A new species Allium sulaimanicum is described from northern Balochistan and southern Khyper Pakhtunkhwa in Pakistan based on morphological, molecular and cytological studies. The new species is characterised by long runner-like cylindrical rhizomes of adult plants, cylindrical bulbs, linear leaves with minute soft hairs along veins, campanulate perigonium, white to creamy white, ovate to elliptical, 4.5–5 mm long acute tepals, with brownish to purplish nerves, stamens as long as to slightly longer than tepals, yellow to brick red anthers, hexagonal ovary, white and papillate/warty along angles. The presence of long herbaceous rhizome indicated serious isolation of the new species hence, a new section Sulaimanicum is proposed to accommodate the new species. The new species is diploid with a chromosome number of 2n =16. Detailed morphological description, illustrations, phylogenetic analyses based on sequences of plastid spacers (rpl32-trnL (UAG) and trnQ-rps16) and nuclear ITS, karyotype features and a distribution map of the new species are provided.

Phylogenetic analyses—Both data sets (nrITS and the cpDNA rpl32-trnL markers) were analysed separately for position identification in the third evolutionary lineage and to find the closest relatives of A. sulaimanicum through Fitch parsimony with the heuristic search option in PAUP* version 4.0b10 (Swofford, 2002) with MULTREES, TBR branch swapping and 100 replicates of random addition sequence. Gaps were treated as missing data. The consistency index (CI) (Kluge & Farris, 1969) was calculated to estimate the amount of homoplasy in the character set. The most parsimonious trees returned by the analysis were summarized in one consensus tree using the strict consensus method. Bootstrap analyses (BS) using 1,000 pseudoreplicates were performed to assess the support of the clades (Felsenstein, 1985). Bayesian phylogenetic analyses were also performed using MrBayes 3.1.23 (Ronquist & Huelsenbeck, 2003). The sequence evolution model was chosen following the Akaike Information Criterion (AIC) obtained from jModelTest2 (Dariba et al. 2012). Two independent analyses with four Markov chains were run for 10 million generations, sampling trees every 100 generations. The first 25% of trees were discarded as burn-in. The remaining 150,000 trees were combined into a single data set, and a majority-rule consensus tree was obtained along with posterior probabilities (PP).