diff --git a/cdot/__init__.py b/cdot/__init__.py
index d7fd326..f6184b5 100644
--- a/cdot/__init__.py
+++ b/cdot/__init__.py
@@ -1,5 +1,5 @@
 __version__ = "0.2.26"
-
+# Data version is kept in generate_transcript_version.json_schema_version
 
 def get_data_schema_int(version: str) -> int:
     """ Return an int which increments upon breaking changes - ie anything other than patch """
diff --git a/generate_transcript_data/ensembl_transcripts_chm13v2.sh b/generate_transcript_data/ensembl_transcripts_chm13v2.sh
index 3a833e0..7313308 100755
--- a/generate_transcript_data/ensembl_transcripts_chm13v2.sh
+++ b/generate_transcript_data/ensembl_transcripts_chm13v2.sh
@@ -13,7 +13,7 @@ fi
 
 merge_args=()
 for release in 2022_06 2022_07; do
-  filename=Homo_sapiens-GCA_009914755.4-${release}-genes.gff3.gz
+  filename=Homo_sapiens-GCA_009914755.4-${release}-genes.gtf.gz
   url=https://ftp.ensembl.org/pub/rapid-release/species/Homo_sapiens/GCA_009914755.4/ensembl/geneset/${release}/${filename}
   cdot_file=cdot-${CDOT_VERSION}.ensembl.$(basename $filename .gz).json.gz
 
@@ -21,7 +21,7 @@ for release in 2022_06 2022_07; do
     wget ${url}
   fi
   if [[ ! -e ${cdot_file} ]]; then
-    ${BASE_DIR}/cdot_json.py gff3_to_json "${filename}" --url "${url}" --genome-build=${GENOME_BUILD} --output "${cdot_file}" --gene-info-json="${GENE_INFO_JSON}"
+    ${BASE_DIR}/cdot_json.py gtf_to_json "${filename}" --url "${url}" --genome-build=${GENOME_BUILD} --output "${cdot_file}" --gene-info-json="${GENE_INFO_JSON}"
   fi
   merge_args+=(${cdot_file})
 done
diff --git a/generate_transcript_data/ensembl_transcripts_grch37.sh b/generate_transcript_data/ensembl_transcripts_grch37.sh
index 0d7db04..c42f29f 100755
--- a/generate_transcript_data/ensembl_transcripts_grch37.sh
+++ b/generate_transcript_data/ensembl_transcripts_grch37.sh
@@ -19,14 +19,14 @@ fi
 merge_args=()
 for release in 82 85 87; do
   # Switched to using GTFs as they contain protein version
-  filename=Homo_sapiens.GRCh37.${release}.gff3.gz
-  url=ftp://ftp.ensembl.org/pub/grch37/release-${release}/gff3/homo_sapiens/${filename}
+  filename=Homo_sapiens.GRCh37.${release}.gtf.gz
+  url=ftp://ftp.ensembl.org/pub/grch37/release-${release}/gtf/homo_sapiens/${filename}
   cdot_file=cdot-${CDOT_VERSION}.ensembl.$(basename $filename .gz).json.gz
   if [[ ! -e ${filename} ]]; then
     wget ${url}
   fi
   if [[ ! -e ${cdot_file} ]]; then
-    ${BASE_DIR}/cdot_json.py gff3_to_json "${filename}" --url "${url}" --genome-build=GRCh37 --output "${cdot_file}" --gene-info-json="${GENE_INFO_JSON}"
+    ${BASE_DIR}/cdot_json.py gtf_to_json "${filename}" --url "${url}" --genome-build=GRCh37 --output "${cdot_file}" --gene-info-json="${GENE_INFO_JSON}"
   fi
   merge_args+=(${cdot_file})
 done
diff --git a/generate_transcript_data/ensembl_transcripts_grch38.sh b/generate_transcript_data/ensembl_transcripts_grch38.sh
index 1cee0ba..9ded072 100755
--- a/generate_transcript_data/ensembl_transcripts_grch38.sh
+++ b/generate_transcript_data/ensembl_transcripts_grch38.sh
@@ -26,16 +26,16 @@ fi
 #81 is first GFF3 for GRCh38
 merge_args=()
 for release in 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112; do
-  # Switched to using GTFs as they contain protein version
-  filename=Homo_sapiens.GRCh38.${release}.gff3.gz
-  url=ftp://ftp.ensembl.org/pub/release-${release}/gff3/homo_sapiens/${filename}
+  # Switched to using GTFs as they contain protein version while Ensembl GFF3s do not (required for c_to_p)
+  filename=Homo_sapiens.GRCh38.${release}.gtf.gz
+  url=ftp://ftp.ensembl.org/pub/release-${release}/gtf/homo_sapiens/${filename}
   cdot_file=cdot-${CDOT_VERSION}.ensembl.$(basename $filename .gz).json.gz
 
   if [[ ! -e ${filename} ]]; then
     wget ${url}
   fi
   if [[ ! -e ${cdot_file} ]]; then
-    ${BASE_DIR}/cdot_json.py gff3_to_json "${filename}" --url "${url}" --genome-build=GRCh38 --output "${cdot_file}" --gene-info-json="${GENE_INFO_JSON}"
+    ${BASE_DIR}/cdot_json.py gtf_to_json "${filename}" --url "${url}" --genome-build=GRCh38 --output "${cdot_file}" --gene-info-json="${GENE_INFO_JSON}"
   fi
   merge_args+=(${cdot_file})
 done
diff --git a/generate_transcript_data/gff_parser.py b/generate_transcript_data/gff_parser.py
index 958e0fa..a6c4978 100644
--- a/generate_transcript_data/gff_parser.py
+++ b/generate_transcript_data/gff_parser.py
@@ -40,7 +40,7 @@ def __init__(self, filename, genome_build, url,
         self.transcript_proteins = {}
         # Store features in separate dict as we don't need to write all as JSON
         self.transcript_features_by_type = defaultdict(lambda: defaultdict(list))
-
+        self._warned_about_htseq_tag_attributes = False
 
         name_ac_map = {}
         if not no_contig_conversion:
@@ -345,6 +345,40 @@ def _get_gene_accession(feature) -> Optional[str]:
                 gene_accession = gene_id
         return gene_accession
 
