapolis, MN), cleav ed-447 caspase 3 (Cl -Cas3; 1:400; rabbit polyclonal; #9661, Cell Signaling, Danvers, MA), and 448 cleaved PARP-1 (1:400; Abcam, #ab32064). The tissue sections were washed 1x with PBS, 449 followed by incubation with fluorescent dye -conjugated secondary antibodies at r oom 450 temperature for 1 h and again 1x wash in PBS. Secondary antibodies were: Alexa Fluor TM 451 594 rabbit anti -goat antibody (1:500; #A11080), and Alexa Fluor TM 647 goat anti -rabbit 452 antibody (1:500; #A21245, all Invitrogen TM, Carlsbad, CA). After 1x wash in PBS, tissue 453 sections were stained with DAPI (1:1000; # D9542, Sigma -Aldrich) for 10 min. Following 454 another 1x wash in PBS, tissue sections were mounted with Fluoromount -GTM without 455 DAPI (Invitrogen TM, Carlsbad, CA), imaged by confocal microscopy (Nikon Scanning 456 Confocal A1Rsi+, Tokyo, Japan) and analyzed using ImageJ. Representative images were 457 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint selected from randomly acquired fields using systematic sampling across each section. For 458 each group, the representative field was chosen from an animal with a quantified value 459 closest to the group median. Display settings were applied identically across groups. 460 Single cell RNA sequencing 461 The dataset used in this study was generated in a previous study from lungs of C57BL/6J 462 mice treated with L-NAME and HFD (analogous to our murine HFpEF model) for 2 weeks 26, 463 and is publicly available through the Gene Expression Omnibus under accession number 464 GSE244309. The full code used for the original scRNA -seq analyses is available at 465 https://github.com/kropskilab/myeloid_il1b. We re-analyzed this single-cell RNA sequencing 466 (scRNA-seq) dataset using the Seurat R package (v4) alongside supporting packages dplyr, 467 tidyr, ggplot2, and viridis for data processing and visualization. The Seura t object was first 468 loaded, and samples were annotated into experimental groups ("Control" or "HFpEF") based 469 on sample identifiers. Gene expression was normalized using Seurat’s NormalizeData 470 function (log -normalization). For targeted analyses of integrated apoptotic and autophagic 471 markers, curated gene sets related to regulated cell death pathways and autophagy were 472 selected based on existing literature. The applied gene sets are given in Suppl. Tables 1A and 473 1B. Individual cell types were identified as in the original publication 26 and expression 474 matrices for the selected gene sets were extracted for specific cell types. For each cell, a "net 475 expression" score was calculated as the sum of expression values across all genes in the set. 476 Individual samples wer e grouped by experimental condition. Data are displayed as boxplots 477 and scatterplots of net expression. 478 Data analyses and statistics 479 Statistical analyses were performed using GraphPad Prism 9 software (GraphPad Software, 480 Inc., Boston, MA). Results are rep orted as median with interquartile range (IQR) or 481 mean±SEM. Normality was assessed by Shapiro -Wilk normality test. For 2 -group 482 comparisons, parametric data sets were analyzed by unpaired parametric two-tailed t -test; 483 non-parametric data sets by Mann -Whitney U-test. P-values were adjusted for multiple 484 testing using the false discovery rate (FDR) method. For comparisons between multiple 485 groups, one-way analysis of variance (ANOV A) was used for parametric data sets, followed 486 by Dunn´s test. For correlations analysis, the Pearson correlation coefficient r was calculated. 487 Statistical significance was assumed at p < 0.05. 488 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 489 490

491 Impaired oxygenation in HFpEF patients 492 To assess parameters of pulmonary oxygenation in clinical HFpEF, we first analyzed data 493 from 234 HFpEF patients enrolled in the clinical cohort of the CRC1470 at the German Heart 494 Center of the Charité, Berlin. Patients were stratified based on NYHA classification, and 495 echocardiographic evaluation confirmed increased HFpEF severity – namely a higher ratio of 496 early to late LV filling velocity (E/A), a higher ratio of early diastolic mitral inflow velocity 497 over early diastolic mitral annular tissue velocity (E/e´), and a lower global longitudinal strain 498 (GLS) - with higher NYHA class (Fig. 1A). LV eje ction fraction (LVEF) did not differ 499 between classes and was consistently >50% confirming the diagnosis of HFpEF. LV end -500 diastolic volume (EDV) showed a trend to lower values with higher NYHA class that, 501 however, did not reach the level of significance. Co mplete patient demographics (Suppl. 502 Table 2) and the full set of echocardiographic parameters (Suppl. Table 3) are provided as 503 Supplementary Material. 504 Inversely to HFpEF severity, arterial oxygen saturation (SaO 2) measured either at rest or 505 during cardiopulmonary exercise testing at maximal load decreased with advancing NYHA 506 class, with patients of NYHA class III showing significantly lower SaO 2 levels than those 507 with NYHA class I (Fig. 1B). These findings not only consolidate previous reports of 508 systemic hypoxemia in HFpEF patients, but demonstrate a direct association between disease 509 severity and the degree of hypoxemia both at rest and – even more pronounced - during 510 exercise. 511 Consistent with impaired blood oxygenation during pulmonary transit, assessment of single -512 breath carbon monoxide uptake in the same patient cohort revealed a decreased diffusing 513 capacity of the lungs (DLCO -SB) in NYHA cl ass III relative to NYHA class I patients 514 (Suppl. Fig. 1A). Alveolar volume (V A) and hemoglobin (Hb) concentration, which may 515 affect DLCO, did not differ between groups (Suppl. Fig. 1B,C), suggesting impaired gas 516 diffusion across the alveolo-capillary membrane in severe HFpEF. This notion was confirmed 517 by lower DLCO -SB after correction for Hb (DLCOc -SB; Fig. 1C) and a lower transfer 518 coefficient for CO (KCOc -SB), which corrects DLCOc -SB for V A (Suppl. Fig. 1D). In line 519 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint with this notion, estimated dead space (VD/VT) increased with NYHA class in exercise, 520 although it did not differ at rest (Fig. 1D). Finally, DLCOc -SB correlated with parameters of 521 LV diastolic function, specifically with E/e´ and EDV indicating that impaired gas diffusion 522 in the lung was directly related to LV diastolic dysfunction in HFpEF (Fig. 1E). 523 524 Lung microvascular rarefaction in HFpEF animals 525 To probe for micro vascular rarefaction as p otential mechanism of impaired pulmonary gas 526 exchange, we next characterized microvascular remodeling across two different preclinical 527 models of HFpEF. First, we used the established model of SU5416 -treated ZSF1 obese rats 528 (Fig. 2A). Consistent with HFpEF , rats sh owed characteristics of LV diastolic dysfunction, 529 namely an altered E/A ratio and a reduced GLS, while the increase in E/e´ ratio did not reach 530 the level of significance (Fig. 2B). LV EF remained unchanged at approximately 60%. 531 Invasive hemodynamic measure ments demonstrated an increase in RVSP in line with 532 secondary pulmonary hypertension (PH). µCT imaging (a representative recording of which 533 is shown as Suppl. Video 1) revealed a decrease in pulmonary vascular volume that was 534 particularly prominent in vessels <250 µm in diameter (Fig. 2D ; Suppl. Fig. 2 A), and a 535 decrease in capillary volume density within the interalveolar septa of the lung as determined 536 by stereological assessment of TEM images (Fig. 2E). A full set of echocardiographic, 537 hemodynamic an d gravimetric data on SU5416 -treated ZSF1 obese rats is given in Suppl. 538 Table 4, and additional stereological parameters are given in Suppl. Table 5. 