背景与目的

重症社区获得性肺炎（severe community acquired pneumonia，SCAP）是重症监护病房（intensive care unit，ICU）内最常见的疾病之一，具有发病率高、死亡率高的特点，在全球范围内造成了极大的卫生和经济负担。对SCAP患者进行早期评估预警，识别预后不良的风险，进而开展个体化精准分层诊疗，可以有效改善患者预后、减轻医疗负担。因此，探究SCAP患者的预后不良危险因素，建立性能较好的预测模型是临床的迫切需求。

近年来的研究进展颠覆了传统对肺炎发病机制的理解：SCAP发病过程不是简单的致病微生物对无菌空间的入侵，而是致病微生物、宿主反应及固有肺微生物组稳态失衡三者相互作用的结果，伴随微生物的入侵，宿主反应及肺微生物组失衡均可能与SCAP预后相关。目前用于预后模型建立的参数主要来源于基础疾病、炎症指标、影像学特征等常规临床资料，无法反映宿主反应性差异及宿主微生物组失衡信息，使得这些模型在临床应用中群体偏倚度大，无法准确预测所有SCAP患者的临床预后。因此，本研究基于大样本SCAP前瞻性多中心队列，利用高通量测序的多组学技术研究SCAP患者的宿主反应和肺微生物组特征，建立多维度的SCAP预后模型，并对其预后不良机制进行研究。

方法

第一部分

此部分为一项前瞻性多中心队列研究，纳入2021年1月1日至2022年6月30日中国大陆地区8个省/直辖市共10家三级甲等医院ICU内239例SCAP患者，收集患者的临床资料并采集支气管肺泡灌洗液（bronchoalveolar lavage fluid，BALF）样本进行宏基因组二代测序（metagenomic next generation sequencing，mNGS）及常规病原学检测，以28天死亡率为主要研究终点，分析比较预后不同的SCAP患者的临床特征和病原学分布差异，并建立基于临床数据的预后预测模型。同时，分析比较了免疫抑制和非免疫抑制SCAP患者的临床特征，在两组人群中分别建立预后预测的临床模型。

第二部分

此部分研究分析了第一部分SCAP患者BALF标本进行的DNA和RNA的mNGS检测结果。从宿主转录组和肺微生物组角度探究SCAP患者的特征，基于宏转录组测序数据，分析SCAP患者的宿主转录组特征（差异基因，免疫细胞组成，差异基因富集的信号通路等）；分析宏基因组测序数据，探究SCAP患者的肺部微生物组特征（相对丰度和构成，组间多样性比较和特征微生物等）。以28天死亡率为主要研究终点，分别探究预后不同的免疫抑制SCAP患者和非免疫抑制SCAP患者的宿主转录组和肺微生物组特征。最后基于宿主转录组测序数据，将患者以7：3随机分为训练集和内部验证集，运用机器学习随机森林算法分别构建免疫抑制和非免疫抑制SCAP患者28天死亡率的预后预测模型。以本研究团队牵头的另一项前瞻性多中心SCAP队列研究纳入的2023年4月1日至2023年7月31日期间全国22家医院ICU内的105例SCAP患者作为外部验证队列，对模型进行外部验证。

第三部分

前瞻性纳入2022年9月至2023年2月中日友好医院普通病房肺炎克雷伯菌（Klebsiella pneumoniae，KP）-社区获得性肺炎（community acquired pneumonia，CAP）和内科重症监护室（medical intensive care unit，MICU）内KP-SCAP的患者，收集两组患者的BALF样本，进行非靶向代谢组学分析，筛选差异代谢产物及调控信号通路，同时运用ELISA方法进行琥珀酸验证和炎症因子检测。运用细菌内毒素脂多糖（lipopolysaccharides，LPS）刺激上皮细胞检测琥珀酸和炎症因子表达水平以进行体外验证；同时用琥珀酸刺激肺泡巨噬细胞作为实验组与未干预对照组进行转录组学测序，筛选关键调控信号通路；运用免疫荧光、免疫印迹杂交和逆转录-聚合酶链反应（reverse transcription-polymerase chain reaction，RT-PCR）方法等检测琥珀酸刺激肺泡巨噬细胞后琥珀酸受体1（succinatereceptor1，SUCNR1）和铁死亡的关键基因蛋白与mRNA的表达。

