How to optimize inefficient query in PostGIS

I'm doing some queries on a PostGIS, and some of them take a really long time ( > 60 secs). My queries are like this, except querying different tables (like osm_placesbelow).

SELECT row_to_json(fc) FROM ( 
  SELECT 
   ST_AsGeoJSON(ST_Transform(ST_Intersection(lg.geometry, bbox.geom), 4326))::json 
   As geometry, ( 
   SELECT row_to_json(t) FROM ( SELECT "name","type","population","z_order") t )
   As properties FROM osm_places As lg, (
   SELECT ST_Transform(ST_MakeEnvelope(-2.8125,54.162433968067795,25.312499999999986,
    67.60922060496382, 4326),3857) as geom) 
    as bbox WHERE ST_Intersects(lg.geometry, bbox.geom)  
    AND (type = 'country' OR type = 'city') ) fc;

This is the EXPLAIN of the above query:

 Nested Loop  (cost=18.51..1159.41 rows=2 width=107)
       Output: row_to_json(ROW((_st_asgeojson(1, st_transform(st_intersection(lg.geometry, ('0103000020110F0000010000000500000080107C45F81B13C1CF57E2E334785B4180107
       ->  Result  (cost=0.00..0.01 rows=1 width=0)
             Output: '0103000020110F0000010000000500000080107C45F81B13C1CF57E2E334785B4180107C45F81B13C117F1A707D8B46341A3922B4E777F454117F1A707D8B46341A3922B4E777F
       ->  Bitmap Heap Scan on public.osm_places lg  (cost=18.51..1158.81 rows=1 width=75)
             Output: lg.id, lg.osm_id, lg.name, lg.type, lg.z_order, lg.population, lg.geometry
             Recheck Cond: (lg.geometry && ('0103000020110F0000010000000500000080107C45F81B13C1CF57E2E334785B4180107C45F81B13C117F1A707D8B46341A3922B4E777F454117F1A
             Filter: ((((lg.type)::text = 'country'::text) OR ((lg.type)::text = 'city'::text)) AND _st_intersects(lg.geometry, ('0103000020110F00000100000005000000
             ->  Bitmap Index Scan on osm_places_geom  (cost=0.00..18.51 rows=279 width=0)
                   Index Cond: (lg.geometry && ('0103000020110F0000010000000500000080107C45F81B13C1CF57E2E334785B4180107C45F81B13C117F1A707D8B46341A3922B4E777F45411
       SubPlan 1
         ->  Subquery Scan on t  (cost=0.00..0.02 rows=1 width=24)
               Output: row_to_json(t.*)
               ->  Result  (cost=0.00..0.01 rows=1 width=0)
                     Output: lg.name, lg.type, lg.population, lg.z_order
    (15 rows)

How can I optimize this query? Are there ways I can query more efficiently? Should I create more indexes? The data is the OSM PostGIS database.