539 To further consolidate these results in a different species and to allow for subsequent 540 mechanistic interrogation in genetically modified animals, we next probed for microvascular 541 rarefaction in a murine HFpEF model of metabolic and hypertensive stress (Fig. 2F). In line 542 with previous studies, the combination of high fat diet (HFD) with the nitric oxide syntha se 543 inhibitor L-NAME induced characteristic signs of diastolic dysfunction, reflected by a change 544 in E/A ratio, an increased E/e´ ratio and a corresponding decrease in GLS (Fig. 2G). In line 545 with a HFpEF phenotype, LVEF was again preserved (Fig. 2G), yet he modynamic signs of 546 overt PH were absent (Fig. 2H). Flow cytometric analyses revealed a time -dependent, 547 exponential loss of lung microvascular endothelial cells and stereological assessment 548 confirmed a reduced number of capillaries per lung in HFpEF mice (Fig. 2I). Finally, HFpEF 549 mice showed characteristic signs of exercise intolerance, including reduced running distance 550 and lower maximal oxygen consumption (VO ₂max) during treadmill exercise (Fig. 2J). Full 551 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint sets of echocardiographic , hemodynamic and gravimetric data as well as stereological data 552 are given in Suppl. Tables 6&7. 553 554 555 556 Lung microvascular rarefaction in AoB rats 557 To probe whether lung microvascular rarefaction is specific for HFpEF, we next characterized 558 the pulmonary microvasculature in a preclinical heart failure model independent of prototypic 559 triggers of HFpEF, namely AoB rats (Fig. 3A). After 9 weeks, AoB rats also showed signs of 560 LV diastolic dysfunction including an increase in E/A and E/e´ ratio and a reduced GLS (Fig. 561 3B). While LVEF showed a slight trend to decline, this effect did not reach statistical 562 significance. Similar to SU5416 -treated ZSF1 obese rats, AoB rats had an increased RVSP 563 (Fig. 3C). A full set of echocardiographic, hemodynamic and gravimetric data is given in 564 Suppl. Table 8. µCT imaging did not detect significant differences in tissue or vascular 565 volume at the total lung level between AoB rats and controls (Suppl. Fig. 2 B); yet, l ung 566 vascular volume of vessels <250 µm in diameter was again significantly reduced consistent 567 with loss of pulmonary microvessels (Fig. 3D). BD efflux analyses, flow cytometry (gating 568 strategy shown in Suppl. Fig. 3) and quantitative stereology demonstrate d a corresponding 569 decrease in pulmonary capillary surface area, the total number of lung microvascular 570 endothelial cells, and the total capillary number per lung (Fig. 3E), consolidating lung 571 microvascular rarefaction in AoB rats. A full set of stereological parameters is given in Suppl. 572 Table 9. Taken together, th ese findings identify pulmonary microvascular rarefaction as a 573 characteristic pathological feature of left heart failure in different species and preclinical 574 models. 575 576 HFpEF drives autophagy-dependent cell death in lung microvascular endothelial cells 577 In recent work, we identified chronic hypoxia to cause autophagy -driven cell death in lung 578 microvascular endothelial cells 24. To probe for a potentially similar scenario in HFpEF, we 579 accessed a single cell RNA sequencing (scRNAseq) dataset from a previous study by 580 Agrawal and coworkers26 which analyzed murine lungs in a similar model of HFD/L-NAME-581 induced HFpEF at an earlier timepoint (2 weeks). Unbiased clustering reproduced the same 582 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint cell types as originally reported by the authors including different endothelial subsets, namely 583 mesothelial, arteriolar, general capillary (gCap), aerocytes (aCap) and venular endothelial 584 cells (Fig. 4A). Relative to controls, HFpEF mice showed a significant loss of gCaps while a 585 concomitant decrease in aCaps did not reach significance (Fig. 4B). Conversely, the 586 proportion of pericytes tended to increase. Consistently, an aggregate set of genes involved in 587 regulated cell death pathways increased significantly in gCaps while aCaps did again not 588 reach significance (Fig. 4C). This effect was associated with a parallel increase in an 589 aggregate set of autophagy marker genes in both aCaps and gCaps, but not in pericytes (Fig. 590 4D). When compared at the single cell level, aggregate autophagy marker genes correlat ed 591 with markers of regulated cell death pathways in gCaps (rs=0.38) and to a lesser extent also in 592 aCaps (rs=0.26; Fig. 4E). These findings indicate a propensity for regulated lung endothelial 593 cell death in HFpEF that is most pronounced in gCaps and less so in aCaps. 594 Evidence for increased endothelial autophagy and regulated cell death was next validated at 595 several levels in our HFpEF mouse model. First, we monitored autophagic flux in CD31+GS-596 IB4+ lung microvascular endothelial cells of HFpEF mice by flow cytometric detection of 597 autophagic vacuoles with monodansylcadaverine. In line with our scRNAseq analyses, 598 autophagic flux in lung endothelial cells increased within one week of HFpEF induction and 599 reached its plateau between weeks 6 - 9 (Fig. 5A,B). Consistently, TEM identified endothelial 600 autophagosomes in lung microvessels of HFpEF mice but not in control mice (Fig. 5C). 601 Subsequent immunohistological analyses revealed an increased abundance of apoptot ic 602 endothelial cells, detected as CD31 + cells staining positive for cleaved caspase 3 (Fig. 5D) or 603 cleaved-PARP1 (Fig. 5E), respectively, in HFpEF relative to control mice at week 12 , and – 604 in case of cleaved caspase 3 – as early as week 1. 605 To experimentally test for a functional role of increased endothelial autophagy in capillary 606 rarefaction in HFpEF, we next subjected mice with an endothelial -specific knockout of the 607 essential autophagy gene Atg7 (Atg7EN-KO) to our murine HFpEF model (Fig. 5F). In 608 stereological analyses, Atg7EN-KO HFpEF mice had hi gher numbers of capillaries per lung 609 relative to their corresponding Atg7EN-WT controls, indicating that capillary rarefaction is 610 prevented in the absence of excessive autophagic flux (Fig. 5 G). A similar increase in 611 capillary number in Atg7EN-KO relative to Atg7EN-WT mice was not evident under control 612 conditions, consistent with the view that baseline autophagic flux does not cause excessive 613 endothelial cell death. 614 615 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint Systemic hypoxemia aggravates experimental HFpEF 616 To probe for functional effects of capillary rarefaction, we assessed parameters of gas 617 exchange in both rat and mouse models. In AoB rats, microvascular rarefaction secondary to 618 HF was associated with systemic hypoxemia as evidenced by reduced arterial partial pressure 619 of oxygen (PaO2) and arterial oxygen saturation (SaO 2), respectively (Fig. 6A,B). In contrast, 620 venous partial pressure of oxygen (PvO 2) remained unchanged (Fig. 6B), indicating impaired 621 pulmonary gas exchange rather than increased peripheral oxygen extraction as cause of 622 arterial hypoxemia. Analogous systemic hypoxemia was evident in HFpEF mice both at rest 623 and after treadmill exercise (Fig. 6 C,D). Conversely, attenuation of lung capillary rarefaction 624 in Atg7EN-KO mice with HFpEF improved oxygenation both at rest and after exercise (Fig. 625 6E,F), and this was associated with an extended running distance and improved maximal 626 oxygen consumption (VO ₂max; Fig. 6 G) relative to Atg7EN-WT mice. Not ably, these effects 627 were reversed in control, non -HFpEF mice, in that Atg7EN-KO mice had worse oxygenation 628 and exercise tolerance compared to their Atg7EN-WT counterparts. As we will discuss below, 629 these findings contrast the physiological relevance of bas eline autophagy versus the 630 pathological implications of excessive autophagy in HFpEF. Overall , these data identify 631 pulmonary microvascular rarefaction as a consequence of excessive endothelial autophagy 632 and cell death and as a characteristic pathological f eature of heart failure that contributes to 633 systemic hypoxemia and exercise intolerance. 