结果

第一部分

1. 本部分研究纳入的239例SCAP患者中，66例（27.61%）在入ICU后28天内死亡。与存活组患者相比，28天死亡组患者年龄更大（P<0.001），入ICU时急性生理学与慢性健康状况II评分（acute physiology and chronic health evaluation Ⅱ，APACHE Ⅱ）和序贯器官功能衰竭评分（sequential organ failure assessment，SOFA）分数更高（P<0.001），发生脓毒症、感染性休克、急性肾损伤、使用血管活性药物和有创机械通气（invasive mechanical ventilation，IMV）比例更高（P<0.05）；实验室检查方面，氧合指数更低，乳酸和D-二聚体水平更高（P<0.05）；预后方面，28天死亡组患者ICU期间接受IMV治疗患者比例更高（P<0.001）。

2. SCAP患者的BALF病原学检测阳性率为79.50%，发生混合感染的患者比例为37.66%（90/239）。与存活组患者相比，28天死亡组患者发生细菌感染的比例更高（P<0.05），发生细菌-细菌混合感染的比例更高（P<0.05）。本研究SCAP患者中排名前五位的病原体为耶氏肺孢子菌、肺炎克雷伯菌、巨细胞病毒、曲霉菌和金黄色葡萄球菌。

3. SCAP患者28天死亡率的独立危险因素包括年龄（OR：1.03，95% CI：1.01-1.06，P=0.015）、乳酸水平（OR：1.31，95% CI：1.02-1.68，P=0.035）、入ICU时接受IMV治疗（OR：2.88，95% CI：1.17-7.09，P=0.021）和入ICU时急性肾损伤（OR：2.20，95% CI：1.01-4.81，P=0.049），以此建立的Logistic回归预后预测模型ROC曲线下面积（area under ROC curve，AUC）为0.831。

4. 免疫抑制和非免疫抑制SCAP患者在临床特征和病原学分布方面具有显著差异。免疫抑制SCAP患者28天死亡率的独立危险因素是D-二聚体水平，平均动脉压增高是保护性因素，建立的预后预测模型AUC为0.910；非免疫抑制SCAP患者28天死亡率的独立危险因素包括年龄、乳酸水平和入ICU时接受IMV治疗，以此建立的预后预测模型AUC为0.825。

第二部分

1. SCAP患者宿主转录组特征：共筛选出4193个差异表达基因，差异基因富集分析（gene set enrichment analysis，GSEA）提示免疫抑制患者趋化因子信号通路、细胞因子-细胞因子受体相互作用通路、IL-17信号通路和NF-κB 信号通路等免疫和炎症反应相关通路表达下调（P<0.05）；免疫细胞分析提示，免疫抑制患者中性粒细胞表达降低，但调节性T细胞表达增加（P<0.05）。

SCAP患者肺微生物组特征：免疫抑制患者以链球菌属、葡萄球菌属和普雷沃氏菌属为主，非免疫抑制患者以链球菌属、克雷伯菌属和不动杆菌属为主，免疫抑制患者的 多样性更高（P<0.05），两组患者的 多样性具有显著差异（P=0.001）。

2. 不同预后的免疫抑制SCAP患者宿主转录特征：共筛选到差异表达基因共1415个，差异基因GSEA富集分析提示28天死亡组患者早期氧化磷酸化等代谢相关通路被激活，而细胞因子受体相互作用、抗原加工提呈、T细胞受体信号通路和NK细胞介导细胞毒性通路等免疫反应通路被抑制（P<0.05）。

