Abstract

   Long noncoding RNAs (lncRNA) are crucial yet underexplored regulators
   of human immunity. Here we develop GRADR, a method integrating gradient
   profiling with RNA-binding proteome analysis, to map the protein
   interactomes of all expressed RNAs in a single experiment to study
   mechanisms of lncRNA-mediated regulation of human primary macrophages.
   Applying GRADR alongside CRISPR-multiomics, we reveal a network of
   NFκB-dependent lncRNAs, including LINC01215, [70]AC022816.1 and ROCKI,
   which modulate distinct aspects of macrophage immunity, particularly
   through interactions with mRNA-processing factors, such as hnRNP
   proteins. We further uncover the function of ROCKI in repressing the
   messenger of the anti-inflammatory GATA2 transcription factor, thus
   promoting macrophage activation. Lastly, all data are consolidated in
   the SMyLR web interface, a searchable reference catalog for exploring
   lncRNA functions and pathway-dependencies in immune cells. Our results
   thus not only highlight the important functions of lncRNAs in immune
   regulation, but also provide a rich resource for lncRNA studies.

   Subject terms: Monocytes and macrophages, Long non-coding RNAs, Systems
   analysis, Gene regulation in immune cells
     __________________________________________________________________

   Long non-coding RNAs (lncRNA) are known immune regulators, yet a
   systemic understanding of their regulatory crosstalk is still elusive.
   Here the authors establish a platform integrating RNA-binding proteome
   analyses and protein interactomes to map the lncRNA target network in
   human primary macrophages to build a searchable atlas of lncRNA.

Introduction

   The human genome is pervasively transcribed, producing a plethora of
   RNAs, most of which remain uncharacterized^[71]1–[72]3. Among these,
   long non-coding RNAs (lncRNAs), arbitrarily defined as
   non-protein-coding RNAs longer than 200 nucleotides^[73]4, have
   garnered increasing attention due to their critical roles in various
   cellular processes, including X chromosome inactivation and immune
   responses to infectious agents^[74]5,[75]6. Despite the known
   importance of lncRNAs in the immune system, our current understanding
   of host-defense against pathogens predominantly rests on the functions
   of proteins. In the innate immune system, which constitutes the first
   line of defense against infections, various protein signaling pathways
   pivotal to pathogen recognition have been determined. For instance,
   macrophages, which are among the first cells to sense an infection, use
   pattern recognition receptors (PRRs) such as Toll-like receptor TLR4,
   which detects lipopolysaccharide (LPS), a major component of
   gram-negative bacterial cell walls. TLR4 activation triggers the
   MyD88-NFκB signaling pathway, essential for the production of
   pro-inflammatory immune mediators like IL-1β and IL-8^[76]7. In
   parallel, TLR4 triggers the TRIF-IRF3 signaling pathway, which promotes
   the production of type I interferons (IFN-I), conferring e.g.,
   protection against intracellular infections^[77]7. A multitude of
   additional PRRs contribute to pathogen detection by the immune system.

   Recent studies have highlighted significant involvement of specific
   lncRNAs, such as MaIL1 and LUCAT1, in these innate immune-signaling
   pathways. MaIL1 enhances TRIF-IRF3 signaling by binding and stabilizing
   the protein Optineurin (OPTN), leading to IFN-I production^[78]8.
   LUCAT1 acts as a negative feedback regulator of this pathway, likely
   via hnRNP and STAT1 protein interactions^[79]9–[80]12. Other lncRNAs
   such as GAPLINC, PACERR, and NEAT1 are known to contribute to
   antimicrobial defense by regulating the pro-inflammatory immune
   response^[81]13–[82]16. Additional lncRNAs have been implicated in
   human innate immunity^[83]6,[84]17. However, despite these discoveries,
   the vast majority of lncRNAs expressed in human immune cells remains
   uncharacterized^[85]8, likely due to the challenging discovery of
   lncRNA-interactions with other biomolecules, required for deciphering
   their modes of action. While methods such as RAP-MS or SHIFTR exist to
   explore individual lncRNA interactions with proteins^[86]18,[87]19,
   they typically lack high-throughput capability. Recently, we used
   Grad-seq to categorize immune-activated macrophage lncRNAs into various
   groups based on their co-sedimentation with major protein
   machineries^[88]8, yet this approach falls short of providing detailed
   interaction data for individual lncRNAs, needed for deeper mechanistic
   investigations. These limitations coupled with a lack of comprehensive
   reference data detailing the molecular pathways controlling lncRNA
   expression, impede advances in our understanding of the timing and
   functioning of lncRNAs in defense against infections and
   immune-pathologies in humans.

   In this work, we map the lncRNA landscape of primary human macrophages
   and introduce GRADR, a method that globally predicts RNA-protein
   interactions based on gradient co-sedimentation. By coupling GRADR with
   CRISPR multiomics, we reveal regulatory functions for multiple
   previously uncharacterized lncRNAs in innate immunity. Our findings are
   consolidated in a web interface, the Searchable Myeloid LncRNA Registry
   (SMyLR; [89]rna-lab.org/smylr). We believe that this resource will
   significantly advance the characterization of RNA-mechanisms in human
   immunity. Our results may also guide future efforts to target lncRNA
   networks in immune-related diseases.

Results

Human lung immunity involves numerous largely uncharacterized lncRNAs

   Long non-coding RNAs (lncRNAs) are emerging as pivotal regulators in
   eukaryotic biology, yet their functions in human immunity are only
   beginning to be defined. In this study, we present a complex multiomics
   framework for large-scale lncRNA characterization in human cells,
   revealing mechanisms of RNA-dependent immune-regulation. We further
   introduce the Searchable Myeloid lncRNA Registry, SMyLR
   ([90]rna-lab.org/smylr), an open-access resource detailing lncRNA
   pathway dependencies, subcellular localization and interactions with
   proteins during primary macrophage immune activation (Fig. [91]1A, B).
   We specifically examined key cell types of the delicate alveolar
   barrier, a primary site of pathogen-attack, to define core lncRNA
   networks relevant to human host defense.

Fig. 1. Mapping of lncRNAs implicated in human lung innate immunity.

   [92]Fig. 1
   [93]Open in a new tab

   A Illustration of key alveolar cell types, highlighting macrophage
   lncRNA networks. B Overview of the SMyLR database, detailing lncRNA
   pathway dependencies, cell-type specificity, and protein interactions.
   C Number and fold changes (≥2-fold up or down) of mRNAs in specified
   cell types following 4 h of flagellin (FLA) (AECII) or LPS (aMФ, G-MФ,
   M-MФ) stimulation; data averaged from two RNA-seq replicates. D 3D PCA
   analysis of RNA-seq samples from (C). E Heatmap of all lncRNAs
   up-regulated ≥ 2-fold in both RNA-seq replicates in aMФ or G-MФ (data
   from C, D AECII data included for comparison). Top section highlights
   lncRNAs also upregulated in AECII ≥ 2-fold. F Volcano plot displaying
   lncRNA changes (4 h LPS vs control treatment) in aMФ, highlighting
   those also regulated in G-MФ (data from C-E). Two-tailed Student’s
   t-test p-values are shown. G Illustration of PCLS preparation and
   culture. H RT-qPCR analysis of IL8 and lncRNA regulation in PCLS post
   stimulation (results relative to RPS18 and 4 h mock). Three independent
   experiments; mean values +/- SD. Asterisks denote statistical
   significance (One-way ANOVA test, p ≤ 0.05). Exact p-values: IL8: 0.017
   (4 h LPS), 0.024 (8 h LPS), 0.039 (8 h FLA); ROCKI: 0.034 (8 h FLA);
   [94]AC010980: 0.042 (8 h FLA); LINC01215: 0.045 (4 h FLA), 0.048 (8 h
   LPS), 0.017 (8 h FLA); [95]AC022816.1: 0.025 (4 h LPS), 0.011 (8 h
   LPS). I Left: Illustration of BAL procedure. Right: Linear regression
   plots with 95% confidence intervals (dashed lines) and Pearson
   correlation statistics (two-tailed test), comparing the
   RT-qPCR-determined levels of the indicated lncRNAs with IFNB1 (log2
   2^-ΔCT values) in bronchoalveolar lavage cell pellets. Samples size
   variations in individual plots (n ≤ 24) reflect cases where the
   respective lncRNA was below the detection limit.

   Our study utilized human type II alveolar epithelial cells (AECII)
   derived from resected lung tissue (Supplementary Fig. [96]1A) and
   alveolar macrophages (aMФ) from healthy volunteers via bronchoalveolar
   lavage. We also differentiated CD14+ monocytes into macrophages using
   M-CSF or GM-CSF (M-MФ or G-MФ) to simulate infection-driven monocyte
   recruitment and differentiation. While optimizing conditions for
   RNA-seq profiling of mRNAs and lncRNAs, we observed distinct immune
   responses of these cells to bacterial stimuli. Both AECII and
   macrophages responded to pathogenic Legionella pneumophila and
   non-pathogenic Escherichia coli (Supplementary Fig. [97]1B), yet
   through different pathways. Our findings show that macrophages display
   a robust response to several Toll-like receptor ligands, notably TLR4,
   TLR5, and TLR2, while AECII cells only responded significantly to TLR5
   stimulation (Supplementary Fig. [98]1B and Fig. [99]1C, D). As
   expected, RNA-seq revealed the three investigated types of macrophages
   (aMФ, M-MФ and G-MФ) to display a more similar mRNA response to
   TLR-stimulation than AECII cells (Fig. [100]1C, D, Supplementary
   Data [101]1). Yet, a limited set of key NFκB-driven immune genes were
   activated both in AECII and aMФ (e.g. CSF2, IL6, IL23A), underscoring
   the division of labor between epithelial cells and macrophages during
   human lung immunity (Supplementary Fig. [102]1C, D).

   Comparative analysis suggested that M-MФ and G-MФ recapitulate
   different aspects of the aMФ response to LPS. For example, expression
   of the cytokine IL10 was induced in both aMФ and M-MФ, while the
   coagulation factor SERPINB2 was induced in aMФ and G-MФ (Supplementary
   Fig. [103]1E-F). Despite the recorded differences, all three macrophage
   types shared the induction of several key immune response factors, such
   as CCL4 or IL6 (Supplementary Fig. [104]1E, F). Examining lncRNA
   regulation across the different macrophage types, we found that most of
   the 39 lncRNAs upregulated ≥ 2-fold in aMФ following LPS stimulation
   (in both RNA-seq replicates) were also induced in M-MФ and G-MФ
   (Supplementary Fig. [105]2A). Thus, although considerable differences
   exist, all three macrophage types exhibit a common immune-response
   signature at both the mRNA and lncRNA levels. For this study, we
   selected G-MФ as our primary in vitro model, due to its scalable
   availability and the essential role of GM-CSF in alveolar macrophage
   replenishment in vivo^[106]20. Among the lncRNAs regulated in all
   analyzed cell types (Supplementary Fig. [107]2A), 24 were consistently
   up-regulated ≥ 2-fold in both RNA-seq replicates in aMФ and G-MФ
   following LPS stimulation. This group included known immune-regulatory
   lncRNAs (e.g. MaIL1 and PACERR), as well as several previously poorly
   characterized lncRNAs, such as ROCKI, [108]AC010980, LINC00158,
   LINC01215, and [109]AC022816.1 (Fig. [110]1E, F, Supplementary
   Fig. [111]2A). Re-analysis of published RNA-seq
   datasets^[112]21–[113]23 further verified the up-regulation of these
   lncRNAs (≥2-fold on average) upon macrophage and monocyte stimulation
   with LPS (Supplementary Fig. [114]2B), with varying levels of
   significance (Supplementary Fig. [115]2C).

