**1. Introduction**

Rice (*Oryza sativa* L.) is a staple crop feeding over half of the global population [1]. The Green Revolution accelerated the productivity of rice cultivation across Asia by focusing on irrigated, high input systems [2]. Intensification and expansion into new suboptimal cultivation areas, coupled with changing climatic conditions, however, necessitate a shift towards low input systems. Under resource- and water-limiting conditions, tolerance of both abiotic and biotic stress factors is crucial to ensure productivity.

Traditionally, rice was grown in areas naturally irrigated by seasonal floods [3]. Pre-domesticated rice is essentially a wetland species, making rice more sensitive to drought stress than most other staple crops [4]. Particularly, during the reproductive stage, drought typically causes yield reduction of 50% or more [5–10]. Consequently, water limitation is a major environmental constraint to rice production [11].

Successful strategies to identify factors contributing to drought tolerance involved mapping of quantitative trait loci (QTLs) for grain yield under drought conditions, so-called DTY (drought tolerant yield) QTLs [12]. By crossing the drought-tolerant donor N22 with Swarna, several major-e ffect DTY QTL, among them *qDTY1.1*, and *qDTY3.2* were identified as having consistent e ffects on grain yield under reproductive-stage drought stress (RDS) and no apparent yield or performance penalty under non-stress conditions [7]. These were subsequently introgressed into drought susceptible elite parents through backcrossing [13], resulting in the release of several drought-tolerant rice varieties. For example, "Bahuguni dhan-1", a sister line of DTY-IL used in this study, was recently released in Nepal [13]. In addition, several other large-e ffect DTY QTLs were identified from other populations and utilized for their potential to confer drought tolerance [14–17]. Gene discovery work in *qDTY12.1* resulted in the identification of a NAM transcription factor as an intra-QTL hub gene [18,19].

To unravel specific drought responses in rice, transcriptome studies across di fferent cultivars and drought stress conditions identified hundreds of di fferentially expressed genes (DEGs) in an organ- and time-specific fashion [1,10,20–27]. Collectively these studies pinpointed key transcription factors (TFs) involved in ABA-dependent and ABA-independent pathways to be upregulated during water deficit stress, e ffecting osmolyte production, reactive oxygen species (ROS) scavenging and ion transportation [10,22,28].

After an initial focus on studying drought during vegetative stages, the importance of RDS was soon realized [8,23,29–33]. From an applied perspective, grain yield under drought is the key trait, making yield-associated developmental processes and responses under RDS a focus of drought research in rice [8]. Knowledge of broad level biological mechanisms governing RDS responses, however, is still limited [34]. Convincing concepts on how drought-stress-related genes are regulated are still in their infancy. DEG analysis, based on single-genes, often failed to result in meaningful biological interpretations [35,36], prompting the development of network-based techniques that consider complex relationships among genes [37,38].

Gene co-expression networks (GCNs) are increasingly employed to explore system-level functionality of genes and have been found useful for describing the pairwise relationships among genes [39]. GCNs provide a structured pathway for extracting modular responses from large datasets that are often missed by DEG or ANOVA approaches [40]. It shifts the focus from single candidate genes to groups of related genes that likely operate together within a tissue or in response to a stimulus [41]. Genes clustering in a module, provide insight into potential regulatory functions [40,42,43]. An essential application of GCN analysis is to identify functional gene modules, which are a group of nodes that have high topological overlap [44]. Weighted gene correlation network analysis (WGCNA) can be used for co-expression network analysis of gene expression data to find modules of highly correlated genes [45]. In rice GCN analysis provided some insights into gene regulation under drought stress, including (i) consensus modules of downregulated and upregulated genes [46]; (ii) a module enriched for genes involved in water homeostasis and embryonic development, including a heat shock TF [47] and (iii) new candidates involved in drought response [48].

While a number of major QTL for DTY have been discovered, knowledge regarding the underlying physiological mechanisms is largely lacking. On the other hand, while a number of transcriptome studies provided some general insights into drought responses of rice, they did not take presence of specific DTY QTL into account. In the 2014 and 2015 drought field trials at the International Rice Research Institute (IRRI), a DTY introgression line (DTY-IL) performed well under drought without showing a penalty under irrigated conditions. We decided to investigate this line further in a comparative transcriptomic approach against its drought susceptible recurrent parent Swarna. Our rationale for this study was that combining a functional genomics with a classical genetics approach would improve resolution on drought tolerance mechanisms. By applying a gene co-expression network analysis and focusing on key source (flag-leaf) and sink (emerging panicle) tissues at reproductive stages, which had previously been demonstrated as critical for drought response [43], we speculated that adaptive mechanisms that drive yield under drought could be captured. By comparing the differential expression responses and overlaying genetic variation within the introgressed DTY QTL we further aimed to demonstrate that the differences at the genome-wide transcriptome level are modulated by the introgression segments. Our results support previous findings in respect to general mechanisms underlying drought response, and in addition sugges<sup>t</sup> specific mechanisms underpinning DTY QTL.