+    def _handle_protein_version(self, transcript_accession, feature):
+        # RefSeq GFF3: CDS protein_id = NP_001659.1
+        # RefSeq GTF:  CDS protein_id = NP_001659.1
+        # Ensembl GTF: CDS protein_id = ENSP00000477624 protein_version = 1
+        # Ensembl GTF: CDS protein_id = ENSP00000477624  <----- can't use this one as no version
+        if feature.type == "CDS":
+            if protein_accession := feature.attr.get("protein_id"):
+                if protein_version := feature.attr.get("protein_version"):
+                    protein_accession = f"{protein_accession}.{protein_version}"
+                if not "." in protein_accession:
+                    raise ValueError(f"Protein '{protein_accession}' missing version")
+                self.transcript_proteins[transcript_accession] = protein_accession
+
+    def _add_tags_to_transcript_data(self, transcript_data, feature):
+        # Ideally we only want to get this once per transcript
+        # So we want to pick something like mRNA or transcript not CDS or exon
+        if feature.type in ('mRNA', 'transcript'):
+            attr_tuples = getattr(feature, "attr_tuples", None)
+            if attr_tuples is None:
+                attr_tuples = feature.attr.items()
+                if not self._warned_about_htseq_tag_attributes:
+                    self._warned_about_htseq_tag_attributes = True
+                    htseq_version = importlib.metadata.version('HTSeq')
+                    logging.warning("Your version of HTSeq (%s) can not handle duplicated tags. Some will be lost "
+                                    "See https://github.com/htseq/htseq/issues/83", htseq_version)
+
+            attr_list_vals = defaultdict(list)
+            for tag, value in attr_tuples:
+                attr_list_vals[tag].append(value)
+
+            if tag_list := attr_list_vals.get("tag"):
+                transcript_data["tag"] = ",".join(tag_list)
+
+
     def get_genes_and_transcripts(self):
         self._parse()
         self._finish()
@@ -356,6 +390,9 @@ class GTFParser(GFFParser):
     """ GTF (GFF2) - used by Ensembl, @see http://gmod.org/wiki/GFF2
 
         GFF2 only has 2 levels of feature hierarchy, so we have to build or 3 levels of gene/transcript/exons ourselves
+
+        We *have* to use GTF as Ensembl GFF3s don't include the protein version (just the ID)
+
     """
     GTF_TRANSCRIPTS_DATA = GFFParser.CODING_FEATURES | {"exon"}
     FEATURE_ALLOW_LIST = GTF_TRANSCRIPTS_DATA | {"gene", "transcript"}
@@ -392,14 +429,8 @@ def handle_feature(self, feature):
                 gene_data["biotype"].add(biotype)
                 transcript["biotype"].add(biotype)
 
-            if feature.type == "CDS":
-                if protein := feature.attr.get("protein_id"):
-                    if protein_version := feature.attr.get("protein_version"):
-                        protein = f"{protein}.{protein_version}"
-                    self.transcript_proteins[transcript_accession] = protein
-            elif feature.type == "transcript":
-                if tag := feature.attr.get("tag"):
-                    transcript["tag"] = tag
+            self._handle_protein_version(transcript_accession, feature)
+            self._add_tags_to_transcript_data(transcript, feature)
 
 
 class GFF3Parser(GFFParser):
@@ -461,15 +492,11 @@ def handle_feature(self, feature):
                     # Some exons etc may be for miRNAs that have no transcript ID, so skip those (won't have parent)
                     if parent_id:
                         transcript_accession = self.transcript_accession_by_feature_id.get(parent_id)
+                        self._handle_protein_version(transcript_accession, feature)
                     else:
                         logging.warning("Transcript data has no parent: %s" % feature.get_gff_line())
                         transcript_accession = None
 
-                    if feature.type == "CDS":
-                        dbxref = self._get_dbxref(feature)
-                        if genbank := (dbxref.get("Genbank") or dbxref.get("GenBank")):
-                            self.transcript_proteins[transcript_accession] = genbank
-
                 if transcript_accession:
                     transcript = self.transcript_data_by_accession.get(transcript_accession)
                     if not transcript:
@@ -516,8 +543,7 @@ def _handle_transcript(self, gene_data, transcript_accession, feature):
             if feature.attr.get("partial"):
                 transcript_data["partial"] = 1
 
-            if tag := feature.attr.get("tag"):
-                transcript_data["tag"] = tag
+            self._add_tags_to_transcript_data(transcript_data, feature)
 
             self.transcript_data_by_accession[transcript_accession] = transcript_data
         self.transcript_accession_by_feature_id[feature.attr["ID"]] = transcript_accession
diff --git a/generate_transcript_data/json_schema_version.py b/generate_transcript_data/json_schema_version.py
index 267d03b..ff84f8b 100644
--- a/generate_transcript_data/json_schema_version.py
+++ b/generate_transcript_data/json_schema_version.py
@@ -1,3 +1,3 @@
 # After 0.2.22 we split version into separate code (pip) and data schema versions
 # The cdot client will use its own major/minor to determine whether it can read these data files
-JSON_SCHEMA_VERSION = "0.2.26"
+JSON_SCHEMA_VERSION = "0.2.27"