634 To probe whether the resulting systemic hypoxemia may in turn affect severity and 635 progression of heart failure, we next combined the HFpEF mouse model with a final 2 -week 636 exposure to hypoxia (10% O ₂) starting after week 10 of HFD/L -NAME treatment (Fig. 6 H). 637 Echocardiographic assessment after 2 weeks of hypoxia revealed an exacerbated HFpEF 638 phenotype with preserved LVEF but signs of aggravate d LV dysfunction, indicated by an 639 increased E/e´ and a further decrease in GLS while E/A ratio remained pathologically altered 640 (Fig. 6 I). Complete echocardiographic data are given in Suppl. Table 10. These findings 641 demonstrate that hypoxemia can aggravate LV dysfunction, suggesting a potential detrimental 642 feed-forward loop between HFpEF, lung capillary rarefaction and hypoxemia as driver of 643 disease progression. 644 645 Loss of endothelial autophagy and moderate hyperoxia mitigate experimental HFpEF 646 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint Contrasting the effects of hypoxia, prevention of capillary rarefaction and (partial) restoration 647 of normoxemia in Atg7EN-KO HFpEF mice was associated with improved LV diastolic 648 function as demonstrated by lower E/A (yet without reaching significance at p=0.059) and 649 E/e´ ratios as well as GLS relative to their corresponding Atg7EN-KO controls (Fig. 7A,B). 650 Complete echocardiographic data are given in Suppl. Table 11. To consolidate the link 651 between restored normoxemia and improved LV function in a translational ly relevant 652 scenario, we finally tested whether moderate hyperoxia may similarly mitigate LV 653 dysfunction in HFpEF mice. Analogous to our hypoxia experiments, we exposed HFpEF 654 mice to 40% O ₂ for the last 2 weeks of the HFD/L -NAME protocol (Fig. 7C). Similar t o our 655 findings in Atg7EN-KO mice, hyperoxia normalized E/A and E/e´ ratios as well as GLS (Fig. 656 7D) while LVEF remained unchanged (Fig. 7D). Stereological assessment of lung tissue 657 sections and arterial pulse oximetry at rest and after exercise further rev ealed that improved 658 LV function was in turn associated with a restoration of lung capillarization and gas exchange 659 (Fig. 7E,F). At the functional level, this resulted in restored exercise tolerance, evident as 660 normalized running distance in treadmill tests (Fig. 7G). The complete set of 661 echocardiographic data are given in Suppl. Table 12. 662 663

664 In the present study, we identif ied in three different animal models of left heart disease 665 extensive structural remodeling of pulmonary microvessels that was associated with impaired 666 blood oxygenation and exercise performance. The latter findings were validated in a clinical 667 HFpEF cohort in that disease severity and parameters of LV dysfunction correlated with 668 impaired alveolo -capillary gas exchange. Mechanistic ally, microvascular rarefaction in 669 HFpEF was driven by excessive autophagy and subsequent autophagy -associated cell death 670 of lung capillary endothelial cells. At the functional level, the resulting hypoxemia aggravated 671 LV dysfunction and thus, accelerated HFpEF progression. Conversely, moderate hyperoxia 672 restored LV function, lung capillarization, gas exchange and exercise tolerance. These 673 findings identify pulmonary microvascular rarefaction as a previously unrecognized 674 pathological feature of HFpEF that drives disease progression and may contribute to recurrent 675 oxygen desaturations and dyspnea in HFpEF patients (Fig. 8). Moderate oxygen therapy may 676 present a promising avenue to decelerate or even reverse this disease trajectory. 677 678 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint Impaired pulmonary gas exchange in HFpEF 679 Dyspnea or “air hunger” is one of the cardinal symptoms of HFpEF. Importantly, dyspnea is 680 alternatively triggered via inputs from arterial chemosensors or lung mechanosensors27, rather 681 than by changes in cardiac fun ction per se . Indeed, arterial desaturations are frequently 682 observed in HFpEF patients at rest and more characteristically during exercise 5,6. This effect 683 cannot be directly attributed to impaired hemodynamics in heart failure , as r educed CO and 684 increased left atrial pressures do by themselves not decrease, but potentially even increase 685 arterial oxygenation as they extend the contact time for blood with the oxygenation unit in the 686 alveolo-capillary compartment28. Increased peripheral oxygen extraction due to low CO may 687 potentially cause venous desaturation, which was, however, not evident in AoB rats and 688 should be readily compensated for by the physiological oxygenation reserve capacity of the 689 lung as long as alveolo -capillary gas exc hange is uncompromised. As such, dyspnea and 690 arterial hypoxemia in HFpEF point to structural alterations or functional deficiencies of the 691 gas-exchange units in the lung. 692 Indeed, patients with HFpEF or other forms of heart failure frequently show characteristics of 693 impaired alveolo-capillary gas exchange, evident as reduced DLCO 29-31. Further expanding 694 on these findings, we show for our clinical HFpEF cohort that arterial oxygen saturation at 695 rest and during exercise, as well as DLCO decreased as a function of NYHA class, 696 highlighting a direct association between disease severity and impaired pulmonary oxygen 697 exchange. DLCO changes were independent of alveolar volume or hemoglobin 698 concentration, consolidating the notion that impaired oxygenation and hypoxemi a reflect a 699 loss of effective alveolo -capillary diffusion capacity rather than changes in ventilated lung 700 volume or blood oxygen -carrying capacity. Further, DLCO correlated with 701 echocardiographic indices of diastolic dysfunction, indicating that declining efficiency of 702 pulmonary gas exchange parallels progressive worsening of LV filling in HFpEF patients. As 703 such, impaired pulmonary gas exchange emerges as a characteristic co -morbidity of HFpEF 704 that not only parallels the severity of HFpEF, but – as we will discuss later – may also 705 critically influence its progression. 706 Compromised lung diffusion capacity in HFpEF or other forms of heart failure has been 707 alternatively attributed to the formation of cardiogenic edema 32, structural thickening of the 708 alveolo-capillary barrier33, and/or ventilation-perfusion mismatching9,34-36. The extent of both 709 interstitial and alveolar edema in chronic heart failure patients is, however, relatively modest 710 due to a series of effective adaptive, counter -regulatory mechanisms such as a reduction in 711 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint endothelial permeability and an increase in alveolar fluid clearance 37,38. Similarly, definite 712 data showing remodeling of the alveolo -capillary barrier or impaired ventilation -perfusion 713 matching in HFpEF patients and their contribution to impaired pulmonary gas exchange are 714 so far lacking. 715 In the present work, we provide first evidence for a decline in lung vascular gas exchange 716 surface area as pathomechanism for impaired alveolo -capillary gas exchange in HFpEF. 