   To validate the engagement of these five lncRNAs under conditions that
   better mimic the native immune-contexts, we conducted 4-and 8-hour
   stimulations (both LPS and flagellin) of human precision-cut lung
   tissue slices (PCLS) which preserve native tissue architecture and
   inter-cell-type communication (Fig. [116]1G). These experiments
   confirmed the up-regulation of the selected lncRNAs following either
   TLR4 (LPS) or TLR5 (flagellin) stimulation, with statistically
   significant induction observed under at least one condition in all
   cases except for LINC00158 (Fig. [117]1H). Furthermore, we quantified
   the expression of these lncRNAs in bronchoalveolar lavage (BAL)-derived
   cells from 24 patients undergoing BAL for pulmonary disease assessment.
   Our recent work demonstrated that IFNB1 serves as a sensitive marker of
   pulmonary immune activation in BAL studies^[118]8. Notably, all 5
   lncRNAs - but not the control transcripts U6 snRNA or RPS18 mRNA -
   showed significant linear correlation with IFNB1 expression in the
   patient cohort, further supporting their involvement in pulmonary
   immunity (Fig. [119]1I, Supplementary Fig. [120]2D and E).

   Together, these results underscore the engagement of various macrophage
   lncRNAs in human anti-pathogen responses. To promote further research
   into their diverse roles we systematically dissected their
   pathway-dependencies, conservation, functions and protein-interactions
   (Fig. [121]2A).

Fig. 2. Pathway dependency of macrophage lncRNAs.

   [122]Fig. 2
   [123]Open in a new tab

   A Illustration of attributes analyzed to characterize immune-responsive
   lncRNAs in macrophages. B Left: Heatmap (row Z-scores) showing lncRNA
   expression changes in G-MФ following 2 h and 4 h stimulation with
   Pam3csk4 (PAM), LPS, or interferon-α (IFN), based on averaged RPKMs
   from two independent RNA-seq datasets. Right: Top: Illustration of
   pathways relevant for LPS-dependent lncRNA regulation. Bottom: RNA-seq
   analysis of the dependence of lncRNAs shown in the heatmap (left) on
   LPS-induced pathways. Bay = BAY 11-7082 (NFκB inhibitor); Rux =
   Ruxolitinib (STAT-inhibitor); BX = BX795 (TBK1-IRF3 inhibitor).
   Fold-changes relative to base-mean, averaged from two independent
   RNA-seq experiments. C RT-qPCR analysis of lncRNA responses to LPS
   (4 h) plus either DMSO or inhibitors used in B) at three increasing
   concentrations; results from three independent experiments; mean values
   +/- SD. Exact p-values (in the same order as the asterisks): ROCKI:
   0.001, 0.013; [124]AC010980: <0.001; LINC00158: <0.001, <0.001;
   LINC01215: 0.001; [125]AC022816.1: <0.001, <0.001. D Time-series
   RT-qPCR analysis of IL8, IFNB1, and lncRNA expression in G-MФ treated
   with LPS; data from three independent experiments, showing mean values
   +/- SD. Exact p-values: ROCKI: <0.001; LINC00158: <0.001;
   [126]AC010980: <0.001; LINC01215: <0.001; [127]AC022816.1: 0.004. E
   Heatmap of lncRNA changes in G-MФ exposed to various live or
   UV-inactivated bacterial pathogens for 4 h; three independent
   experiments; RT-qPCR. C, D: One-way ANOVA tests were conducted. C: all
   inhibitor treatments were compared to the LPS-treated control. D (left
   panel): P-values were calculated for the linear trends of column
   averages from left to right. Statistical significance (p ≤ 0.05) is
   indicated by asterisks.

LncRNAs are embedded in different pathways and phases of the immune response

   To elucidate the pathway dependencies of the identified
   immune-associated lncRNAs, we exposed G-MФ to various immune agonists
   and pathway inhibitors, followed by RNA-seq. Furthermore, we tracked
   lncRNA temporal regulation by performing a time-course LPS stimulation
   experiment coupled to RNA-seq at intervals from 0.5 to 12 hours. Most
   lncRNAs found to be up-regulated by LPS in aMФ and G-MФ (Fig. [128]1E,
   Supplementary Fig. [129]2A), also responded to TLR2 agonist Pam3CSK4
   but not to type I IFN (IFNα) (Fig. [130]2B, Supplementary
   Fig. [131]3A–D, Supplementary Data [132]2). Inhibition of
   NFκB-dependent gene expression with BAY-11-7082 globally reduced
   up-regulation of these lncRNAs. In contrast, inhibition of STAT and IRF
   signaling with Ruxolitinib and BX795 respectively, did not affect
   global lncRNA induction in macrophages in response to LPS
   (Fig. [133]2B, Supplementary Fig. [134]3E, Supplementary Data [135]3).
   These findings suggest that most immune-inducible lncRNAs are regulated
   in an NFκB-dependent manner. RT-qPCR validations for the five lncRNAs
   of primary interest (see above) confirmed a consistent NFκB dependency,
   with only LINC01215 showing additional sensitivity to the TBK1-IRF3
   pathway in BX795 inhibitor treatments (Fig. [136]2C and Supplementary
   Fig. [137]3E). Temporally, these lncRNAs followed regulation kinetics
   similar to the classic immune response gene IL8 (CXCL8) in G-MФ
   (Fig. [138]2D, Supplementary Fig. [139]3C and F). However, other
   lncRNAs, such as MaIL1 and PACERR, exhibited earlier expression peaks
   (e.g. at 1 hour post-LPS challenge) (Supplementary Fig. [140]3F),
   suggesting that lncRNAs play distinct roles at different stages of the
   macrophage immune response. Notably, similar lncRNA regulation kinetics
   were observed in M-MФ, with the exception of LINC01215, which was not
   regulated in this macrophage type (Supplementary Fig. [141]3G). RT-qPCR
   experiments further confirmed up-regulation of the five lncRNAs in
   focus in G-MФ in response to multiple live or UV-inactivated bacterial
   pathogens, including S. Typhimurium, K. pneumoniae, M. catarrhalis, and
   S. pneumoniae (Fig. [142]2E), underscoring their involvement in host
   defense. Comparative genomic analysis revealed that while the five
   lncRNAs in focus, as well as macrophage LPS-inducible lncRNAs in
   general (lncRNAs from Fig. [143]1E), are conserved among primates,
   their conservation is limited in mice. However, better conservation was
   observed in species with longer generation times (Fig. [144]3A).

Fig. 3. Conservation and functional impact of macrophage lncRNAs.

   [145]Fig. 3
   [146]Open in a new tab

   A Left: Evolutionary tree of selected mammalian species. Right: Percent
   base conservation of human LPS-inducible macrophage lncRNA sequences
   across the genomes of indicated species, sorted by evolutionary
   distance; top panel includes all 24 lncRNAs from Fig. [147]1E and data
   are shown as box plots (median and 75th-25th percentile interquartile
   range, whiskers indicate minimum and maximum data values), with species
   generation time overlayed; bottom panel focuses on specific lncRNAs. B
   Illustration of genomic locations of specified lncRNAs relative to
   nearest neighboring genes (distance in kilobases [kb] provided). Red
   triangles indicate transcriptional start site positions targeted by
   CRISPRi for lncRNA silencing. C Volcano plots from CRISPRi-based lncRNA
   loss-of-function experiments in THP1 cells stimulated with LPS for 8 h.
   Fold-changes (fc) compare lncRNA-knockdown cells to empty vector
   control cells. Results from three independent experiments. Two-tailed
   Student’s t-test p-values are shown. D Cytoscape network of lncRNAs
   (blue) and mRNAs from panel C, regulated upon lncRNA silencing (≥2-fold
   up or down, p ≤ 0.05, two-tailed Student’s t-test). E Pie charts
   showing proportions of all expressed mRNAs or LPS-responsive mRNAs
   (≥2-fold up or down, p ≤ 0.05, two-tailed Student’s t-test) affected by
   silencing of one or more lncRNAs. C–E: 8 h LPS-stimulated THP1 cells
   and three independent replicates.

   Taken together, these findings suggest that the lncRNA responses
   depicted here reflect a common core program with minor variations
   across macrophage types. The limited evolutionary conservation
   highlights a potential challenge in using murine models to study the in
   vivo functions of many human macrophage lncRNAs, including those
   central to this study.

CRISPR-multiomics uncovers a broad lncRNA network regulating innate immunity

   To estimate the regulatory potential and redundancy among various
   immune-responsive lncRNAs, we conducted loss-of-function studies
   focusing on the lncRNAs selected for validation above (ROCKI,
   [148]AC010980, LINC00158, LINC01215, and [149]AC022816.1). Using CRISPR
   interference (CRISPRi), we silenced each lncRNA using ≥ 2 independent
   guideRNA designs in THP1 macrophages, followed by LPS stimulation,
   RT-qPCR and RNA-seq analysis in triplicates. All five lncRNAs were
   located several kilobases away from neighboring genes, suggesting that
   directing the CRISPRi machinery to the transcriptional start site
   regions of these lncRNAs is unlikely to directly affect the expression
   of adjacent genes (Fig. [150]3B). RT-qPCR and RNA-seq analysis
   confirmed successful lncRNA silencing (Supplementary Fig. [151]4A,
   Fig. [152]3C, Supplementary Data [153]4) and revealed a broad mRNA
   network significantly influenced by the five lncRNAs in
   immune-challenged cells (Fig. [154]3C, D, Supplementary Fig. [155]4B).
   Notably, substantial overlap in mRNA regulation was observed between
   LINC00158 and [156]AC010980, suggesting partial redundancy between
   these two lncRNAs, whereas [157]AC022816.1, LINC01215, and ROCKI
   influenced more distinct mRNA sets (Fig. [158]3D). Overall, ~8% of all
   expressed mRNAs (RPKM ≥ 0.5) and ~16% of all LPS-responsive mRNAs
   (fold-change ≥ 2) were under regulatory influence by the 5 lncRNAs
   (≥2-fold change and p ≤ 0.05 upon loss of ≥ 1 lncRNA) (Fig. [159]3E).

   To correlate mRNA regulation with protein abundance, we supplemented
   our lncRNA loss-of-function RNA-seq data with quantitative
   whole-proteome analysis. Significant correlations between regulated
   mRNAs and their corresponding proteins were noted for LINC01215 and
   [160]AC022816.1 (p < 10^-4), with ROCKI approaching significance
   (p = 0.06) (Fig. [161]4A, B, Supplementary Data [162]5). LINC00158 and
   [163]AC010980 showed a less pronounced impact at the protein level
   (Fig. [164]4A, B). Notably, when restricting the analysis to factors
   regulated both at the RNA and the protein level upon lncRNA-loss, the
   five lncRNAs impacted the expression of distinct sets of genes, with
   LINC01215 and [165]AC022816.1 mostly impinging on immune-related
   pathways and LINC00158 and [166]AC010980 on RNA-processing and
   metabolism-related pathways (Fig. [167]4B, C, Supplementary
   Fig. [168]5A–C). When limiting the analysis to LPS-responsive genes,
   ROCKI, LINC01215, and [169]AC022816.1 were found to significantly
   influence the LPS response at the mRNA level, with LINC01215 also
   exerting significant effects at the protein level (Fig. [170]4D).
   Specifically, genes that were down-regulated upon ROCKI silencing were
   up-regulated in response to LPS, while genes up-regulated upon ROCKI
   silencing exhibited the opposite trend. In contrast, genes regulated by
   LINC01215 or [171]AC022816.1 showed the opposite behavior
   (Fig. [172]4D), suggesting that ROCKI acts to enhance, whereas
   LINC01215 and [173]AC022816.1 function to suppress the LPS-response.