不同预后的免疫抑制SCAP患者肺微生物组特征：两组患者多样无显著差异，而多样性存在差异（P=0.011）。小韦荣氏球菌等定植菌与APACHE II评分和IMV时间等预后指标显著相关（P<0.05）。

预后预测模型：基于宿主转录组建立预测28天死亡率的7-基因分类器的AUC为0.947（95% CI，0.928-0.966），准确率为0.893，敏感性为86.40%，特异性为89.40%，内部验证集AUC为0.722，外部验证集AUC为0.830。

3. 不同预后的非免疫抑制SCAP患者宿主转录特征：宿主转录组共筛选到差异表达基因共2130个，差异基因GSEA富集分析提示TNF信号通路、IL-17信号通路和氧化磷酸化在28天死亡组中高表达，而ABC转运蛋白通路低表达（P<0.05）。

不同预后的非免疫抑制SCAP患者肺微生物组特征：两组患者肺微生物组构成不同，28天死亡组患者BALF中病原体相对丰度较高，多样无显著差异，多样性具有显著差异（P=0.001）。KP、基质金属蛋白酶8（matrix metallopeptidase 8，MMP8）基因和集落刺激因子3（colony stimulating factor 3，CSF3）基因与APACHE II 和SOFA评分以及临床预后指标显著相关（P<0.05）。

预后预测模型：基于宿主转录组建立预测28天死亡率的7-基因分类器的AUC为0.839（95% CI，0.817-0.861），准确率为0.786，敏感性为91.60%，特异性为63.40%，内部验证集AUC为0.838，外部验证集AUC为0.739。

第三部分

1. 本研究共纳入12例KP-CAP和13例KP-SCAP患者，分离患者BALF样本上清进行非靶向代谢组学测序结果发现，琥珀酸在SCAP组的BALF中显著上升（P=0.02）。正离子代谢产物富集结果主要与谷胱甘肽代谢和NOD样受体信号通路相关，而负离子代谢产物富集结果与HIF-1信号通路、PI3K-Akt信号通路、cAMP信号通路和cGMP-PKG信号通路相关（P<0.05）。

2. 体外实验部分，随着LPS浓度增加，肺泡上皮细胞内和向细胞外分泌的琥珀酸水平均升高，向细胞外分泌的炎症因子白细胞介素1β（interleukin 1β，IL-1β）和肿瘤坏死因子α（tumor necrosis factor α，TNF-α）水平增高（P<0.05）。

3. 琥珀酸刺激肺泡巨噬细胞与未干预对照组的转录组学测序共筛选出792个差异基因，富集到PI3K-Akt信号通路和MAPK信号通路等，将差异基因在铁死亡基因集中进行分析显示，差异基因在铁死亡调节基因集和未分类的铁死亡基因集中显著富集（P<0.05），提示铁死亡可能是琥珀酸在肺泡巨噬细胞中的作用机制。

4. 琥珀酸刺激后的肺泡巨噬细胞中，SUCNR1定位于细胞膜，在蛋白和mRNA水平表达增加，铁死亡关键基因酰基辅酶A合成酶长链家族成员4（acyl-CoA synthetase long chain family member 4，ACSL4）和铁蛋白重链1（ferritin heavy chain 1，FTH1）在蛋白和mRNA水平表达均增高，溶质载体家族7成员11（solute carrier family 7 member 11，SLC7A11）和谷胱甘肽过氧化物酶4 （glutathione peroxidase 4，GPX4）在蛋白和mRNA水平表达下降（P<0.05），验证了琥珀酸可促进肺泡巨噬细胞发生铁死亡。