717 While surface area as a key determinant of alveolo -capillary gas diffusion is commonly 718 discussed in terms of alveolar surface, it is important to consider the equal relevance of 719 capillary surface area as its vascular counterpart. In the present stu dy, lung microvascular 720 rarefaction was evidenced in three different animal models of heart failure in two different 721 species –SU5416-treated ZSF1 obese rats, HFD/L-NAME-treated mice, and AoB rats – by a 722 set of different methodologies: µCT revealed a loss o f lung vascular volume in rat lungs that 723 was most pronounced in the microvascular compartment comprised of vessels <250 µm in 724 diameter. As µCT can, however, not resolve single alveolar capillaries, we consolidated 725 microvascular rarefaction at the capillary level by quantitative stereology, commonly 726 considered as the gold standard of lung morphometry 39. Specifically, we calculated the total 727 number of capillaries per lung from the Euler -Poincaré characteristic based on recognizable 728 topological constellations termed “islands”, “bridges” and “holes” in an unbiased manner 20. 729 Compared to their respective controls, both HFpEF mice and AoB rats showed a significant 730 reduction in total capillary number per lung, consolidating that microvascular rarefaction 731 extended to the functional units of gas exchange. Consistent with capillary rarefaction, flow 732 cytometric analyses revealed a time -dependent loss of microvascular EC in lungs with the 733 progression of HFpEF in mice that exceeded 50% after 12 weeks of HFD and L -NAME 734 treatment. Microvascular EC loss of similar magnitude was evident in AoB rats after 9 weeks. 735 In AoB rats, we further consolidated the actual loss of capillary gas exchange area by 736 measuring the efflux over time of a dye previously loaded to capillary EC in isolated perfused 737 lungs25. As such, we provide multiple levels of evidence for a structural loss of functional gas 738 exchange area in different animal models. The fact that microvascular rarefaction was equally 739 evident in classic rodent models of HFpEF as well as in a metabolic stress-independent model 740 of pressure overload in AoB rats highlights loss of pulmonary gas exchange units as universal 741 feature of heart failure, suggesting in turn that microvascular rarefaction is primarily driven 742 by changes in pulmonary hemodynamics common to different forms of heart failure rather 743 than by their individual pathogenetic mechanisms. In line with this notion, clinical finding of 744 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint reduced DLCO and symptoms of dyspnea span across the entire spectrum of heart failure 745 patients29-31. 746 The extent of EC loss required to impair alveolo -capillary gas exchange is presently 747 unknown. The network of alveolar capillaries is extremely dense; that notwithstanding any 748 decrease in capillary gas exchange area will directly translate int o a proportional decrease in 749 oxygen diffusion as described by Fick´s first law of diffusion. At rest, impaired diffusion may 750 in part be compensated by the fact that blood oxygenation is completed before the blood exits 751 the pulmonary capillary bed 40, yet these compensatory mechanisms tend to fail with shorter 752 capillary transit times as seen e.g. in exercise. In line with this concept, arterial desaturation 753 in patients with NYHA class II or III became more prominent during exercise as compared to 754 rest. 755 (Micro)vascular rarefaction has previously been reported as a characteristic feature in patients 756 with and animal models of pulmonary arterial hypertension 24,41,42. To our knowledge, the 757 present work constitutes, however, the first report of a similar ph enomenon in left heart 758 disease. This finding is unexpected as the increased left atrial pressures seen in HF will 759 acutely recruit previously unperfused alveolar capillaries and further distend open 760 microvascular segments 28, resulting in a net increase in c apillary gas exchange surface. Our 761 present data show, however, that this acute effect is subsequently counteracted by a 762 progressive loss of EC and lung capillaries. 763 Further mechanistic interrogation revealed increased autophagic flux and a higher propensity 764 for apoptotic cell death in lung microvascular endothelial cells of HFpEF mice as compared 765 to controls. Specifically, analysis of a single -cell transcriptomic dataset of murine lungs at an 766 early stage of HFpEF (2 weeks after start of HFD/L -NAME) yielded increased expression of 767 regulated cell death pathway and autophagy related genes predominantly in gCaps that was 768 associated with a loss in total gCap cell numbers. The fact that these effects were more 769 prominent in gCap s than aCaps may at first seem counterintuitive in the context of HFpEF 770 associated arterial desaturation, as aCap s form the primary vascular unit for pulmonary gas 771 exchange43. However, as gCap s serve as progenitor pool for the largely non -proliferative 772 aCap population44, loss of gCaps at an early stage (shown in the present scRNAseq dataset to 773 occur within 2 weeks of HFpEF) is expected to have profound effects also on aCaps over the 774 subsequent progression of HFpEF. Consistently, flow cytometric and immunohistological 775 analyses of lungs from HFpEF mice revealed a progressive increase in autophagic flux over 776 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint time and elevated markers of endothelial apoptosis, specifically cleaved caspase -3 and 777 cleaved-PARP1 in CD31+ cells at 12 weeks HFD/L -NAME. While physiological autophagy 778 constitutes an important cell survival mechanism, an emerging body of evidence shows that 779 excessive autophagy can trigger apoptotic cell death (autophagy -dependent cell death, 780 ADCD)45. In the present study, this dichotomy likely accounts for the finding that control 781 mice with EC -specific loss of the es sential autophagy gene Atg7 only managed a shorter 782 running distance compared to their wild type counterparts in treadmill tests, yes performed 783 better in the HFpEF model. 784 In previous work, we reported ADCD as a driver of lung microvascular EC loss in chronic 785 hypoxic PH24. In the present study, we identified an analogous mechanism in HFpEF, in that 786 deficiency of the essential autophagy gene Atg7 protected HFpEF mice from lung EC 787 apoptosis and capillary loss. Notably, this finding is in line with the results from a cardiome -788 directed network analysis of RNAseq data in HFpEF rats which identified apoptosis and 789 autophagy as one of five core disease processes 46. As such, excessive autophagy and ADCD 790 may constitute key drivers of both heart and lung disease in HFpEF. In parallel, ADCD of 791 lung microvascular endothelial cells emerges as a principal characteristic across different 792 types of PH (namely PH due to chronic hypoxia or HFpEF). 793 Importantly, the development of lung capillary rarefaction and subsequent hypoxemia 794 promote the progression of ventricular dysfunction in HFpEF in a detrimental feedforward 795 loop. Exposing HFpEF mice to hypoxia worsened characteristic signs of LV diastolic 796 dysfunction, while moderate hyperoxia improved LV function and exercise tolerance. 797 Notably, hyperoxia also restored pulmonary capillarization and normalized arterial 798 oxygenation, indicating that the "vicious circle" between HFpEF and lung c apillary 799 rarefaction can be reversed into a "virtuous circle". As such, ambulatory oxygen therapy e.g. 800 by high flow nasal oxygen may not only alleviate symptoms of dyspnea and increase exercise 801 capacity in HFpEF patients 47, but may provide a therapeutic me ans to decelerate disease 802 progression. Beyond oxygen therapy, interventions targeting the underlying mechanisms of 803 microvascular rarefaction may guide future therapeutic strategies with the aim to reduce 804 endothelial apoptosis and/or modulate excessive endo thelial autophagy to preserve lung 805 microvascular integrity. 