Fig. 4. Comparative analysis of lncRNA loss-of-function effects at RNA and
protein levels.

   [174]Fig. 4
   [175]Open in a new tab

   A Pearson correlation analysis (p-values shown, two-tailed test)
   comparing RNA-seq-determined significant mRNA fold-changes (from
   Fig. [176]3) and corresponding protein fold-changes. B Clustered
   heatmap showing aggregated factors regulated at both RNA and protein
   levels (≥2-fold up or down, p ≤ 0.05) upon lncRNA silencing. Clusters
   (A-H) and their primary Reactome pathways (Supplementary Fig. [177]5A)
   are indicated. C ENRICHR-based UMAP visualization of the “Biological
   Process” GO-term network, highlighting GO-terms associated with gene
   products that are significantly regulated at the RNA and protein levels
   (≥2-fold upregulation [red lines] or downregulation [blue lines],
   p ≤ 0.05, two-tailed Student’s t-test) upon silencing of the depicted
   lncRNAs. D Top left: correlation of mRNA (≥2-fold up or down, p ≤ 0.05,
   two-tailed Student’s t-test) and corresponding protein fold-changes
   (RNA-seq and proteomics, ≥ 3 independent experiments) in response to
   LPS (8 h, THP1 cells). Pearson correlation p-value (two-tailed test) is
   shown. Violin plots: fold-changes at the RNA- and protein-level in
   response to LPS of factors regulated upon lncRNA silencing (≥2-fold up
   [red] or down [blue], p ≤ 0.05 in data from A, B). Two-tailed Student’s
   t-test p-values are shown. E Schematic summary of hypothesized
   functions for the five studied lncRNAs in macrophages. All panels: 8 h
   LPS-stimulated THP1 cells and ≥ three independent replicates.

   Collectively, our comprehensive, comparative lncRNA loss-of-function
   studies highlight LINC01215 and [178]AC022816.1 as negative regulators
   (suppressors) of macrophage immunity, while suggesting that ROCKI
   serves as a positive regulator (enhancer) (Fig. [179]4E). However,
   unlike LINC01215, the regulatory effects observed at the RNA level for
   ROCKI and [180]AC022816.1 appear to translate to the protein level only
   to a limited extend. During the preparation of our manuscript, ROCKI
   was independently identified as a positive regulator of macrophage
   immunity, corroborating our findings^[181]24. To explore the potential
   modes of action of macrophage lncRNAs, we determined their cooperation
   with protein partners.

GRADR charts the global lncRNA-protein interaction landscape in macrophages

   A primary challenge in lncRNA research is identifying protein
   interaction partners, a labor-intensive process reliant on individual
   affinity purification of each lncRNA. To accelerate the
   characterization of lncRNAs compiled in our catalog, we developed a
   streamlined approach termed GRADR (Gradient-based RNA interactor
   profiling), predicting the interactomes of all expressed lncRNAs in a
   single experiment. GRADR integrates several techniques (Fig. [182]5A):
   initially, Grad-seq separates cellular RNA-protein complexes by size on
   a linear glycerol gradient, followed by fractionation. Subsequent
   RNA-seq and proteomics analyses reveals the global co-sedimentation
   landscape, providing a comprehensive view of possible RNA and protein
   interactions. Additionally, OOPS-MS (Orthogonal Organic Phase
   Separation-Mass Spectrometry) refines this data by pinpointing those
   co-sedimenting proteins that have RNA binding capacity and thus are
   likely interactors. Parallel cytoplasm/nucleus fractionations coupled
   to RNA-seq and proteomics focus the analysis on proteins and RNAs
   present in the same subcellular compartment. This approach narrows the
   protein interactor space for each expressed RNA to a few dozen
   candidates, as demonstrated below.

Fig. 5. Development of GRADR for predicting lncRNA-protein interactions in
macrophages.

   [183]Fig. 5
   [184]Open in a new tab

   A Illustration of the GRADR methodology. B Z-score heatmap and violin
   plot (overlayed averages as bars) displaying differential enrichment of
   RNA-binding and non-RNA-binding protein groups in OOPS-MS eluates
   (violin plot: averages from three independent experiments). C
   Illustration of the distribution of representative protein classes
   within cytoplasmic and nuclear fractions; data averaged from three
   independent proteomics experiments. D Percentage of ACTB mRNA and RMRP
   lncRNA across cytoplasmic (C) and nuclear (N) fractions (RNA-seq,
   averages and individual data points from two independent experiments).
   E Panels 1-2: GRADR-derived protein interactome for ACTB mRNA and RMRP
   lncRNA, plotting co-sedimentation Pearson r values against -log10
   OOPS-MS RNA-binding p-values (two-tailed Student’s t-test). Proteins
   with r ≥ 0.5 and p ≤ 0.05 highlighted in blue. Panel 3: Reactome
   pathway analysis for proteins associated with RMRP (blue in panel 2). F
   Same as D), but for the indicated lncRNAs. G Grad-seq Pearson
   correlation matrix for proteins identified by GRADR (r ≥ 0.5 and
   p ≤ 0.05, two-tailed Student’s t-test) as potential interactors of the
   five lncRNAs shown in F (Pearson r color coded from red to white
   [1–0]). % cytoplasmic localization from C indicated to the left,
   protein-lncRNA interactions are noted below. Relevant clusters were
   subjected to Bioplanet pathway analysis (shown to the right). H UMAP
   plot displaying the glycerol gradient distribution (Grad-seq) of the
   five lncRNAs (highlighted as circles with black outlines) and proteins
   belonging to several cellular machineries. Relevant machineries
   highlighted for context. I Grad-seq Pearson r values determined for the
   indicated splicing-associated proteins (from G) and the five lncRNAs.
   All panels: 4 h LPS-stimulated G-MФ.

   For the initial step of GRADR, we revisited our previously published
   Grad-seq co-sedimentome data from LPS-treated G-MФ^[185]8. We
   complemented this dataset with OOPS-MS analysis using RNase-based RBP
   elution, which delineated the RNA-binding proteome in G-MФ. The results
   successfully discriminated between known RNA-binding and
   non-RNA-binding protein classes as positive and negative controls,
   respectively (Fig. [186]5B, Supplementary Fig. [187]6A). Further,
   subcellular fractionation combined with proteomics accurately mapped
   proteins to their respective cellular compartments, e.g., identifying
   chromatin factors in the nucleus and Golgi proteins in the cytoplasm
   (Fig. [188]5C, Supplementary Fig. [189]6B). To validate the efficacy of
   GRADR, we examined its ability to pinpoint proteins known to
   co-localize with cytoplasmic ACTB mRNA and the mitochondrial RMRP
   lncRNA (Fig. [190]5D, Supplementary Fig. [191]6C, Supplementary
   Data [192]6–[193]8). GRADR accurately predicted interactions between
   ACTB mRNA and ribosomal proteins, and between RMRP and mitochondrial
   proteins involved in pathways such as glycolysis and the TCA cycle
   (Fig. [194]5E). To further benchmark GRADR against known interactomes
   of immune-regulatory lncRNAs in macrophages, we compared its
   predictions for the interferon-regulatory lncRNA LUCAT1 (Supplementary
   Fig. [195]6C and D) with ChIRP-MS affinity purifications conducted both
   in this study and previously published by others^[196]10
   (Fig. [197]6E–G, Supplementary Data [198]9). Our own ChIRP-MS data
   independently confirmed the previously reported association of LUCAT1
   with hnRNP proteins and generally with splicing-related factors
   (Supplementary Fig. [199]6E and F)^[200]10. Consistently, GRADR also
   predicted interactions between LUCAT1, hnRNPs and splicing factors
   (Supplementary Fig. [201]6D and F), several of which had previously
   been identified as LUCAT1 interactors by Vierbuchen et al.^[202]10.
   (Supplementary Fig. [203]6G).

Fig. 6. Functional dissection of lncRNA ROCKI in macrophages.

   [204]Fig. 6
   [205]Open in a new tab

   A GRADR-derived interactome for ROCKI, analogous to Fig. [206]5E
   (two-tailed Student’s test p-values and Pearson r). B Illustration of
   the ChIRP methodology. C Left: Volcano plot highlighting ChIRP-MS
   identified ROCKI interactors in THP1 cells (fold-changes comparing
   ROCKI to control ChIRP captures; two-tailed Student’s t-test p-values).
   Right: Venn diagram showing overlap between GRADR and ChIRP-MS
   interactome predictions for ROCKI. D Top: Representative Western blot
   of hnRNP L in control (-) and hnRNP L ( + ) CLIP eluates (size marker
   positions in kDa indicated). Bottom: RT-qPCR analysis showing
   enrichment of ROCKI in hnRNP L CLIP eluates from THP1 and G-MФ cells
   (two-tailed Student’s t-test; three independent experiments; mean +/-
   SD). E Left: Row Z-score heatmap of protein abundance in two THP1 hnRNP
   L CLIP eluates (C = control, L = hnRNP L CLIP). Middle: iBAQ heatmap
   showing averaged total protein abundance in the hnRNP L CLIP. Right:
   Fold-enrichment plot for the top 7 eluted proteins. F Left: IGV plots
   showing control and ROCKI ChIRP-seq signals on both DNA strands at
   specified genomic loci (THP1 cells). Right: qPCR quantification of
   ROCKI and MARCKS locus enrichment in ROCKI- compared to control-ChIRP
   captures. MARCKS primer locations indicated by triangles below IGV plot
   (colors correspond to replicates shown in qPCR quantifications; three
   independent experiments; mean +/- SD). Significant differences
   (p ≤ 0.05, One-way ANOVA test) indicated. G Volcano plot of mRNA
   regulation upon ROCKI silencing (data from Fig. [207]3C). Two-tailed
   Student’s t-test p-values are shown. H RT-qPCR analysis of GATA2 mRNA
   levels in THP1 cells following the indicated ROCKI and hnRNP knockdowns
   (KD) (three independent experiments; mean +/- SD). Primers targeting
   the specified GATA2 exon-exon junctions were used. All significant
   changes (p ≤ 0.05, One-way ANOVA test) are denoted by blue asterisks.
   Exact p-values (in the same order as the asterisks): Exon 3-4: 0.004,
   0.002. Exon 5-6: 0.032. I Left: GATA2 overexpression strategy. Right:
   Comparison of protein regulation upon ROCKI silencing (data from
   Fig. [208]4, two-tailed Student’s t-test p-value ≤ 0.05)) versus GATA2
   overexpression (THP1 cells). Pearson correlation analysis was
   performed, and r and p-value (two-tailed test) are shown. J Left:
   averaged fold-changes of LPS-inducible (≥2-fold induction) proteins
   upon ROCKI and GATA2 silencing (ranked by changes observed in the ROCKI
   silencing experiment). Right: Regulation of the respective proteins in
   response to LPS (data from Fig. [209]4D). K Summarizing model of ROCKI
   function in conjunction with GATA2 in macrophages. L Illustration of
   the SMyLR results page. All panels: G-MФ treated with LPS for 4 h, THP1
   cells for 8 h.