结论

1. 预后不同的SCAP患者在临床特征和病原学分布等方面存在显著差异，以临床指标建立的模型能较好的预测SCAP患者的预后。

2. 基于宿主转录组数据构建的SCAP患者预后预测模型具有较高的敏感性和特异性，优于现有临床评分，且经过外部队列验证预测性能较好。

3. 琥珀酸可能是CAP患者病情进展、预后不良的生物标志物之一，其作用机制可能是调控肺泡巨噬细胞发生铁死亡。

Background and purpose

Severe community-acquired pneumonia (SCAP) is one of the most common diseases in intensive care unit (ICU), characterized by high morbidity and mortality, resulting in a significant health and economic burden worldwide. Early assessment and early warning of SCAP patients, identification of the risk of poor prognosis, and then individualized and precise stratified diagnosis and treatment can effectively improve patient prognosis and reduce medical burden. Therefore, it is an urgent clinical need to explore the risk factors of poor prognosis in patients with SCAP and establish a better predictive model.

Recent research advances have overturned the traditional understanding of the mechanism of pneumonia: the pathogenesis of SCAP is not the invasion of pathogens into sterile space, but the result of the interaction of pathogens, host response and the homeostasis imbalance of the inherent lung microbiome. With the invasion of pathogens, host response and the imbalance of the lung microbiome may be related to the prognosis of SCAP. At present, the parameters used to establish prognostic models are mainly derived from routine clinical data such as underlying diseases, inflammatory indicators, and imaging features, which cannot reflect host response differences and host microbiome imbalances. As a result, these models have a large population bias in clinical application and cannot accurately predict the clinical prognosis of all SCAP patients. Therefore, based on a prospective multicenter cohort of large samples of SCAP, this study utilized multi-omics technology of high-throughput sequencing to explore the host response and lung microbiome of SCAP patients, established multidimensional prognostic models of SCAP, and studied the mechanism of its poor prognosis.

Method

Part I

This part was a prospective multicenter cohort study. 239 patients with SCAP were enrolled from January 1, 2021 to June 30, 2022 in 10 ICUs within China. The clinical data of the patients were collected, and the bronchoalveolar lavage fluid (BALF) samples were collected for metagenomic next-generation sequencing (mNGS) and conventional microbiological tests. With 28-day mortality as the primary endpoint, the clinical characteristics and pathogen distribution of SCAP patients with different prognosis were compared, and a prognostic predictive model based on clinical data was established. At the same time, the clinical characteristics of immunocompromised and non-immunocompromised SCAP patients were compared, and the clinical models for predicting the prognosis of the two groups were established.

Part II

This part analyzed the mNGS results based on DNA and RNA of BALF samples from SCAP patients in the Part I. The characteristics of SCAP patients were explored from the perspective of host transcriptome and lung microbiome. Based on metatranscriptomic analyses, host transcriptome characteristics of SCAP patients were analyzed (differential genes, immune cell composition, and signaling pathway of differential gene enrichment). Metagenomic sequencing data were analyzed to explore lung microbiome characteristics (relative abundance and composition, comparison of diversity between groups, and characteristic microorganisms) of SCAP patients. With 28-day mortality as the primary endpoint, host transcriptomic and lung microbiome characteristics of immunocompromised and non-immunocompromised SCAP patients with different prognoses were investigated. Finally, based on the host transcriptomic analyses, patients were randomly divided into the training set and the internal validation set (7:3), to construct prognostic models in machine learning random forests algorithm for 28-day mortality in immunocompromised and non-immunocompromised SCAP patients, respectively. Another multicenter study led by our team prospectively included 105 SCAP patients in the ICU of 22 hospitals across the country between April 1, 2023 and July 31, 2023 as an external validation cohort for external validation of the models.

Part III

Patients with Klebsiella pneumoniae (KP)-CAP and KP-SCAP in the MICU of China-Japan Friendship Hospital from September 2022 to February 2023 were prospectively included. BALF samples from the two groups were collected for untargeted metabolomics to screen for differential metabolites and regulatory signaling pathways. Meanwhile, succinate verification and inflammatory factor detection were performed by ELISA. Bacterial endotoxin lipopolysaccharide (LPS) was used to stimulate epithelial cells to detect the expression levels of succinate and inflammatory factors for in vitro verification. At the same time, succinate stimulated alveolar macrophages as the experimental group and the control group were used for transcriptomic sequencing to screen key regulatory signaling pathways. Immunofluorescence, western blot and RT-PCR were used to detect the protein and mRNA expression of succinate receptor1 (SUCNR1) and ferroptosis key genes in alveolar macrophages stimulated by succinate.