806 Several methodological aspects and limitations of the present study merit specific 807 consideration. First, pressure -overload models are commonly considered as HFrEF rather 808 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint than HFpEF models. Hence, although the majority of our AoB rats had an EF >50% and 809 showed signs of diastolic dysfunction such as increased E/A, E/e´ and a reduced GLS, 810 caution is required in the interpretation of this model as HFpEF, as it may transition to a 811 HFrEF phenotype at later stages. In the present manuscript, we thus refer to AoB rats as a 812 model of heart failure due to pressure overload . In contrast, the SU5416 -treated ZSF1 obese 813 rat and the HFD/L -NAME murine model present well -established models of LV HFpE F, 814 which was confirmed in the present study by echocardiographic signs of diastolic dysfunction 815 in the presence of a sustained EF. Importantly, microvascular rarefaction and impaired gas 816 exchange were consistent across models irrespective of LVEF, suggesti ng that this 817 pulmonary phenotype reflects fundamental consequences of chronic pulmonary congestion 818 secondary to LV failure rather than a response unique to a specific subtype of HF. Second, 819 E/A ratios tended to change in different directions across models, with E/A increasing in 820 HFpEF patients, AoB rats, and HFD/L -NAME mice while they decreased in SU5416 -treated 821 ZSF1 obese rats and in hypoxia -exposed HFD/L -NAME mice. This seemingly discrepant 822 effect is well -recognized in the literature: As the transmitral f low pattern depends both on 823 loading conditions and diastolic function, E/A may decrease in HF due to impaired relaxation 824 or increase due to restrictive filling. Importantly, all models exhibited clear evidence of 825 diastolic dysfunction based on changes in G LS and/or E/e′, validating E/A as an informative 826 but context-dependent indicator of diastolic abnormalities. Third, in ZSF1 rats image quality 827 and the resolution at light microscopic level was not sufficient to allow for the stereological 828 assessment of cap illary number per lung. We thus adopted an alternative, TEM based 829 stereological approach to estimate capillary volume density within the interalveolar septa of 830 the lung in these animals. The identified loss of capillary structures was nevertheless 831 internally consistent in both quality and quantity with results obtained by classical stereology 832 in AoB rats or HFpEF mice. Fourth, it should be emphasized that the degree of hypoxia used 833 (10% O₂) exceeds the level of hypoxemia typically observed in human HFpEF. T he results of 834 this intervention should therefore be interpreted primarily as proof-of-principle demonstrating 835 the impact of impaired gas exchange on cardiac function rather than as a direct clinical 836 analogue. Finally, it should be noted that in HFpEF mice, capillary rarefaction was not 837 associated with a corresponding increase in RVSP. This observation aligns with prior work 838 by Duncan Stewart and colleagues showing that RVSP only starts to increase after more than 839 2/3 of the pulmonary microvascular bed has b een lost 48,49. As such, impairments in gas 840 exchange may precede overt PH in the trajectory of HFpEF. 841 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint In summary, our clinical and experimental data converge on a new and previously 842 unrecognized mechanism of exertional dyspnea in heart failure with important functional and 843 therapeutic consequences for disease progression. Through autophagy - and apoptosis-driven 844 loss of lung endothelial cells, HFpEF causes a significant loss in surface area for alveolo -845 capillary gas exchange, thus contributing to impaired diffusion capacity, systemic hypoxemia 846 and exercise intolerance. Hypoxemia, in turn, aggravates LV dysfunction, thereby 847 establishing a feed -forward loop between cardiac and pulmonary pathology. The ability of 848 moderate hyperoxia or inhibition of endothelial autophagy to restore not only blood 849 oxygenation, but also pulmonary capillarization and – importantly - improve cardiac diastolic 850 performance highlights the therapeutic potential of targeting the lung microvasculature and/or 851 systemic oxygenation in heart failure. These findings broaden the mechanistic framework of 852 heart failure beyond the myocardium and identify the pulmonary microcirculation as a critical 853 — and potentially modifiable — determinant of disease progression. 854 855

856 The authors thank all Kuebler and Grune lab members for discussion and insightful 857 comments. We thank Kerstin Riskowski, John Horn, and Katja Dörfel for technical assistance 858 and Dr. Sara Timm and Petra Schrade from the Core Facility of Electron Microscopy f or 859 support. We thank the Advanced Medical Bioimaging Core Facility (AMBIO) of the Charité 860 for support in acquisition of imaging data. The graphical abstract has been created in 861 BioRender. Kocana, C. (2026) https://BioRender.com/rio5nw2. Dyspnea figure in the 862 graphical abstract was created with Illustrae.co. 863 864 Funding 865 C.K., J.L., J.G. and W.M.K. were funded by the Deutsche Forschungsgemeinschaft (DFG, 866 German Research Foundation) - SFB-1470 (Project ID 437531118) A04, B05, Z02. W.M.K. 867 was supported by the DZHK (German Centre for Cardiovascular Research) , funding code s: 868 81Z0100214 and 81X2100294 and by DFG operational grants KU1218/11 -2, KU1218/12-1, 869 KU1218/14-1 and the SFB-1449 (Project ID 431232613 ) B01. J.G., D.F., L.v.O. and N.H. 870 were funded by the Corona-Stiftung Grant S199/10086/2022. L.v.O. was supported by a 871 scholarship from the Sonnenfeld Foundation. J.G. and A.W. were supported by the DZHK , 872 funding code: 81X3100305. C.B. was funded by DFG operation al grant BR5347/4-1. G.G.S 873 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint was supported by grants from DZHK , funding code: 81X3100210, 81X2100282; the DFG 874 SFB-1470 (Project ID 437531118 ) A02, Z01; the European Research Council – ERC StG 875 101078307; and HI-TAC (Helmholtz Institute for Translational AngioCardiScience). 876 877 Author contribution 878 C.K. and L.J. performed and analyzed experiments and interpreted data. C.K. made the 879 figures. N.H., P.S., Q.L., K.K., A.M., D.F., L.v.O., P. -L.P., J.L.G. and A.W. performed 880 experiments and collected data. A.M.C., K.F., E.R., A.K., L.K., D.Z. and V.Z. provided 881 human specimen and data. S.M.T., M.A.H., B.K.B. and S.A.H. performed single -cell RNA 882 sequencing analysis. J.G. and W.M.K. designed experiments. M.U., C.B., J.G. and W.M.K. 883 discussed results and strategy. C.K. and W.M.K. wrote the manuscript with input from all 884 authors. W.M.K. conceived and directed the study. 885 886 Conflict of interest 887 The other authors declare no conflict of interest. G.G.S. reports research agreements and/or 888 speaker honoraria and/or consultancy agreement from Boehringer Ingelheim, Pfizer, Novo 889 Nordisk, e-Therapeutics, Sanofi not related to this work. 890 891 892 893 894 895 896 897 898 899 900 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 901 902 903 904 905 906 907 908 909 910