   Having established the capability of GRADR to correctly predict known
   RNA-protein interactions, we applied it to identify proteins and
   protein machineries associated with the five immune-related lncRNAs in
   the focus of this study. Initial mapping of RNA subcellular
   localization via RNA-seq showed ROCKI, [210]AC022816.1, LINC00158, and
   LINC01215 to be primarily nuclear, while [211]AC010980 was found mainly
   in the cytoplasm (Fig. [212]5F). GRADR identified numerous potential
   interactors for these lncRNAs, notably chaperones and splicing factors
   (Fig. [213]5G–I, Supplementary Fig. [214]6H, Fig. [215]6A). Grad-seq
   further revealed a substantial cluster of co-sedimenting proteins among
   the GRADR-predicted interactors of these five lncRNAs, predominantly
   associated with splicing (Fig. [216]5G). UMAP analysis of the Grad-seq
   sedimentation patterns of the five lncRNAs and proteins belonging to
   various intracellular machineries supported the association of these
   lncRNAs with splicing factors (Fig. [217]5H). Within the Bioplanet
   splicing pathway, GRADR identified nine proteins, primarily from the
   hnRNP family, as potential core-interactors for the five lncRNAs
   (Fig. [218]5I). These findings are consistent with the established
   function of hnRNPs in lncRNA-mediated immunoregulation^[219]10,[220]25.

LncRNA ROCKI cooperates with hnRNPs to remove a GATA2 dependent brake on
immunity

   In our CRISPR loss-of-function analyses, the lncRNA ROCKI emerged as a
   potential positive regulator of the macrophage immune response
   (Fig. [221]4). To further substantiate this role and its
   GRADR-predicted interaction with hnRNP proteins, we investigated the
   influence of ROCKI on macrophages and explored its mechanisms of
   action. Although ROCKI did not significantly alter the overall LPS
   response at the protein level (Fig. [222]4), detailed analysis revealed
   the down-regulation of key pro-inflammatory mediators upon ROCKI
   silencing, including IL1b, CCL20, IL8, and PTGS2 (Supplementary
   Fig. [223]7A). This regulation was confirmed through independent IL1β
   ELISA assays (Supplementary Fig. [224]7B), further suggesting that
   ROCKI acts as an NFκB-inducible feed-forward regulator of the
   inflammatory response (Supplementary Fig. [225]7C).

   To elucidate its mechanism, we focused on the GRADR-predicted
   interaction of ROCKI with hnRNP proteins (Fig. [226]6A). To test the
   validity of the GRADR predictions, we employed ChIRP-MS RNA-antisense
   purification to determine ROCKI-bound proteins. ChIRP-MS validated the
   primary association of ROCKI with several hnRNPs, confirming our GRADR
   predictions (Fig. [227]6A–C, Supplementary Fig. [228]8A and B,
   Supplementary Data [229]9). Subsequent hnRNP L UV-crosslinking and
   co-immunoprecipitation (UV-CLIP) experiments confirmed the interaction
   of ROCKI with hnRNP L (Fig. [230]6D, Supplementary Fig. [231]8C, D) - a
   protein previously implicated in immune-regulation^[232]25. Mass
   spectrometry analysis of hnRNP L interacting proteins further suggested
   that ROCKI may be part of a larger hnRNP complex, additionally
   comprising e.g. hnRNP A2B1 and hnRNP D, predicted by GRADR and ChIRP-MS
   to bind to ROCKI, respectively (Fig. [233]6D, E, Supplementary
   Fig. [234]8E, Supplementary Data [235]10). Binding motif analysis and
   established machine learning models (GraphProt^[236]26 and
   RNAprot^[237]27) trained on publicly available hnRNP CLIP data
   furthermore revealed various potential binding sites for these hnRNPs
   in the ROCKI RNA sequence (Supplementary Fig. [238]9A, B). Beyond
   hnRNPs, ENCODE eCLIP data analysis substantiated the interaction of
   ROCKI with various proteins involved in splicing (Supplementary
   Fig. [239]9C), further supporting the suspected role for ROCKI in the
   nuclear mRNA processing machinery. While a report published during
   preparation of our manuscript suggested that ROCKI acts at the
   chromatin level, binding upstream of the MARCKS gene^[240]24, our
   ChIRP-seq and -qPCR analysis did not detect ROCKI-binding at this site
   (Fig. [241]6F), supporting a role independent of direct chromatin
   interaction. Of note, it did, however, confirm the expected
   crosslinking of ROCKI to its own site of transcription, as a positive
   control for the ChIRP-seq assay (Fig. [242]6F). Closer examination of
   our ROCKI CRISPRi RNA-seq dataset revealed that the messenger of the
   GATA2 transcription factor was the most highly and significantly
   upregulated mRNA upon ROCKI silencing (Fig. [243]6G). Up-regulation of
   GATA2 mRNA upon ROCKI silencing could additionally be confirmed in
   siRNA-based experiments with G-MФ and M-MФ (Supplementary
   Fig. [244]9D). RT-qPCR analysis across all expressed exon-exon
   junctions of GATA2 mRNA indicated that ROCKI specifically suppresses
   the maturation of the GATA2 3’ exons 5 and 6 (Fig. [245]6H,
   Supplementary Fig. [246]9E–G). GapmeR-mediated knockdown of several
   GRADR- and ChIRP-MS-predicted ROCKI-interacting hnRNPs identified
   hnRNPs A3 and A2B1 as being required for establishment of the GATA2
   mRNA exon 3-4 boundary (Fig. [247]6H, Supplementary Fig. [248]9E–G).
   Thus, ROCKI and hnRNP proteins appear to impact distinct GATA2 mRNA
   exons, suggesting a model in which ROCKI piggybacks on an hnRNP complex
   required for specific GATA2 mRNA processing steps, eventually enabling
   ROCKI to suppress the inclusion of 3’ exons and ultimately reduce GATA2
   mRNA abundance (Supplementary Fig. [249]9H). Investigating a direct
   connection between GATA2-inhibition and the pro-inflammatory activity
   of ROCKI, we force-expressed GATA2 via a lentiviral vector, thus
   overriding suppression by ROCKI. Intriguingly, proteomics analysis
   revealed that GATA2 transcription factor overexpression phenocopies
   ROCKI deficiency, entailing the suppression of the major
   pro-inflammatory mediators IL1b, CCL20, IL8, and PTGS2, alongside other
   ROCKI-regulated proteins (Fig. [250]6I, Supplementary Fig. [251]10A,
   Supplementary Data [252]11). When focusing on LPS-response genes, ROCKI
   silencing and GATA2 over-expression were found to increase the
   expression of several proteins associated with the interferon-response,
   while down-regulating pro-inflammatory factors (Fig. [253]6J,
   Supplementary Fig. [254]10B). Collectively, these findings reveal ROCKI
   as an hnRNP-associated, NFκB-inducible lncRNA that reinforces the
   inflammatory response by blunting a GATA2 transcription factor
   dependent anti-inflammatory program in macrophages (Fig. [255]6K).

   Overall, our study establishes a mechanistic framework for
   understanding the roles of lncRNAs in human immunity, systematically
   charting their regulation, pathway dependencies and protein-interaction
   networks. By revealing RNA-driven control mechanisms in professional
   immune cells, our results position lncRNAs, such as ROCKI, LINC01215
   and [256]AC022816.1 as central players in coordinating innate immune
   responses. To accelerate progress in this emerging field, we made our
   comprehensive datasets available through the SMyLR web interface
   ([257]rna-lab.org/smylr) (Fig. [258]6L and Supplementary Fig. [259]11),
   providing an open-access resource for functional lncRNA exploration. We
   believe, our findings will facilitate future work aimed at deepening
   our understanding of RNA-mediated regulation in human health and
   disease.

Discussion

   Our findings suggest that lncRNAs exert more extensive control over
   cellular innate immune responses than previously thought. The five
   lncRNAs ROCKI, [260]AC022816.1, LINC00158, LINC01215, and
   [261]AC010980, investigated in this study, collectively influence the
   expression of at least 16% of the mRNAs regulated in the macrophage LPS
   response (Fig. [262]3E). These are complemented by additional
   immune-regulatory lncRNAs such as LUCAT1 and MaIL1^[263]8–[264]12.
   Notably, the five lncRNAs in the focus of the present study were
   selected arbitrarily, implying that many of the remaining
   LPS-responsive lncRNAs may also have regulatory functions in
   macrophages, potentially influencing an even greater proportion of the
   immune-responsive transcriptome. Our LPS time-course RNA-seq analysis
   further supports that not all LPS-responsive lncRNAs operate at the
   same time but rather at various stages, with some, like MaIL1, acting
   early in TLR signaling, and others modulating the sustained immune
   response (Supplementary Fig. [265]3F). These observations underscore
   the complexity of lncRNA-mediated regulatory networks in macrophages
   and their potential to orchestrate distinct layers of innate immune
   control across different temporal and functional axes.

   Our findings particularly underscore the role of ROCKI as a
   pro-inflammatory lncRNA in humans. Although we cannot rule out that
   ROCKI regulates chromatin, as previously speculated^[266]24, we found
   no evidence supporting its postulated binding to the MARCKS gene
   region. Instead, ROCKI appears to inhibit GATA2 mRNA expression
   (Fig. [267]6G, H). Expression of the GATA2 transcription factor is
   typically restricted to hematopoietic progenitor cells^[268]28 and
   overexpression of GATA2 disrupts normal macrophage function, including
   the ability to express inflammation mediators (Fig. [269]6I–K) and to
   perform efferocytosis^[270]29. This suggests, that ROCKI helps
   maintaining the functional integrity of terminally differentiated
   macrophages by preventing the re-expression of progenitor state genes.
   Mechanistically, ROCKI appears to interact with hnRNP proteins and
   modulate GATA2 mRNA processing. The hnRNPs A3 and A2B1, which appear to
   be part of a ROCKI-binding complex (Fig. [271]6A–E), promote the
   establishment of the GATA2 exon 3-4 junction, while ROCKI itself
   appears to inhibit primarily the formation of the exon 5-6 junction
   (Fig. [272]6H). We hypothesize, that ROCKI is brought into proximity to
   GATA2 mRNA via the hnRNP complex, where it may either shift the
   activity of this complex or recruit additional factors suppressing the
   maturation of GATA2 mRNA 3’ sequence elements. The precise mode of
   action of the ROCKI-hnRNP complex, potentially in cooperation with
   other RNA processing factors, remains to be elucidated. Furthermore,
   the specific GATA2 protein binding sites in the genome and epigenetic
   alterations that contribute to reduced pro-inflammatory activity of
   macrophages upon ROCKI silencing warrant further investigation.

   Our study suggests that beyond ROCKI, numerous other lncRNAs
   participate in macrophage immunity. The introduction of GRADR marks a
   major advancement in identifying their interaction partners, a crucial
   step for lncRNA functional analysis. By integrating co-sedimentation,
   RNA-binding potential, and subcellular localization, GRADR enables the
   identification of potential interactors for a large number of lncRNAs
   in a single experiment. However, it does not confirm direct
   interactions, which is a key limitation compared to targeted affinity
   purification methods such as eCLIP or RAP-MS. The strengths and
   limitations of GRADR are reflected in our findings on the protein
   interactomes of LUCAT1 and ROCKI. GRADR identified only a fraction of
   the interactors detected by ChIRP-MS (~14-15% for LUCAT1 and ~19% for
   ROCKI) (Supplementary Fig. [273]6G, Fig. [274]6C). However, for LUCAT1,
   among these were interactors previously reported by others^[275]10,
   reinforcing the ability of GRADR to highlight relevant candidates.
   Given the interexperimental variability in RNA affinity purifications,
   as demonstrated for LUCAT1 (Supplementary Fig. [276]6G), we believe
   that this capacity of GRADR to refine likely interactors and pathways
   for further investigation is particularly valuable. As a stand-alone
   tool, GRADR is best suited for identifying broader cellular machineries
   (i.e. through pathway analysis) that lncRNAs associate with rather than
   selecting individual interactors. We recommend that interactor
   selection from GRADR predictions be complemented either by RNA affinity
   purification studies as discussed above, or alternative approaches,
   such as machine-learning models and eCLIP-based predictions, as
   demonstrated for ROCKI in this study (Supplementary Fig. [277]9B, C).