Result

Part I

       1. Of the 239 patients with SCAP included in this part of the study, 66 (27.61%) died within 28 days after admission to the ICU. Compared with the survival group, patients in the 28-day mortality group were older (P<0.001), had higher APACHE II scores and SOFA scores at ICU admission (P<0.001), and had higher rates of sepsis, septic shock, acute kidney injury, use of vasopressors and invasive mechanical ventilation (IMV) on ICU admission (P<0.05). Laboratory tests showed lower oxygenation index and higher levels of lactic acid and D-dimer (P<0.05). In terms of prognosis, the proportion of patients with IMV treatment in the 28-day mortality group during ICU was higher (P<0.001).

       2. The positive rate of BALF etiological detection in SCAP patients was 79.50%, and the proportion of mixed infection was 37.66% (90/239). Compared with the survival group, the proportion of bacterial infection was higher in the 28-day mortality group (P<0.05). A higher proportion of mixed bacterial infections occurred (P<0.05). The top five pathogens of SCAP in this study were Pneumocystis jiroveci, Klebsiella pneumoniae, cytomegalovirus, Aspergillus and Staphylococcus aureus.

       3. Independent risk factors for 28-day mortality in patients with SCAP included age (OR: 1.03, 95%CI: 1.01-1.06, P=0.015), lactate level (OR: 1.31, 95%CI: 1.02-1.68, P=0.035), and IMV treatment at ICU admission (OR: 2.88, 95%CI: 1.17-7.09, P=0.021) and acute kidney injury at ICU admission (OR: 2.20, 95%CI: 1.01-4.81, P=0.049), the area under curve of the receiver operating characteristic (AUC) of Logistic regression prognosis predictive model was 0.831.

       4. Immunocompromised and non-immunocompromised SCAP patients showed significant differences in clinical features and etiological distribution. The independent risk factor for 28-day mortality in immunocompromised SCAP patients was D-dimer level, and increased mean arterial pressure was a protective factor. The AUC of prognosis predictive model was 0.910. Independent risk factors for 28-day mortality in non-immunocompromised SCAP patients were age, lactate level, and IMV treatment at ICU admission, with an AUC of 0.825 for prognosis predictive model.

Part II

       1. Host transcriptomic characteristics of SCAP patients: A total of 4193 differentially expressed genes were screened, and GSEA enrichment suggested that the expression of chemokine signaling pathway, cytokine-cytokine receptor interaction pathway, IL-17 signaling pathway and NF-κB signaling pathway in immunocompromised group was down-regulated (P<0.05). Immune cell analysis indicated that neutrophils were decreased in immunocompromised patients, but regulatory T cells were increased (P<0.05).

Lung microbiome of SCAP patients: Streptococcus, Staphylococcus and Prevotella were predominant in immunocompromised SCAP patients, while Streptococcus, Klebsiella and Acinetobacter were predominant in non- immunocompromised SCAP patients. The α diversity of immunocompromised SCAP patients was higher (P<0.05). There was a significant difference in β diversity between the two groups (P=0.001).

2. Host transcriptomic characteristics of immunocompromised SCAP patients with different prognoses: A total of 1415 differentially expressed genes were screened from the transcriptome. GSEA enrichment analysis of differentially expressed genes suggested that metabolic pathways such as early oxidative phosphorylation were activated in the 28-day mortality group, while immune response pathways such as cytokine receptor interaction, antigen processing presentation, T cell receptor signaling pathway and NK cell-mediated cytotoxicity pathway were inhibited (P<0.05).