911 1. Dunlay SM, Roger VL , Redfield MM. Epidemiology of heart failure with preserved ejection 912 fraction. Nat Rev Cardiol. 2017;14(10):591-602. 913 2. Borlaug BA, Sharma K, Shah SJ, Ho JE. Heart failure with preserved ejection fraction: JACC 914 Scientific Statement. J Am Coll Cardiol. 2023;81(18), 1810–1834. 915 3. Ferreira JP, Packer M, Butler J, Vasques -Novoa F, Marques P, Pocock S, Filippatos G, 916 Zannad F, Anker SD. Serum magnesium, outcomes, and the effect of empagliflozin in heart 917 failure with mildly reduced and preserved ejection fraction: findings from EMPEROR -918 Preserved. JACC Heart Fail. 2026;14(2), 102889. 919 4. Solomon SD, Vaduganathan M, Claggett BL, de Boer RA, DeMets D, Hernandez AF, 920 Inzucchi SE, Kosiborod MN, Lam CSP, Martinez F, et al. Baseline characteristics of patients 921 with hf with mildly reduced and preserved ejection fraction : DELIVER trial. JACC Heart 922 Fail. 2022;10(3), 184–197. 923 5. Kitzman DW, Higginbotham MB, Cobb FR, Sheikh KH, Sullivan MJ. Exercise intolerance in 924 patients with heart failure and preserved left ventricular systolic function: failure of the Frank-925 Starling mechanism. J Am Coll Cardiol. 1991;17(5), 1065–1072. 926 6. Munger MA, Stanek EJ, Nara AR, Strohl KP, Decker MJ, Nair RN. Arterial oxygen 927 saturation in chronic congestive heart failure. Am J Cardiol. 1994;73(2), 180–185. 928 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 7. Omar M, Omote K, Sorimachi H, Popovic D, Kanwar A, Alogna A, Reddy YNV, Lim KG, 929 Shah SJ, Borlaug BA. Hypoxaemia in patients with heart failure and preserved ejection 930 fraction. Eur J Heart Fail. 2023;25(9), 1593–1603. 931 8. Moesgaard J, Kristensen JH, Malczynski J, Holst -Hansen C, Rees SE, Murley D, Andreassen 932 S, Frokjaer JB, Toft E. Can new pulmonary gas exchange parameters contribute to evaluation 933 of pulmonary congestion in left-sided heart failure? Can J Cardiol. 2009;25(3), 149–155. 934 9. Fermoyle CC, Stewart GM, Borlaug BA, Johnson BD . Simultaneous measurement of lung 935 diffusing capacity and pulmonary hemodynamics reveals exertional alveolar -capillary 936 dysfunction in heart failure with preserved ejection fraction . J Am Heart Assoc . 2021;10(16), 937 e019950. 938 10. Olson TP, Johnson BD, Borlaug BA. Impaired pulmonary diffusion in heart failure with 939 preserved ejection fraction. JACC Heart Fail. 2016;4(6), 490–498. 940 11. Fayyaz AU, Edwards WD, Maleszewski JJ, et al. Global pulmonary vascular remodeling in 941 pulmonary hypertension associated with heart failure and preserved or reduced ejection 942 fraction. Circulation. 2018;137(17):1796-1810. 943 12. Hunt JM, Bethea B, Liu X, Gandjeva A, Mammen PP, Stacher E, Gandjeva MR, Parish E, 944 Perez M, Smith L, et al. Pulmonary veins in the normal lung and pulmonary hypertension due 945 to left heart disease. Am J Physiol Lung Cell Mol Physiol. 2013;305(10), L725–736. 946 13. Jaeschke L, Koçana C, Chitroceanu AM, Winkler A, Kleitke H, Fahjen P, Faidel D, Asmus E, 947 Meiser M, Stamatiades EG, et al. Myeloid cell recruitment propels right ventricular 948 dysfunction in HFpEF via sterile inflammation. bioRxiv [Preprint]. 949 2025;22:2025.11.21.689863. 950 14. Lai YC, Tabima DM, Dube JJ, Hughan KS, Vanderpool RR, Goncharov DA, St Croix CM, 951 Garcia-Ocaña A, Goncharova EA, Tofovic SP, et al. SIRT3-AMP-activated protein kinase 952 activation by nitrite and metformin improves hyperglycemia and normalizes pulmonary 953 hypertension associated with heart failure with preserved ejection fraction . Circulation. 954 2016;133(8):717-31. 955 15. Yin J, Kukucka M, Hoffmann J, Sterner-Kock A, Burhenne J, Haefeli WE, Kuppe H, Kuebler 956 WM. Sildenafil preserves lung endothelial function and prevents pulmonary vascular 957 remodeling in a rat model of diastolic heart failure. Circ Heart Fail. 2011;4: 198-206 958 16. Sang P, Kucherenko MM, Yao J, Li Q, Simmons S, Kuebler WM, Knosalla C. A model of 959 reverse vascular remodeling in pulmonary hypertension due to left heart disease by aortic 960 debanding in rats. J Vis Exp. 2022;1;(181). 961 17. Scherle W. A simple method for volumetry of organs in quantitative stereology. Mikroskopie. 962 1970;26(1):57-60. 963 18. Mattfeldt T, Mall G, Gharehbaghi H, Möller P. Estimation of surface area and length with the 964 orientator. J Microsc. 1990;159(Pt 3):301-17. 965 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 19. Schneider JP, Ochs M. Alterations of mouse lung tissue dimensions during processing for 966 morphometry: a comparison of methods. Am J Physiol Lung Cell Mol Physiol . 967 2014;306(4):L341-50. 968 20. Kling KM, Lopez -Rodriguez E, Pfarrer C, Mühlfeld C, Brandenberger C. Aging exacerbates 969 acute lung injury-induced changes of the air-blood barrier, lung function, and inflammation in 970 the mouse. Am J Physiol Lung Cell Mol Physiol. 2017;312(1):L1-L12. 971 21. Willführ A, Brandenberger C, Piatkowski T, Grothausmann R, Nyengaard JR, Ochs M, 972 Mühlfeld C. Estimation of the number of alveolar capillaries by the Euler number (Euler -973 Poincare characteristic). Am J Physiol Lung Cell Mol Physiol. 2015;309(11), L1286–1293. 974 22. Sterio DC. The unbiased estimation of number and sizes of arbitrary particles using the 975 disector. J Microsc. 1984;134(Pt 2):127-36. 976 23. Tschanz SA, Burri PH, Weibel ER. A simple tool for stereological assessment of digital 977 images: the STEPanizer. J Microsc. 2011;243(1):47-59. 978 24. Zhang Q, Yaoita N, Tabuchi A, Liu S, Chen SH, Li Q, Hegemann N, Li C, Rodor J, Timm S, 979 Laban H , et al. Endothelial heterogeneity in the response to autophagy drives small vessel 980 muscularization in pulmonary hypertension. Circulation. 2024;150(6), 466–487. 981 25. Nelin LD, Roerig DL, Rickaby DA, Linehan JH, Dawson CA. Influence of flow on 982 pulmonary vascular surface area inferred from blue dextran efflux data. J Appl Physiol 983 (1985). 1992;72(3), 874–880. 984 26. Agrawal V, Kropski JA, Gokey JJ, Kobeck E, Murphy MB, Murray KT, Fortune NL, Moore 985 CS, Meoli DF, Monahan K , et al . Myeloid cell derived IL1β contributes to pulmonary 986 hypertension in HFpEF. Circ Res. 2023;133(11), 885–898. 987 27. O'Donnell DE, Banzett RB, Carrieri -Kohlman V, Casaburi R, Davenport PW, Gandevia SC, 988 Gelb AF, Mahler DA, Webb KA. Pathophysiology of dyspnea in chronic obstructive 989 pulmonary disease: a roundtable. Proc Am Thorac Soc. 2007;4(2), 145–168. 990 28. Presson RG Jr , Todoran T M, De Witt BJ, McMurtry IF, Wagner WW Jr. Capillary 991 recruitment and transit time in the rat lung. J Appl Physiol (1985). 1997;83(2), 543–549. 992 29. Lewis NP, Banning AP, Cooper JP, Sundar AS, Facey PE, Evans WD, Henderson AH. 993 Impaired matching of perfusion and ventilation in heart failure detected by 133xenon. Basic 994 Res Cardiol. 1996;91 Suppl 1:45-9. 995 30. Wada O, Asanoi H, Miyagi K, Ishizaka S, Kameyama T, Seto H, Sasayama S. Importance of 996 abnormal lung perfusion in excessive exercise ventilation in chronic heart failure. Am Heart J. 997 1993;125(3):790-8. 998 31. Agostoni P, Magini A, Andreini D, Contini M, Apostolo A, Bussotti M, Cattadori G, Palermo 999 P. Spironolactone improves lung diffusion in chronic heart failure. Eur Heart J. 1000 2005;26(2):159-64. 1001 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 32. Guazzi M, Pontone G, Brambilla R, Agostoni P, Reina G. Alveolar--capillary membrane gas 1002 conductance: a novel prognostic indicator in chronic heart failure. Eur Heart J . 2002;23(6), 1003 467–476. 1004 33. Kingsbury MP, Huang W, Donnelly JL, Jackson E, Needham E, Turner MA, Sheridan DJ. 1005 Structural remodelling of lungs in chronic heart failure. Basic Res Cardiol. 2003;98(5), 295–1006 303. 1007 34. Deis T, Balling L, Rossing K, Wolsk E, Perch M, Gustafsson F. Lung diffusion capacity in 1008 advanced heart failure: relation to central haemodynamics and outcome. ESC Heart Fail . 1009 2019;6(2), 379–387. 1010 35. Hoeper MM, Meyer K, Rademacher J, Fuge J, Welte T, Olsson KM. Diffusion capacity and 1011 mortality in patients with pulmonary hypertension due to heart failure with preserved ejection 1012 fraction. JACC Heart Fail. 2016;4(6), 441–449. 1013 36. Agostoni P, Bussotti M, Cattadori G, Margutti E, Contini M, Muratori M, Marenzi G, 1014 Fiorentini C. Gas diffusion and alveolar -capillary unit in chronic heart failure. Eur Heart J . 1015 2006;27(21), 2538–2543. 