   Interestingly, most of the immune-inducible lncRNAs catalogued by us
   seem to be poorly conserved beyond primates (Fig. [278]3A). This
   limited conservation might explain why regulators such as ROCKI or
   LINC01215 have been overlooked in traditional immunology studies
   focusing on murine models. The greater sequence flexibility of lncRNAs,
   which allows them to evolve new functions more rapidly compared to
   coding genes^[279]30 might facilitate the development of new
   species-specific regulators, that tailor mammalian immune systems to
   diverse ecological niches. Thus, while rodent models are invaluable for
   elucidating fundamental principles of immunity, a thorough
   understanding of lncRNA functions may be crucial for fully grasping
   immune-associated mechanisms contributing to antimicrobial defense and
   diseases in humans.

   As lncRNAs gain recognition as key players in immune-related diseases,
   including severe COVID-19, COPD, and IBD^[280]9–[281]12,[282]24, future
   iterations of the SMyLR web interface could be expanded to incorporate
   genomic and expression data from diverse patient cohorts to assess
   clinical relevance. Although this study lays substantial groundwork,
   more research is required to fully elucidate the implications of
   lncRNAs in human immunity and disease. We anticipate that the resources
   and insights provided here will catalyze further work, advancing our
   understanding of the roles of lncRNAs in immune-regulation and opening
   new opportunities for disease biomarker discovery and therapeutic
   intervention.

Methods

Cell and tissue isolation and cultivation

   All uses of human material were approved by the ethics committees of
   the University of Marburg Medical Faculty (UMR-FB20) or the Justus
   Liebig University of Giessen Medical Faculty (JLU-FB11). All recruited
   volunteers provided written informed consent. All work with primary
   human samples is described in detail below.

   Alveolar type II epithelial cells (AECII) and precision cut lung slices
   (PCLS): For the isolation of AECII, healthy human lung tissue (e.g.
   healthy edge tissue from tumour resections; UMR-FB20 ethics approval
   number: Studie 224/12 and Studie 122/19) was sliced and washed with
   balanced salt solution buffer (BSSB) pH 7.4 (137 mM NaCl, 5 mM KCl,
   0.7 mM Na[2]HPO[4], 10 mM HEPES, 5.5 mM Glucose, 1.8 mM CaCl[2], 0.7 mM
   MgSO[4]). Tissue was digested in 40 ml BSSB containing 2.86 mg trypsin
   (#T4799, Sigma) and 300 µl of 5 U/ml elastase (#[283]LS002292,
   Worthington Biochemical) at 37  °C for 45 min while shaking. Digestion
   was stopped by transferring the tissue sample into 40 ml inhibition
   solution (30 ml DMEM/F-12 (#11320033, Gibco), 10 ml FCS (#S0615,
   Biochrom) and 1 ml DNaseI (10,000 U/ml, #4716728001, Sigma)) and cells
   were detached from the tissue by pipetting and passing the cell
   suspension successively through a 100 µm and 40 µm cell strainer
   (#431752, #431750, Corning), followed by centrifugation at 400 x g for
   10 min. Cells were resuspended in adhesion medium (22.5 ml SAGM
   (#11645490, Lonza), 22.5 ml DMEM/F-12, 10% FCS, 0.5 ml DNAseI 10,000
   U/ml), seeded and incubated at 37  °C for 2 h. The supernatant was
   collected and the adhesion step was repeated a total of three times,
   followed by centrifugation at 400 x g for 10 min. Cell pellet was
   resuspended in 2 ml SAGM medium. A discontinuous Percoll gradient was
   prepared by pipetting 10 ml low density Percoll in a tube and layering
   10 ml high density Percoll below. Resuspended cell pellet was layered
   on top followed by centrifugation at 400 x g, 20 min, 4  °C. Interface
   was collected, washed with PBS, and pelleted^[284]31. AECIIs were
   cultured for 7 days in SAGM supplemented with 2% FCS and 1%
   penicillin/streptomycin (#15140122, Thermo Fisher) on a 0.4 µm ThinCert
   transwell (#665641, Greiner Bio-One) at a density of 5 ×105 cells /
   113.1 mm^2, followed by stimulations and cell harvesting, as indicated.
   For the generation and stimulation of PCLS, lung tissue (JLU-FB11
   ethics approval number: AZ31/93 and AZ58/15) with healthy appearance
   (e.g. tumor edge tissue) from patients undergoing surgery was
   processed. Tissue was filled with 3% low-melting agarose diluted in
   medium with PBS (1:1) and filled through the bronchi. Sections of 300
   to 400 µm thickness were generated using a vibratome (Leica
   Biosystems). PCLS were cultured in 12-well dishes with DMEM,
   supplemented with GlutaMAX (#35050061, Thermo Fisher), 10% FCS, and
   1:500 Normocin (#ant-nr-05, Invivogen) for 4 days and one additional
   day without antibiotics prior to further treatment.

   Alveolar macrophages and patient BAL-pellets: Bronchoalveolar lavage
   (BAL) fluid (BALF) samples were collected from healthy donors or from
   patients who underwent BAL for pulmonary disease assessment at the
   University Clinics Giessen and Marburg. Each person, undergoing
   bronchoscopy including bronchoalveolar lavage (BAL), gave both oral and
   written informed consent. The study received approval from the local
   ethics committee (UMR-FB20 ethics approval number: Studie 87/12 and
   Studie 168/12). During bronchoscopy, a flexible fiberoptic bronchoscope
   was used, and BAL was performed by instilling 150 ml of pre-warmed
   sterile 0.9% NaCl solution. BAL was done in the right middle lobe or
   the lingula. For alveolar macrophage (aMФ) isolation, BALF from healthy
   donors was centrifuged at 400 x g for 10 min and cell pellet was washed
   once with PBS, followed by resuspension in X-Vivo 15 medium
   (#BEBP02-061Q, Lonza) supplemented with 5% FCS, 0.1%
   penicillin/streptomycin and 25 µg/ml gentamicin (#15750060, Gibco).
   Cells were then filtered through a 100 µm cell strainer. aMФs were
   counted based on morphology, seeded at a density of 4 ×105 cells/ml,
   and incubated for 2 h at 37  °C and 5% CO[2]. Medium was then replaced
   with fresh medium and cells were stimulated as indicated and collected.
   BALF from patients (Supplementary Table [285]1) was centrifuged at 300
   x g for 10 min and cell pellet was snap-frozen in liquid nitrogen,
   followed by −80 °C storage and eventually RNA-isolation and RT-qPCR as
   described below.

   Blood-derived and THP1 macrophages: Buffy coats were obtained from the
   Transfusion Medicine department, University Clinics Giessen and
   Marburg. Leukocytes were isolated using Lymphoprep (#18061, Stemcell
   Technologies) and MACS-purification (#130-050-201, Miltenyi CD14+
   beads). Isolated monocytes were cultured for 7 days in X-Vivo 15 medium
   supplemented with 5% FCS and 15 ng / ml GM-CSF or 15 ng / ml M-CSF to
   obtain blood-derived macrophages (G-MФ and M-MФ, respectively). THP1
   cells were maintained at a density of 2 × 105 – 1 ×106 cells / ml in
   RPMI-1640 (#31870074, Thermo Fisher), supplemented with 10% FCS, 1%
   sodium pyruvate (#11360070, Thermo Fisher), 1% GlutaMAX (Thermo Fisher)
   and 1% penicillin/streptomycin solution. THP1 cells were differentiated
   into macrophage-like cells with 80 nM phorbol 12-myristate 13-acetate
   (PMA) for 72 h.

   Cell and tissue stimulations and cultivation conditions: If not
   specified otherwise in the figures, cells were treated with the
   following stimuli for the indicated durations: 1 µg/ml flagellin from
   Salmonella enterica serovar Typhimurium (FLA-ST, Invivogen), 100 ng/ml
   LPS from Salmonella enterica serovar Typhimurium, (BioXtra L6143,
   Sigma), 100 ng / ml recombinant human IFNα (rh IFNα 2 A Stemcell) or
   200 ng/ml Pam3CSK4 (tlrl-pms, Invivogen). For PCLS stimulations
   500 ng/ml LPS or FLA-ST were used. For bacterial infections, all
   bacteria were cultured in LB medium, with the exception of S.
   pneumoniae, which was cultured in THY medium. Cells were infected at a
   multiplicity of infection (MOI) of 10. For pathway inhibition, cells
   were pre-treated with the respective inhibitor or DMSO for 2 h followed
   by further treatments. Inhibitors used were: BAY 11-7082 (NFκB and
   NLRP3 inflammasome inhibitor, #tlrl-b82, Invivogen; 1, 5 or 10 µM),
   Ruxolitinib (JAK-STAT inhibitor, #FBM-10-4511, Biozol; 0.1, 1 or 5 µM)
   and BX795 (TBK1/IKKε inhibitor, #tlrl-bx7, Invivogen; 0.1, 1 or 5 µM).
   All cells were maintained at 37  °C in a humidified atmosphere with 5%
   CO[2].

Gene silencing and overexpression

   Stable knockdown cell lines were generated using the CRISPR
   interference (CRISPRi) vector pLV-hU6-sgRNA-hUbC-dCas9-KRAB-T2a-GFP
   (Addgene #71237). Guide RNAs (gRNAs; Supplementary Table [286]2),
   targeting the transcriptional start site regions of genes of interest,
   were cloned into the CRISPRi vector, followed by lentiviral particle
   production and transduction as described below. Control cell lines were
   generated by transducing cells with the same empty CRISPRi vector,
   lacking a guide RNA insert. The transduced cells were sorted for
   GFP-expression (Aria III cell sorter, BD; Supplementary Fig. [287]12)
   and brought back into culture.

   For LNA Gapmer and siRNA based gene silencing, 5 × 10^5 THP1 cells or
   CD14+ monocytes (12-well format) were differentiated into
   macrophage-like cells as described above. After differentiation, medium
   was exchanged with antibiotic free medium. For transfection, 40 pmol
   antisense LNA GapmeR or 50 pmol SilencerSelect siRNA were complexed
   with 1.5 µl Lipofectamine 2000 (#11668019, Thermo Fisher) in 200 µl
   OptiMEM (#31985062, Thermo Fisher) and incubated for 15 min at room
   temperature. Liposomes were added to the cells, followed by
   centrifugation at 850 x g, 37  °C for 2 h. Macrophages were cultured
   for additional 40 h, followed by stimulations and cell harvesting.
   Knockdowns were verified by RT-qPCR. LNA GapmeRs and siRNAs used in the
   present study are listed below (Supplementary Table [288]3).

   For generation of overexpression cell lines, the pLVX-M-puro
   lentivector (Addgene #125839) was used. The GATA2 coding sequence was
   amplified from a pLXV-zsGreen lentiviral plasmid containing the GATA2
   coding sequence (a gift from Bryan Heit^[289]29). For PCR amplification
   the following primers were used:

   5’-tccgatggatcccgATGGAGGTGGCGCCCGAGCAG-3 (forward, primer number
   OBS-4006)

   5’-atcggagaattcCTAGCCCATGGCGGTCACCAT-3’ (reverse, primer number
   OBS-3985)

   The GATA2 coding sequence was cloned into the multiple cloning site of
   the pLVX-M-puro vector (BamHI and EcoRI), followed by lentiviral
   particle production and transduction of THP1 cells as described below.
   The empty pLVX-M-puro vector served as a control. Positive selection of
   transduced cells was achieved using puromycin (2 µg/ml, 3 days).