Lung microbiome of immunocompromised SCAP patients with different prognosis: there was no significant difference in the diversity between the two groups, while there was a difference in the diversity (P=0.011). Veillonella parvula and other colonizing bacteria were significantly correlated with APACHE II score and IMV time and other prognostic indicators (P<0.05).

Prognosis predictive model: The 7-gene classifier established based on transcriptome to predict 28-day mortality of immunocompromised SCAP patients had an AUC of 0.947 (95% CI, 0.928-0.966), accuracy of 0.893, sensitivity of 86.40%, specificity of 89.40%, internal validation set AUC of 0.722, and external validation set AUC of 0.830.

3. Host transcriptomic characteristics of non-immunocompromised SCAP patients with different prognoses: A total of 2130 differentially expressed genes were detected. GSEA enrichment analysis of differentially expressed genes suggested that TNF signaling pathway, IL-17 signaling pathway and oxidative phosphorylation were highly expressed in the 28-day mortality group, while ABC transporter was down-regulated (P<0.05).

Lung microbiome of non-immunocompromised SCAP patients with different prognosis: The lung microbiome composition of the two groups was different, and the relative abundance of pathogens was higher in 28-day mortality group, with diversity showing no significant difference, diversity analysis showed significant difference (P=0.001). KP、MMP8 and CSF3 were significantly correlated with APACHE II and SOFA scores and clinical prognostic indexes (P<0.05).

Prognosis predictive model: The AUC of 7-gene classifier established to predict 28-day mortality of non-immunocompromised SCAP patients based on transcriptome was 0.839 (95% CI, 0.817-0.861), with an accuracy of 0.786, sensitivity of 91.60%, specificity of 63.40%, internal validation set AUC of 0.838 and external validation set AUC of 0.739.

Part III

1. A total of 12 KP-CAP and 13 KP-SCAP patients were included in this study. The results of untargeted metabolomics of BALF sample showed that succinate significantly increased in BALF of KP-SCAP group (P=0.02). The enrichment results of pos metabolites were mainly related to glutathione metabolism and NOD-like receptor signaling pathway, while the enrichment results of neg metabolites were related to HIF-1 signaling pathway, PI3K-Akt signaling pathway, cAMP signaling pathway and cGMP-PKG signaling pathway (P<0.05).

2. In vitro experiments, with the increase of LPS concentration, the levels of succinate in alveolar epithelial cells and in cell increased, the levels of inflammatory cytokines IL-1β and TNF-α secreted by cells increased (P<0.05).

3. Between succinate stimulated alveolar macrophages and control group, 792 differential genes were screened by transcriptomics and enriched into PI3K-Akt signaling pathway, antigen processing and presentation, MAPK signaling pathway and so on. The differential genes and ferroptosis gene sets were analyzed. Differential genes were significantly enriched in ferroptosis regulatory gene set and unclassified ferroptosis gene set (P<0.05), suggesting that f ferroptosis may be the mechanism of action of succinate in alveolar macrophages.

4. In macrophage stimulated by succinate, SUCNR1 was localized to the cell membrane and its expression was increased at the protein and mRNA levels, while the expression of the key genes of ferroptosis, ACSL4 and FTH1 were increased at the protein and mRNA levels, and SLC7A11 and GPX4 were decreased at the protein and mRNA levels (P<0.05), proving succinate promote ferroptosis of alveolar macrophage.

Conclusion

       1. There are significant differences in clinical characteristics and pathogen distribution among SCAP patients with different prognosis. The model established by clinical indicators can predict the prognosis of SCAP patients.

       2. The prognostic predictive model of SCAP patients based on host transcriptome has high sensitivity and specificity, which showed better performance than the existing clinical scores, and the predictive performance is good after external cohort validation.

      3. Succinate may be one of the biomarkers of disease progression and poor prognosis in CAP patients, and its mechanism may be the regulation of alveolar macrophage ferroptosis.