1016 37. Kaestle SM, Reich CA, Yin N, Habazettl H, Weimann J, Kuebler WM. Nitric oxide -1017 dependent inhibition of alveolar fluid clearance in hydrostatic lung edema. Am J Physiol Lung 1018 Cell Mol Physiol. 2007;293(4):L859-69. 1019 38. Kerem A, Yin J, Kaestle SM, Hoffmann J, Schoene AM, Singh B, Kuppe H, Borst MM, 1020 Kuebler WM. Lung endothelial dysfunction in congestive heart failure: role of impaired Ca 2+ 1021 signaling and cytoskeletal reorganization. Circ Res. 2010;106(6), 1103–1116. 1022 39. Mühlfeld C, Weibel ER, Hahn U, Kummer W, Nyengaard JR, Ochs M. Is length an 1023 appropriate estimator to characterize pulmonary alveolar capillaries? A critical evaluation in 1024 the human lung. Anat Rec (Hoboken). 2010;293(7), 1270–1275. 1025 40. Tabuchi A, Styp-Rekowska B, Slutsky AS, Wagner PD, Pries AR, Kuebler WM. Precapillary 1026 oxygenation contributes relevantly to gas exchange in the intact lung. Am J Respir Crit Care 1027 Med. 2013;188(4), 474–481. 1028 41. Yuan K, Shamskhou EA, Orcholski ME, Nathan A, Reddy S, Honda H, Mani V, Zeng Y, 1029 Ozen MO, Wang L , et al . Loss of endothelium -derived Wnt5a is associated with reduced 1030 pericyte recruitment and small vessel loss in pulmonary arterial hypertension . Circulation. 1031 2019;139(14):1710-1724. 1032 42. Kern AL, Park DH, Fuge J, Hohlfeld JM, Wacker F, Hoeper MM, Olsson KM, Vogel -1033 Claussen J. Loss of pulmonary capillaries in idiopathic pulmonary arterial hypertension with 1034 low diffusion capacity is accompanied by early diffuse emphysema detected by 129Xe MRI. 1035 Eur Radiol. 2025;35(6):3010-3020. 1036 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 43. Gillich A, Zhang F , Farmer CG, Travaglini KJ, Tan SY, Gu M, Zhou B, Feinstein JA, 1037 Krasnow MA, Metzger RJ. Capillary cell -type specialization in the alveolus. Nature. 1038 2020;586(7831):785-789. 1039 44. Kolesnichenko O A, Whitsett JA, Kalin TV, Kalinichenko VV. Therapeutic Potential of 1040 Endothelial Progenitor Cells in Pulmonary Diseases. Am J Respir Cell Mol Biol. 1041 2021;65(5):473-488. 1042 45. Denton D, Kumar S. Autophagy -dependent cell death. Cell Death Differ . 2019;26(4), 605–1043 616. 1044 46. Summer G, Kuhn AR, Munts C, Miranda-Silva D, Leite-Moreira AF, Lourenco AP, Heymans 1045 S, Falcao -Pires I, van Bilsen M. A directed network analysis of the cardiome identifies 1046 molecular pathways contributing to the development of HFpEF. J Mol Cell Cardiol . 1047 2020;144, 66–75. 1048 47. Muller J, Lichtblau M, Saxer S, Calendo LR, Carta AF, Schneider SR, Berlier C, Furian M, 1049 Bloch KE, Schwarz EI, Ulrich S. Effect of breathing oxygen -enriched air on exercise 1050 performance in patients with pulmonary hypertension due to heart failure with preserved 1051 ejection fraction: A randomized, placebo-controlled, crossover trial . Front Med (Lausanne) . 1052 2021; 8, 692029. 1053 48. Kuebler WM, Nicolls MR, Olschewski A, Abe K, Rabinovitch M, Stewart D, Chan SY, 1054 Morrell NW, Archer SL, Spiekerkoetter E. A pro -con debate: current controversies in PAH 1055 pathogenesis at the American Thoracic Society International Conference in 2017. Am J 1056 Physiol Lung Cell Mol Physiol. 2018;315(4), L502–L516. 1057 49. Deng Y, Chaudhary KR, Yang A, Rowe KJ, Stewart DJ. Protracted endothelial cell apoptosis 1058 leads to direct microvascular loss as a major mechanism of severe pulmonary arterial 1059 hypertension in the rat SU5416 -hypoxia mode (Abstract). Am J Respir Crit Care Med . 1060 2017;195: A7214. 1061 1062 1063 1064 1065 1066 1067 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 Figures and Figure Legends 1078 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 1079 Figure 1. Impaired oxygenation in HFpEF patients 1080 A total of 234 HFpEF patients of the clinical cohort of the CRC1470 were stratified according to New 1081 York Heart Association (NYHA) class I (n=25), II (n=157), and III (n=52). A. Echocardiographic 1082 analysis of left ventricular (LV) ejection fraction (LVEF), ratio of early to late LV filling velocity (E/A 1083 ratio), ratio of early diastolic mitral inflow velocity over early diastolic mitral annular tissue velocity 1084 (E/e’ ratio), global longitu dinal strain (GLS) and end -diastolic volume (EDV); B. Arterial oxygen 1085 saturation (SaO 2) at rest and during cardiopulmonary exercise testing at maximal load; C. Diffusing 1086 capacity of the lungs for carbon monoxide after correction for hemoglobin (DLCOc -SB); D. 1087 Estimated dead space relative to tidal volume (V D/VT) at rest and during cardiopulmonary exercise 1088 testing at maximal load. E. Correlation analysis of E/e´ ratio and EDV against DLCOc -SB. Dots 1089 represent individual patients. Data are presented as median with interquartile range (IQR) . Kruskal-1090 Wallis test followed by uncorrected Dunn’s multiple comparisons test; r s: Spearman coefficient of 1091 correlation; P-values as indicated. 1092 1093 1094 1095 1096 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 1097 Figure 2. Lung microvascular rarefaction in HFpEF animals 1098 A. Experimental protocol for HFpEF induction by SU5416 (s.c. 100 mg/kg) in ZSF1 obese rats. B. 1099 Echocardiographic analysis of left ventricular (LV) ejection fraction (LVEF), ratio of early to late LV 1100 filling velocity (E/A ratio), ratio of early diastolic mitral inflow velocity over early diastolic mitral 1101 annular tissue velocity (E/e’ ratio) and global longitudinal strain (GLS) in ZSF1 lean and SU5416 -1102 treated ZSF1 obese (n=13 -15 each) rats. C. Right ventricular systolic pressure (RVSP) in ZSF1 lean 1103 and SU5416 -treated ZSF1 obese rats (n=12 each). D. Representative micro -computed tomography 1104 (µCT) reconstructions of the pulmonary vasculature in ZSF1 lean (left) and SU5416 -treated Z SF1 1105 obese (right) rats and quantification of pulmonary vascular volume by µCT in vessels <250 µm in 1106 diameter in ZSF1 lean (n=4) and SU5416-treated ZSF1 obese (n=5) rats. Vessels <110 µm in diameter 1107 were excluded from the analysis due to limited resolution. E. Stereological quantification of capillary 1108 volume density within the interalveolar septa of the lung (V v(cap/sept)) in ZSF1 lean and SU5416 -1109 treated ZSF1 obese rats (n=3 each). F. Experimental protocol for HFpEF induction by high fat diet 1110 and Nω-Nitro-L-Arginin-Methylester-hydrochlorid ( L-NAME) in C57BL/6J mice. G. 1111 Echocardiographic analysis of LVEF, E/A ratio, E/e’ ratio and GLS in control and HFpEF mice (n=15 1112 each). H. RVSP in control and HFpEF mice (n=15 each). I. Flow cytometric quantification of 1113 CD31⁺/GS-IB4+ endothelial cells as a function of HFpEF progression over time (n=6 for each time 1114 point), and stereological quantification of total lung capillary number in control (n=10) and HFpEF 1115 (n=12) mice. J. Running distance and maximal oxygen consumption (VO₂max) of control and HFpEF 1116 mice (n=12 each) in treadmill exercise tests. Dots represent individual animals. Data are presented as 1117 median with interquartile range (IQR ; all data except for D,I ) or mean ± SEM (D,I). Mann–Whitney 1118 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint U-test (all data except for time course in I) or Kruskal -Wallis test followed by uncorrected Dunn’s 1119 multiple comparisons test; P-values as indicated or *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 1120 0.0001. 