Lentiviral transduction

   Calcium phosphate transfection was used for co-transfection of 10 µg
   lentiviral vector, pseudotyping- and helper-plasmid (3.5 µg pVSVG and
   6.5 µg psPAX2) into HEK293T cells. To this end, 3.5 × 10^6 cells were
   seeded in a cell culture dish (100 × 20 mm) with 8.5 ml DMEM
   (#41965039, Gibco), 10% FCS and cultivated overnight, leading to a
   confluence of 70-85%. The plasmids, 106.3 µl 2 M CaCl[2] and 743.7 µl
   H[2]O (minus plasmid volume) were mixed and 2x HBS was added dropwise
   while vortexing. The mixture was incubated for 10 min at room
   temperature, added to the cells and incubated at 37  °C, 5% CO[2] for
   8 h. The cells were then washed with PBS and cultured in DMEM for 2
   days. The supernatant containing the viral particles was passed through
   a 0.45 µM filter and concentrated overnight by ultracentrifugation
   (Sorvall Discovery 90SE, Beckman Coulter Rotor 70 Ti) at 64,000 x g and
   4  °C. 5 ×105 cells were seeded in a 12-well and transduced by adding
   the viral pellet resuspended in RPMI-1640. Medium containing viral
   particles was exchanged after 2 days. 3-4 days after transduction,
   cells were sorted for GFP expression (Supplementary Fig. [290]12) or
   selected with puromycin, as described above.

Subcellular fractionation (RNA and Protein)

   Cytoplasm/nucleus fractionations for RNA measurements were carried out
   by adding 200 µl lysis buffer to 5 × 10^6 cells per condition (10 mM
   Tris pH 8, 140 mM NaCl, 1.5 mM MgCl[2], 0.5% Igepal, 2 mM vanadyl
   ribonucleoside complex), followed by incubation on ice for 5 min.
   Lysate was centrifuged at 1000 x g, 4  °C for 3 min and the supernatant
   (cytosolic fraction) was transferred to a new tube. The pellet was the
   nuclear fraction. The cytosolic fraction was centrifuged again at
   15,000 x g, 4  °C for 10 min and the supernatant was transferred to a
   new tube. The nuclear fraction was washed twice with 160 µl lysis
   buffer (centrifugations at 1000 x g, 4  °C for 3 min) and finally
   resuspended in 100 µl lysis buffer. RNA from both fractions was
   isolated with TRIzol (#15596026, Ambion) as described below. An equal
   amount of Escherichia coli K12 total RNA was spiked in before adjusting
   the concentrations of cytoplasmic and nuclear RNA for RNA-seq library
   preparation, described below.

   Cytoplasm/nucleus fractionations for protein measurements were carried
   out by washing 5 ×106 cells per condition with PBS and collection of
   the cells by centrifugation at 1000 x g for 8 min. Cells were then
   resuspended in 400 µl buffer A (10 mM Hepes pH 7.9, 10 mM KCl, 0.1 mM
   EDTA, 0.1 mM EGTA, 0.5 mM DDT, 1x cOmplete EDTA-free (#11836170001,
   Roche)), and incubated on ice for 15 min. Lysis was achieved by passing
   the cells through a 26 G needle attached to a 1 ml syringe (7-8 times).
   Upon centrifugation at 5000 x g for 2 min, supernatant (cytosolic
   fraction) was transferred to a new tube. The pellet was the nuclear
   fraction. The cytosolic fraction was centrifuged at 20,000 x g for
   20 min and the supernatant was transferred to a new tube. The nuclear
   fraction was washed twice with buffer A and the nuclei pellet was
   resuspended in 100 µl buffer B (20 mM Hepes pH 7.9, 400 mM NaCl, 1 mM
   EDTA, 1 mM EGTA, 0.5 mM DDT, 1x cOmplete EDTA-free (Roche)), followed
   by 1 h incubation on a shaker at 4  °C. To obtain a pure nuclear
   extract, the obtained lysate was centrifuged at 20,000 x g for 20 min
   and the supernatant was transferred to a new tube. Proteins were
   precipitated from the cytosolic and nuclear fractions by adding three
   times the volume of ice cold acetone, followed by overnight incubation
   at −20 °C. Proteins were pelleted by centrifugation at 17,000 x g and 4
   °C for 30 min, air-dried and resuspended in 50 µl 8 M Urea, 0.1 M
   ammonium-bicarbonate.

ChIRP-seq and -qPCR

   DNA antisense probes for ChIRP (Supplementary Table [291]4) were
   synthesized at Metabion AG. For 3’ mono-biotinylation, terminal
   transferase (#M0315, New England Biolabs) and Biotin-11-ddUTP
   (#NU-1619-BIOX-L, Jena Bioscience) were used according to the
   manufacturer’s instructions. 2 × 10^7 cells were used per capture and
   ChIRP-seq was performed as described previously^[292]32. Briefly, to
   preserve RNA–chromatin interactions, cells were cross-linked with 20 ml
   of 1% glutaraldehyde in PBS for 10 min at room temperature while
   shaking. Cross-linking was stopped by adding 1/10^th volume of 1.25 M
   glycine and incubating for 5 min. Cells were pelleted (2000 × g,
   5 min), washed with ice-cold PBS, flash-frozen in liquid nitrogen, and
   stored at –80 °C. Frozen pellets were thawed, weighed, and lysed in
   300 µl buffer (50 mM Tris-Cl pH7.0, 10 mM EDTA, 1% SDS, 1× cOmplete
   EDTA-free (Roche), 1 mM PMSF (#93482, Sigma-Aldrich), 200U/ml murine
   RNase inhibitor (#M0314, NEB)). Lysates were sonicated using a
   Bioruptor Plus (Diagenode) at high intensity (30 s on, 45 s off,
   90 min) in 15 ml tubes. Fragmentation (100–500 bp) was verified by
   treating 5 µl of lysate with 90 µl proteinase K (PK) buffer (100 mM
   NaCl, 10 mM Tris-Cl pH8.0, 1 mM EDTA, 0.5% SDS, 5% [v/v] Proteinase K
   (#AM2546, Ambion)), followed by DNA purification (GeneJET PCR Kit,
   #K0701, Thermo Scientific), and analysis on a 1% agarose gel. Sonicated
   lysates were transferred to 2 ml tubes, adjusted to 1 ml final volume
   with lysis buffer, flash-frozen in liquid nitrogen, and stored at –80
   °C. For hybridization, samples were thawed; 10 µl aliquots were put
   aside as RNA and DNA inputs. The remaining sample was split in half and
   each portion mixed with 2 ml hybridization buffer (750 mM NaCl, 1% SDS,
   50 mM Tris-Cl pH7.0, 1 mM EDTA, 15% formamide, 1× cOmplete EDTA-free,
   1 mM PMSF, 200 U/mL RNase inhibitor). Biotinylated DNA probes
   (100pmol/ml chromatin) were added and incubated at 37 °C for 4 h with
   shaking. C-1 streptavidin beads (100 µl per 100pmol probe, #65001,
   Invitrogen) were washed thrice with 1 ml lysis buffer, resuspended in
   100 µl lysis buffer, added to the lysate, and incubated for 30 min at
   37 °C with shaking. Beads were washed five times with 1 ml pre-warmed
   wash buffer (2× SSC, 0.5% SDS, 10 mM PMSF), including 5-min incubations
   at 37 °C. After the final wash, beads were resuspended and separated
   into 100 µl for RNA and 900 µl for DNA isolation.

   For RNA isolation, 95 µl RNA PK buffer (100 mM NaCl, 10 mM Tris-Cl
   pH7.0, 1 mM EDTA, 0.5% SDS, 5% [v/v] Proteinase K) was added to the
   input aliquot and 100 µl to the bead-bound samples. After incubation at
   50 °C for 45 min and denaturation at 95 °C for 10 min, samples were
   chilled on ice. RNA was isolated and qPCR was performed as described
   below.

   For DNA extraction, beads were resuspended in 150 µl and input aliquot
   in 140 µl DNA elution buffer (50 mM NaHCO[3], 1% SDS, 100 µg/mL RNase
   A). Samples were incubated at 37 °C for 30 min with shaking, and
   supernatants were collected using a magnetic stand. The extraction was
   repeated and pooled. Proteinase K (15 µl, 20 mg/ml) was added, followed
   by incubation at 50 °C for 45 min. DNA was purified via
   phenol:chloroform:isoamyl extraction and used for high throughput
   sequencing and RT-qPCR.

ChIRP-MS

   DNA antisense probes used for ROCKI and LUCAT1 ChIRP-MS are shown in
   Supplementary Table [293]4. Probes were 3’-monobiotinylated as
   described above. 2 × 10^8 cells were used per capture. ChIRP-MS was
   performed as described previously^[294]33, with the following
   modifications: (1) no RNase-treatment control; (2) proteins were
   on-bead digested and eluted from the beads as specified in the mass
   spectrometry section. Briefly, to preserve RNA–protein interactions, 4
   × 10^8 cells were cross-linked in 3% formaldehyde in PBS for 30 min at
   room temperature while shaking. Cross-linking was stopped by the
   addition of 0.125 M glycine and incubation for 5 min. Cells were
   pelleted by centrifugation at 2000 × g for 5 min, washed once with
   ice-cold PBS, and centrifuged again. The supernatant was removed, and
   the pellet was weighed, flash-frozen in liquid nitrogen, and stored at
   –80 °C until further processing. Frozen pellets were dissolved in cell
   lysis buffer (50 mM Tris-Cl pH7.0, 10 mM EDTA, 1% SDS, 1× cOmplete
   EDTA-free protease inhibitor cocktail (Roche), 1 mM PMSF
   (Sigma-Aldrich), and 200 U/mL RNase inhibitor (NEB)) and sonicated
   using a Bioruptor Plus (Diagenode) at high intensity (30 s on / 45 s
   off) until the lysate became clear. DNA fragmentation was assessed by
   mixing 5 µl of lysate with 90 µl PK buffer (100 mM NaCl, 10 mM Tris-Cl
   pH8.0, 1 mM EDTA, 0.5% SDS), followed by centrifugation at 16,000 × g
   for 10 min. The resulting pellet and supernatant were treated
   separately with 5 µl Proteinase K (5% v/v; Ambion) and incubated at 50
   °C for 45 min. DNA was purified using the GeneJET PCR Purification Kit
   (Thermo Scientific), and the fragment size (1–2 kb) was verified by 1%
   agarose gel electrophoresis. Lysates were flash-frozen and stored at
   –80 °C.

   For hybridization, thawed lysates were aliquoted (10 µl) for RNA and
   protein input controls and incubated at 37 °C. The remaining lysate was
   divided equally. To pre-clear the lysate, 30 µl of C-1 streptavidin
   magnetic beads were washed twice with 100 µl lysis buffer, incubated
   with the lysate at 37 °C for 30 min with shaking, and removed using a
   magnetic stand. The supernatant was transferred to a new tube and this
   step was repeated to remove any residual beads. Hybridization was
   carried out by adding 2 volumes of hybridization buffer (750 mM NaCl,
   1% SDS, 50 mM Tris-Cl pH7.0, 1 mM EDTA, 15% formamide, 1× cOmplete
   EDTA-free (Roche), 1 mM PMSF, and 200 U/mL RNase inhibitor) and 1 µl of
   100 µM biotinylated DNA probes per 1 ml of lysate. Samples were
   incubated at 37 °C for 16 h with shaking. Prior to use, 100 µl of C-1
   streptavidin beads per 1 µl probe were washed three times with 1 ml
   lysis buffer. Following hybridization, 1 ml of the lysate was mixed
   with the washed beads and incubated for 30 min at 37 °C with shaking.
   Beads were then separated on a magnetic stand and washed five times
   with 1 ml of wash buffer (2× SSC, 0.5% SDS, 10 mM PMSF) at 37 °C, with
   5 min incubation per wash. The final wash was performed using 2× SSC to
   remove residual detergents.

   Beads were then separated magnetically, with 100 µl retained for RNA
   isolation and the remaining fraction used for protein analysis. RNA was
   extracted as described in the ChIRP-seq section, and proteins were
   processed as detailed in the mass spectrometry protocol.

(RT-)qPCR

   Total RNA was isolated using TRIzol (Ambion) and contaminating gDNA was
   removed by treatment with DNaseI (Thermo Fisher) in the presence of
   recombinant murine RNase inhibitor (NEB), followed by PCI extraction.
   RNA concentration was determined using a Nanodrop 2000 spectrometer
   (Thermo Scientific). cDNA was generated (#4368813, High-Capacity cDNA
   Reverse Transcription Kit, Thermo Fisher), and RT-qPCR was performed
   using Luna Universal qPCR Master Mix (#M3003, NEB). For
   RNA-quantifications in subcellular fractionation and ChIRP-capture
   experiments, the Luna Universal One-Step RT-qPCR Kit (#E3005, NEB) was
   used. For DNA quantifications in ChIRP-capture experiments, the Luna
   Universal qPCR Master Mix (NEB) was used. All (RT-)qPCRs were performed
   on a QuantStudio 3 instrument (Applied Biosystems). Relative expression
   changes were calculated using the 2^-ΔΔCT method^[295]34, based on CT
   values and using U6 snRNA as a housekeeper control, unless stated
   otherwise. For BAL patient samples, equal amounts of RNA were reverse
   transcribed, followed by RT-qPCR analysis as described above. However,
   CT values were not processed using the 2^-ΔΔCT method. Instead, log2
   2^-ΔCT values were calculated for the lncRNAs shown in Fig. [296]1I, as
   well as for U6 snRNA and RPS18 mRNA as controls (Supplementary
   Fig. [297]2D), using patient # 3 (the sample with the highest IFNB1
   CTs), as a reference. These values were then plotted against IFNB1 log2
   2^-ΔCT values, followed by linear regression and Pearson correlation
   analysis. Primer oligos used for RT-qPCR are listed in Supplementary
   Table [298]5.

High Throughput Sequencing

   RNA was isolated using TRIzol reagent as described above and handed
   over to the in-house genomics core facility (Philipps-University
   Marburg, medical faculty) or Genewiz GmbH for library generation and
   sequencing. RNA quality was validated (Agilent RNA 6000 Nano Kit).
   Illumina Truseq stranded total RNA libraries with rRNA depletion
   (subcellular fractionation experiments) or Lexogen QuantSeq 3’ mRNA-Seq
   FWD libraries (all remaining RNA-seq experiments) were generated.
   Libraries were paired end sequenced with 2 × 150 bp (Truseq libraries)
   or single end sequenced with 75-bp read length (Lexogen libraries) on a
   NextSeq 550 instrument. ChIRP-seq stranded genomic DNA libraries were
   generated at Vertis Biotech AG (Freising, Germany) using in-house
   protocols and sequenced on a NexSeq 550. External RNA-seq data
   (Supplementary Table [299]6) were downloaded from the NCBI GEO
   repository using the SRA tool ‘prefetch’. Downloaded.sra files were
   converted to.fastq files using the SRA tool ‘fastq-dump’.

   Demultiplexed Fastq files were imported into the CLC Genomics Workbench
   v 10.0.1. Sequence reads were trimmed using the CLC workbench with
   quality score limit set to 0.05 and 25 nt minimal read length. Trimmed
   sequences were mapped to the GRCh38 human reference genome using the
   CLC workbench with the following parameters: Mismatch cost: 2.
   Insertion cost: 3. Deletion cost: 3. Length fraction: 0.8. Similarity
   fraction: 0.8. Max. hits for a read: 10. Strand specific: both. The
   subcellular fractionation RNA-seq data were additionally normalized to
   Escherichia coli spike-in RNA. To this end, the human GRCh38 (GENCODE)
   and E. coli K12 MG1655 (NCBI:GCF_000005845.2_ASM584v2) reference
   annotation- and genome-files were fused and upon read mapping (with the
   same parameters as above), separate tables of uniquely mapping reads
   were created for both organisms. Human RPKMs were calculated based on
   the human read table and subsequently normalized to the percentage of
   E. coli reads in each sample. For all other RNA-seq experiments,
   normalized gene expression values (e.g. RPKM) were calculated based on
   the raw read counts obtained (uniquely mapping reads). To focus on
   reliably detectable RNAs, all RNA-seq data tables were filtered to
   include only RNAs with RPKM values ≥ 0.5 in all experimental replicates
   under at least one of the compared conditions. This filtering
   criterion, combined with a minimal number of 10 sequencing reads
   detected under at least one condition, was applied uniformly in
   Fig. [300]1, where the lncRNAs central to this study were identified.
   In subsequent analyses, the lncRNAs included in Fig. [301]1E were
   exempted from this filter to ensure their consistent inclusion in all
   downstream analyses, while all other RNAs continued to be subject to
   the original filtering criteria. For experiments where RNA-seq data
   were obtained in triplicates, DeSeq2^[302]35 was executed in R Studio
   to calculate expression changes and p-values. For visualization of read
   coverages, mapped read data were exported in BAM format and files were
   analysed using the Integrated Genomics Viewer (IGV^[303]36).

Orthogonal organic phase separation (OOPS)

   3 × 10^6 monocytes isolated from buffy coats were seeded into a 90-mm
   dish, differentiated, and stimulated with LPS as described above. OOPS
   was conducted as described^[304]37. Protein from the input, eluate
   (RNase-treatment), and control-elution (no RNase treatment) samples was
   precipitated using acetone, collected by centrifugation, and air-dried.
   Pellets were resuspended in 8 M Urea, 0.1 M ammonium-bicarbonate for
   mass spectrometry analysis, as described below. A detailed protocol for
   the OOPS procedure is provided in Supplementary Note [305]1.

Protein mass spectrometry

   For ChIRP-MS and CoIP-MS analysis, capture beads were transferred to
   the mass spectrometry facility, and proteins were digested on-bead and
   eluted as follows: Samples were washed three times with 100 µl
   ammonium-bicarbonate (50 mM, pH 8.0) each. Trypsin (0.1 µg in 50 µl
   ammonium-bicarbonate buffer) was added to the beads and samples were
   incubated at 37  °C for 45 min. The supernatant was transferred into
   fresh tubes and digested over-night at 37  °C. For reduction of
   disulphide bridges 5 mM DTT (Dithiothreitol) was added. Samples were
   then incubated for 15 minutes at 95  °C. Subsequently, the
   sulfhydryl-groups were chemically modified by adding iodoacetamide to a
   final concentration of 25 mM and incubating samples for 45 min at room
   temperature (RT) in the dark. Excess iodoacetamide was quenched by the
   addition of 50 mM DTT and incubation for one more hour at RT.

   For whole proteome analysis, 1 × 10^6 cells were differentiated and
   stimulated as described above. Cells were washed once with PBS,
   pelleted by centrifugation, snap frozen, and stored at -80  °C till
   further use. For proteomics analysis, the cells were lysed (8 M Urea,
   0.1 M Ammoniumbicarbonate) and sonicated (Bioruptor Plus, Diagenode)
   with 2 × 10 seconds on, 30 seconds off and intensity setting “high”.
   The lysate was cleared by centrifugation and protein concentration was
   determined using the BCA method (#23225, Pierce™ BCA Protein Assay Kit,
   ThermoFisher and an Infinite PRO (Tecan) plate reader). Samples were
   adjusted to equal protein concentrations and mass spectrometry analysis
   was done. Specifically, to each sample solubilized in 8 M Urea buffer,
   1.25 µl of a 0.2 M TCEP solution in 0.1 M NH[4]HCO[3] was added and
   samples were incubated for 1 h on a shaker at 37  °C and 1000 rpm.
   After cooling to 25  °C, 1.25 µl of 0.4 M iodoacetamide in bidest water
   was added. Subsequently, samples were incubated in the dark at 25  °C
   and 500 rpm. Then, 1.25 µl of a 0.5 M N-acetyl-cystein solution in
   0.1 M NH[4]HCO[3] was added and samples were incubated for 10 min at
   25  °C and 500 rpm. If necessary, pH was adjusted to 8-9. Then, the
   urea concentration was decreased to 6 M urea by dilution with 0.1 M
   NH[4]HCO[3] and 2.5 µl of a 0.2 µg / µl LysC solution in 0.1 M
   NH[4]HCO[3] were added. Samples were then incubated at 37  °C for 3 h.
   Subsequently, samples were further diluted using 0.1 M NH[4]HCO[3],
   resulting in a final urea concentration of 1.6 M. Then, 4 µl of 0.5 µg
   / µl Sequencing Grade Modified Trypsin (Serva) in 0.1 M NH[4]HCO[3]
   were added. Samples were incubated at 37  °C over-night. TFA was added
   to a final concentration of ~1% to yield a pH below 2. Samples were
   centrifuged at 14,000 rpm.

   Both for “On-Bead digestion” samples and samples solubilized in 8 M
   urea buffer, the resulting peptides were desalted and concentrated
   using Chromabond C18WP spin columns (Macherey-Nagel, Part No. 730522),
   according to the manufacturer’s protocols. Finally, peptides were
   dissolved in 25 µl of water with 5% acetonitrile and 0.1% formic acid.
   Peptide concentration was measured using a Nanodrop (Thermo Scientific)
   and samples were diluted to 100 ng of peptides per µl. Mass
   spectrometric analysis of the samples was performed using a timsTOF Pro
   mass spectrometer (Bruker Daltonic). A nanoElute HPLC system (Bruker
   Daltonics), equipped with an Aurora 25 cm × 75 µm C18 RP column filled
   with 1.7 µm beads (IonOpticks) was connected online to the mass
   spectrometer. A portion of approximately 200 ng of peptides
   corresponding to 2 µl was injected directly on the separation column.
   Sample loading was performed at a constant pressure of 800 bar.
   Separation of tryptic peptides was achieved at 60  °C column
   temperature with either a 30 min gradient or a 100 min gradient of
   water / 0.1% formic acid (solvent A) and acetonitrile/0.1% formic acid
   (solvent B) at a flow rate of 400 nl / min. 30 min gradient: Linear
   increase from 2% B to 17 % B within 18 minutes, followed by a linear
   gradient to 25% B within 9 min and linear increase to 37% solvent B in
   additional 3 min. Finally, B was increased to 95% within 10 min and
   hold at 95% for additional 10 min. 100 min gradient: Linear increase
   from 2% B to 17% B within 60 min, followed by a linear gradient to 25%
   B within 30 min and linear increase to 35% solvent B in additional
   10 min. Finally, B was increased to 95% within 10 min and hold for
   additional 10 min. The built-in “DDA PASEF-standard_1.1sec_cycletime”
   method developed by Bruker Daltonics was used for mass spectrometric
   measurement. Data analysis was performed using MaxQuant (different
   versions) with Andromeda search engine against the human Uniprot
   database. Peptides with a minimum of seven amino-acid lengths were used
   and FDR was set to 1% at the peptide and protein level. Protein
   identification required at least one razor peptide per protein group
   and label free quantification (LFQ) algorithm was applied.
   Perseus^[306]38 was used for further evaluation of MaxQuant results,
   including contaminant removal and calculation of protein fold-changes
   and significances (two-tailed Student’s t-test), based on LFQ and iBAQ
   abundance values. For ChIRP-MS data analysis, the average LFQ values of
   control samples were artificially set to 100 when all values in this
   group were zero, to enable inclusion of proteins exclusively detected
   in the lncRNA capture samples for volcano plot visualizations
   (Fig. [307]6C and Supplementary Fig. [308]6E).