1121 1122 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 1123 Figure 3. Lung microvascular rarefaction in AoB rats 1124 A. Experimental protocol for left heart failure induction by aortic banding (AoB) in rats. B. 1125 Echocardiographic analysis of left ventricular (LV) ejection fraction (LVEF), ratio of early to late LV 1126 filling velocity (E/A ratio), rat io of early diastolic mitral inflow velocity over early diastolic mitral 1127 annular tissue velocity (E/e’ ratio) and global longitudinal strain (GLS) in sham (n=14) and AoB 1128 (n=16) rats. C. Right ventricular systolic pressure (RVSP) in sham (n=14) and AoB (n=1 6) rats. D. 1129 Representative micro -computed tomography (µCT) reconstructions of the pulmonary vasculature in 1130 sham (left) and AoB (right) rats and quantification of pulmonary vascular volume by µCT in vessels 1131 <250 µm in diameter in sham (n=4) and AoB (n=3) ra ts. Vessels <110 µm in diameter were excluded 1132 from the analysis due to limited resolution. E. Pulmonary capillary surface area in sham (n=10) and 1133 AoB (n=10) rats as determined by blue dextran efflux, flow cytometric quantification of CD31⁺/GS -1134 IB4+ endothelial cells in sham (n=6) and AoB (n=7) rats, and stereological quantification of total lung 1135 capillary number in sham (n=7) and AoB (n=7) rats . Dots represent individual animals. Data are 1136 presented as median with interquartile range (IQR ; all data except for D) or mean ± SEM (D). Mann–1137 Whitney U-test; P-values as indicated or *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001. 1138 1139 1140 1141 1142 1143 1144 1145 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 1146 Figure 4. Single-cell RNA sequencing identifies propensity for regulated cell death and 1147 autophagy in general capillary cells 1148 A. Uniform Manifold Approximation and Projection (UMAP) plot shows different cell clusters from 1149 integrated analysis of murine lungs after 2 weeks of HFpEF induction by high fat diet/L -NAME. B. 1150 Proportion of general capillary endothelial cells (gCap), aerocytes (aCap), and pericytes in HFpEF 1151 versus control mice (n=4 each). C. Net expression of an aggregate set of marker genes for regulated 1152 cell death pathway in gCap, aCap, and pericytes in HFpEF versus control mice (n=4 each). D. Net 1153 expression of an aggregate set of autophagy marker genes in gCap, aCap and pericytes in HFpEF 1154 versus control mice (n=4 each). E. Correlation analysis of autophagy marker genes vs. regulated cell 1155 death pathway marker genes in gCap and aCap. Dots represent indivi dual animals (B) or cells (C-E), 1156 respectively. Data are presented as median with interquartile range (IQR) . Mann–Whitney U-test; rs: 1157 Spearman coefficient of correlation; P-values as indicated. 1158 1159 1160 1161 1162 1163 1164 1165 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 1166 Figure 5. HFpEF drives autophagy-dependent cell death in lung microvascular endothelial cells 1167 A. Experimental protocol for HFpEF induction by high fat diet and N ω-Nitro-L-Arginin-Methylester-1168 hydrochlorid (L -NAME) in C57BL/6J mice. B. Flow cytometric analysis of autophagic flux in 1169 CD31⁺/GS-IB4+ lung microvascular endothelial cells as a function of HFpEF progression over time 1170 (n=6-13 for each time point). C. Representative transmission electron micrographs of lung endothelial 1171 cells at (from top to bottom) increasin g levels of magnification showing autophagosomes (marked by 1172 black arrowheads) in HFpEF yet not in control mice. D,E. Representative images and quantitative 1173 analysis of cleaved caspase 3 (Cl -Casp3) ( red; D) and cleaved poly[ADP -ribose] polymerase 1(Cl -1174 PARP1) (red; E) staining in lung CD31 + endothelial cells (green; nuclei counterstained with DAPI in 1175 blue) from control (week 12) and HFpEF mice (week 0, 1, and 12) (n=10 each). Data are shown as 1176 percentage of vessels containing at least one Cl -Cas3 or Cl -PARP1 positive endothelial cell, 1177 respectively. F. Experimental protocol for HFpEF induction by high fat diet and L -NAME in 1178 endothelial cell-specific Atg7-deficient (Atg7EN-KO) and corresponding wild type ( Atg7EN-WT) mice. G. 1179 Stereological quantification of total lung capillary number in Atg7EN-WT (n=5 each) and Atg7EN-KO (n=8 1180 each) HFpEF or control mice. Dots represent individual animals. Data are presented as median with 1181 interquartile range (IQR) . Kruskal-Wallis test followe d by uncorrected Dunn’s multiple comparisons 1182 test; P-values as indicated. 1183 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 1184 Figure 6. Systemic hypoxemia aggravates experimental HFpEF 1185 A. Experimental protocol for induction of left heart failure by aortic banding (AoB) in rats. B. Arterial 1186 oxygen saturation (SaO ₂), partial pressure of arterial oxygen (PaO ₂), and mixed venous oxygen 1187 saturation (PvO ₂) in sham (n=15) and AoB (n=16) rats. C. Experimental protocol for HFpEF 1188 induction by high fat diet and Nω-Nitro-L-Arginin-Methylester-hydrochlorid (L-NAME) in C57BL/6J 1189 mice. D. SaO₂ at rest and after exercise in control and HFpEF mice (n=8 each). E. Experimental 1190 protocol for HFpEF induction by high fat diet and L-NAME in endothelial cell-specific Atg7-deficient 1191 (Atg7EN-KO) and corresponding wild type (Atg7EN-WT) mice. F. SaO₂ at rest and after exercise in Atg7EN-1192 WT and Atg7EN-KO mice (n=5 -7 each). G. Running distance and maximal oxygen consumption 1193 (VO₂max) of Atg7EN-WT(n=5 each) and Atg7EN-KO (n=10 each) HFpEF or control mice in treadmill 1194 exercise tests. H. Experimental protocol for hypoxia exposure (10%O 2 for 2 weeks) in HFpEF mice. 1195 I. Echocardiographic analysis of left ventricular (LV) ejection fraction (LVEF), ratio of early to late 1196 LV filling velocity (E/A ratio), ratio of early diastolic mitral inflow velocity over early diastolic mitral 1197 annular tissue velocity (E/e’ ratio) and glo bal longitudinal strain (GLS) in control, HFpEF and 1198 HFpEF+Hypoxia mice (n=15 -21 each). Dots represent individual animals. Data are presented as 1199 median with interquartile range (IQR) . Mann–Whitney U-test (B,D) or Kruskal -Wallis test followed 1200 by uncorrected Dunn’s multiple comparisons test (F,G,I); P-values as indicated. 1201 1202 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 1203 Figure 7. Loss of endothelial autophagy and moderate hyperoxia mitigate experimental HFpEF 1204 A. Experimental protocol for HFpEF induction by high fat diet and N ω-Nitro-L-Arginin-Methylester-1205 hydrochlorid (L -NAME) in endothelial cell -specific Atg7-deficient ( Atg7EN-KO) and corresponding 1206 wild type ( Atg7EN-WT) mice. B. Echocardiographic analysis of left ventricular (LV) ejection fraction 1207 (LVEF), ratio of early to late LV filling veloc ity (E/A ratio), ratio of early diastolic mitral inflow 1208 velocity over early diastolic mitral annular tissue velocity (E/e’ ratio) and global longitudinal strain 1209 (GLS) in Atg7EN-WT (n=5 each) and Atg7EN-KO (n=10 each) HFpEF or control mice. C. Experimental 1210 protocol for moderate hyperoxia exposure (40% O 2 for 2 weeks) in HFpEF mice. D. 1211 Echocardiographic analysis of LVEF, E/A ratio, E/e’ ratio and GLS in control, HFpEF and 1212 HFpEF+Hyperoxia mice (n=16 each). E. Stereological quantification of total lung capillary number in 1213 HFpEF (n=10) and HFpEF+Hyperoxia (n=10) mice. F. Arterial oxygen saturation (SaO ₂) at rest and 1214 after exercise in control, HFpEF and HFpEF+Hyperoxia mice (n=16 each). G. Running distance of 1215 control, HFpEF and HFpEF+Hyperoxia mice (n=16 each) in treadmill exercise tests. Dots represent 1216 individual animals. Data are presented as median with interquartile range (IQR) . Mann–Whitney U-1217 test (E) or Kruskal -Wallis test followed by uncorrected Dunn’s multiple comparisons test (all data 1218 except E); P-values as indicated. 1219 1220 1221 1222 1223 1224 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint 1225 1226 Figure 8. Graphical abstract 1227 HFpEF causes lung microvascular rarefaction by driving autophagy -dependent endothelial cell death. 1228 Resulting hypoxemia contributes to clinical symptoms of dyspnea and aggravates HFpEF progression 1229 in a vicious circle. 1230 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted March 9, 2026. ; https://doi.org/10.64898/2026.03.05.709974doi: bioRxiv preprint