GRADR

   For GRADR analysis of RNA-protein interactions, Grad-seq data^[309]8,
   OOPS-MS data and subcellular fractionation RNA-seq and MS data were
   combined into one tabular dataset (Supplementary Data [310]8). For RNAs
   annotated as cytoplasmic (Supplementary Data [311]7), the GRADR dataset
   was filtered for RNA-binding proteins (OOPS-MS RNase + vs – elution
   fold-change ≥ 2, p-value ≤ 0.05 [Student’s t-test]) detectable in the
   cytoplasm (proteins detected in the cytoplasmic fractions in all three
   independent replicates and not being ≥ 60% nuclear in all three
   replicates). Pearson correlation analysis was conducted, comparing the
   Grad-seq co-sedimentation profile of the RNA in question to the
   Grad-seq profile of each protein present in the dataset after applying
   the filtering steps described above. The dataset was sorted in
   descending order by the Pearson r values and entries with r
   values ≥ 0.5 were considered as potential protein-interactors of the
   RNA in question. For nuclear RNAs (Supplementary Data [312]7), the
   procedure described above was carried out analogously for nuclear
   proteins and additionally, mitochondrial proteins were removed from the
   dataset due to the implausible interaction of nuclear and mitochondrial
   factors. A detailed step-by-step protocol for the GRADR procedure is
   provided in Supplementary Note [313]1.

UV crosslinking & co-immunoprecipitation

   Co-IP was performed with 2 × 10^7 cells per capture according to a
   previously published procedure^[314]39, with minor modifications.
   UV-cross-linking was carried out with 300 mJ / cm^2 (HL-2000
   HybriLinker) and the protein of interest was isolated from the cell
   lysate with capture antibody-coated protein G dynabeads (#10003D,
   Thermo Fisher). Per Co-IP, 2.5 µg of hnRNP L- and control-IgG
   antibodies were used. RNA was isolated by boiling the beads, followed
   by PCI extraction and 30:1 ethanol / 3 M sodium acetate precipitation.
   For Co-IP with subsequent mass spectrometry analysis, the capture
   dynabeads were washed with 50 mM HEPES, pH 7.5 and resuspended in 0.1 M
   ammonium bicarbonate and further processed (mass spectrometry core
   facility, Philipps-University Marburg) as described above. Antibodies
   used for Co-IP experiments are listed below (Supplementary
   Table [315]7).

Western blotting

   Co-IP samples were separated on a 10% polyacrylamide gel by SDS-PAGE.
   For Western blotting, proteins were transferred onto nitrocellulose
   (Amersham Protran; Sigma-Aldrich; GE10600003) by semidry blotting.
   Blots were blocked with TBST buffer containing 5% milk powder and 3%
   BSA. Detection antibodies were used at a 1:10,000 (anti-hnRNP L) or
   1:1000 (anti-mouse-HRP) dilution. Blot development was carried out
   using the ECL Prime Western Blot Detection kit (GERPN2232, Amersham)
   and a Chemostar Imager (INTAS Science Imaging). Western blot full scans
   can be found in Supplementary Fig. [316]8.

FACS-based ELISA

   Cytokine content in cell culture supernatants was detected using the
   LEGENDplex human inflammation panel 1 multiplex FACS kit (BioLegend,
   Cat: 740809, Lot No. B347171). The experiment was performed according
   to the manufacturer’s protocol, using a BD FACSymphony A1 flow
   cytometer and BD FACS diva 9.0 software. Results were analyzed using
   the LEGENDplex data analysis software suite.

LncRNA conservation analysis

   Conservation analysis was conducted on macrophage lncRNAs identified in
   Fig. [317]1E as being up-regulated ≥ 2-fold in response to LPS.
   ENSEMBL-annotated MANE transcript variants were used, or alternatively
   the transcript variants best matching macrophage RNA-seq coverage. To
   avoid biases associated with coding sequence conservation, lncRNA
   regions overlapping protein-coding genes were clipped. Sequence
   conservation was assessed using NCBI BLASTN, comparing human lncRNA
   sequences against the main RefSeq reference genomes of the species
   included in Fig. [318]3A. BLAST hits with ≥ 20 complementary
   nucleotides were retained, and the sequence window with the longest
   query sequence coverage (within a maximum 100 kb region) was considered
   the orthologous locus. If BLAST hits mapped to multiple loci, syntenic
   regions were prioritized. Percent base conservation was calculated from
   all BLAST hits within this window. For genus-level generation lengths
   (Fig. [319]3A), the available data for all species within each genus
   were averaged^[320]40. Evolutionary distances between species were
   inferred using data from the TimeTree database
   ([321]http://www.timetree.org/home).

Computational prediction of ROCKI-protein interactions

   To identify potential hnRNP binding sites on the ROCKI transcript
   ENST00000434296, we used a combination of searching for known binding
   sequence motifs and predicting binding sites by machine learning models
   trained on publicly available CLIP data.

   For the sequence motif search, two types of motifs were used: sequence
   motifs in MEME motif format^[322]41 (visualized as sequence logos) or
   regular expressions/strings (for hnRNP C and hnRNP D). Sequence motifs
   were extracted from catRAPID omics v2.0 and ATtRACT curated motif
   databases^[323]42,[324]43. MEME motif search was conducted using
   FIMO^[325]44 as part of the RBPBench package
   ([326]https://github.com/michauhl/RBPBench), with a set p-value
   threshold for reporting motif hits of 0.0002. The threshold was chosen
   such that for each RBP ≥ 1 hit was reported. Regular expression search
   was conducted with RBPBench.

   To computationally predict hnRNP binding sites, we used publicly
   available eCLIP data from ENCODE^[327]45 to train predictive models for
   two methods: GraphProt and RNAProt^[328]26,[329]27. Data was available
   for hnRPN C and hnRNP L, while for the remaining hnRNPs there was
   either no eCLIP data available (hnRNP A2B1, hnRNP A3, hnRNP D) or not
   enough recovered binding sites for efficient training (hnRNP A1). IDR
   peak region files were downloaded from ENCODE and used as positive
   instances for training, while the negative instances were generated
   using RNAProt. Both GraphProt and RNAProt were trained using default
   settings (using only sequence features), with variable sequence length
   input and an up- and downstream site extension of 10 nt. For RNAProt,
   hits after applying an internal threshold (--thr 1) were used, while
   for GraphProt we used the top 10 predicted sites ranked by GraphProt
   score.

   To search for further ROCKI binding proteins, we examined the overlap
   of the ROCKI full-length transcript with ENCODE available eCLIP
   data^[330]45 using bedtools intersect –s –wao. Identified peaks were
   ranked by p-value, as reported in Supplementary Fig. [331]9C.

Statistical analysis and data visualization

   Statistical analysis of RNA-seq and proteomics data with ≥ 3
   experimental replicates was performed using Deseq2 and Perseus
   (two-tailed Student’s t-test), respectively, as described above.
   Regulations ≥ 2-fold with p-values ≤ 0.05 were considered statistically
   significant. Further data (e.g. RT-qPCR measurements) were analyzed in
   GraphPad PRISM and two-tailed Student’s t-test was used for two-sample
   comparisons, and One-way ANOVA test was used for multiple-sample
   comparisons. P-values are written out in the figure panels, where
   possible, or where not possible, due to space limitations, statistical
   significance is indicated with an asterisk. Where appropriate,
   individual measurement points are shown. Else, error bars are provided,
   illustrating the standard deviation (SD). For heatmap visualization of
   RNA-seq and proteomics data, hierarchical clustering was done using
   Cluster 3.0^[332]46 and heatmaps were generated using JAVA
   TreeView^[333]47. Pathway enrichment analysis was performed using
   ConsensusPathDB^[334]48 or ENRICHR^[335]49, as indicated. GO-term UMAPs
   (Fig. [336]4C) were generated using ENRICHR and processed as follows:
   For genes up- and down-regulated upon knockdown of each lncRNA, the
   associated “Biological Process” GO-terms were highlighted within the
   pre-computed ENRICHR GO-term UMAP obtained by following the provided
   Appyter link. For each lncRNA, the top 10 GO terms associated with the
   up- and down-regulated gene sets were identified. These GO-terms were
   marked on the UMAP, and lines were drawn in a vector graphic overlay to
   connect each GO-term to the corresponding lncRNA. In this
   representation, each lncRNA is depicted as a white circle with a black
   outline, positioned next to its associated GO-terms in a way allowing
   clear visualization of all connections. For UMAP representation of
   sedimentation similarities of cellular protein machineries and lncRNAs
   (Fig. [337]5H), the “PCA, t-SNE, UMAP” Appyter
   ([338]https://appyters.maayanlab.cloud/)^[339]50 was used. The input
   data table included sedimentation patterns across the 22 Grad-seq
   fractions, expressed as row-Z-scores^[340]8. These data were provided
   for the indicated lncRNAs and diverse proteins classified as components
   of the specified protein machineries (Supplementary Data [341]12). Venn
   diagrams were generated using Venn Diagram Generator
   ([342]http://barc.wi.mit.edu/tools/venn/) or Venny 2.1
   ([343]https://bioinfogp.cnb.csic.es/tools/venny/). PCA analysis was
   performed in R Studio, based on RPKMs using the R-script prcomp (stats)
   and the rgl package. Network plots were generated using Cytoscape
   version 3.7.2. Violin plots were generated using BoxPlotR^[344]51.
   Other plots were generated in GraphPad PRISM or Excel.

Reporting summary

   Further information on research design is available in the [345]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [346]Supplementary Information^ (4.1MB, pdf)
   [347]41467_2025_60084_MOESM2_ESM.pdf^ (197.9KB, pdf)

   Description of Additional Supplementary Files
   [348]Supplementary Data 1^ (3.9MB, xlsx)
   [349]Supplementary Data 2^ (5.8MB, xlsx)
   [350]Supplementary Data 3^ (3.5MB, xlsx)
   [351]Supplementary Data 4^ (8.7MB, xlsx)
   [352]Supplementary Data 5^ (178.2KB, xlsx)
   [353]Supplementary Data 6^ (794.9KB, xlsx)
   [354]Supplementary Data 7^ (2.4MB, xlsx)
   [355]Supplementary Data 8^ (6.3MB, xlsx)
   [356]Supplementary Data 9^ (128.4KB, xlsx)
   [357]Supplementary Data 10^ (28.8KB, xlsx)
   [358]Supplementary Data 11^ (1.2MB, xlsx)
   [359]Supplementary Data 12^ (49.8KB, xlsx)
   [360]Reporting Summary^ (233KB, pdf)
   [361]Transparent Peer Review file^ (620.8KB, pdf)

Source data

   [362]Source Data^ (18.7MB, xlsx)

Acknowledgements